{"agents":{"a01":{"id":"a01","alias":"Kestrel","theme":"Time-series momentum and trend following: which horizons, which filters, how to size","started":"2026-10-11T00:00:07+00:00","rounds":[{"k":1,"minutes":78.0},{"k":2,"minutes":44.9},{"k":3,"minutes":22.4}],"messages":[{"t":"2026-10-11T00:00:16.339Z","text":"I’ll inspect the simulator and data, submit an initial trend strategy, then test broader momentum rules for results that hold up across the visible period. I’ll track each variant and its assumptions in the writeup."},{"t":"2026-10-11T00:01:25.390Z","text":"The first momentum tests lost heavily, even with a Bitcoin trend filter. I’m submitting a low-turnover Bitcoin trend baseline now, then checking whether those losses come from reversals, execution timing, or selection of unstable coins."},{"t":"2026-10-11T00:03:02.676Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% drawdown. Across the broader tests, chasing recent altcoin leaders is still losing; next I’m testing steadier trends and buying pullbacks within them, alongside rotation among BTC, ETH and SOL."},{"t":"2026-10-11T00:05:34.813Z","text":"The channel-breakout tests are stronger than the moving-average tests so far. I also found a start-up issue with flip-only output: a fresh account can miss an existing trend. Candidates will refresh targets periodically so they can begin trading when started mid-trend."},{"t":"2026-10-11T00:08:00.244Z","text":"The new entry was accepted: $100 became $847.49, with a 28.7% drawdown. It buys channel breakouts and holds the strongest risk-adjusted trend among BTC, ETH and SOL. I’m testing broader universes and four-hour execution next, then checking nearby parameters and the trades driving the gains."},{"t":"2026-10-11T00:11:19.915Z","text":"The broader-universe and four-hour breakout tests underperformed the daily entry. I’m now testing whether a small walk-forward model can distinguish persistent demand from isolated price and volume spikes. Each training set will contain only outcomes already known at that retraining point."},{"t":"2026-10-11T00:13:52.711Z","text":"The pooled model did not improve net returns. I’m keeping it out of the entry and separating further return searches from robustness checks on the channel strategy, so a large parameter search doesn’t become our only evidence."},{"t":"2026-10-11T00:15:56.373Z","text":"The current entry still made money with doubled costs and with its ten best days removed. Nearby parameters also remained profitable, though returns varied widely. Most of its gain came from SOL, so I’m testing whether a very small universe selected by persistent trading volume can find other leaders."},{"t":"2026-10-11T00:20:56.683Z","text":"The latest accepted entry turns $100 into $1,752.94, with a 33.8% drawdown. It adds coins only when their median trading volume places them among the top five, allowing new leaders to enter without naming them. SOL still supplied most of the gain, though the additional coins contributed; I’m also testing whether past channel-trading profits make a better leader score than raw price momentum."},{"t":"2026-10-11T00:23:24.804Z","text":"A complementary rule roughly doubled the visible result: while the trend strategy is in cash, it buys very oversold conditions in a major coin and exits the rebound quickly. I’m checking whether that gain survives nearby thresholds and a sharp-loss guard before submitting it."},{"t":"2026-10-11T00:26:49.287Z","text":"The fourth submission is in for review at $3,354.40, with a 34.6% drawdown. Nearby rebound settings generally improved on the trend-only strategy. The rebound coin is selected by volatility among BTC, ETH and SOL, and the rule rejects or exits a daily loss of 15% or more."},{"t":"2026-10-11T00:29:07.784Z","text":"The hybrid entry was accepted and is currently second on the board. The short-horizon learning model also lost money after costs. I’m testing one further approach: choosing momentum horizons from their past net trading results, so the horizon can adapt causally instead of staying fixed."},{"t":"2026-10-11T00:32:31.980Z","text":"The adaptive horizon ensemble underperformed the fixed channel rule. I’m keeping the current entry and testing a smaller extension: a cautious momentum estimate for newly listed liquidity leaders, which currently need six months of price history before they can be selected."},{"t":"2026-10-11T00:35:49.835Z","text":"Early eligibility for young listings weakened results, so I’m retaining the six-month history requirement. A more promising cash rule buys a liquid coin after a sharp intraday dip that largely recovered by the close. I’m checking its parameter neighborhood and earlier-period results before deciding whether to add it."},{"t":"2026-10-11T00:39:13.923Z","text":"The fifth entry was accepted: $100 became $4,254.43, with a 35.4% drawdown. It adds the price-recovery trade only when both earlier rules would hold cash. I’m checking its costs and return attribution now, including how much depends on a few exceptional days."},{"t":"2026-10-11T00:42:13.512Z","text":"The full strategy remained profitable at three times the stated costs. Removing its twenty best days left $237.49 from $100, showing how strongly the result depends on large winners. SOL and PEPE supplied most of the gain. I’m documenting those limits and the full search count in the writeup."},{"t":"2026-10-11T00:45:40.315Z","text":"The writeup now covers the accepted code, equations, parameters, failed approaches, and 3,058 tested configurations. I’m trying a small change to leader selection: combining price momentum with where the coin closes in its daily range, weighted by relative volume. The current entry will stay unless the evidence improves."},{"t":"2026-10-11T00:47:28.394Z","text":"The range-volume proxy reduced visible returns, so I’m leaving it out. I’m testing a causal local-trend filter next: it estimates the current slope and its uncertainty, which may offer a way to identify a new leader sooner than the six-month return score."},{"t":"2026-10-11T00:54:20.550Z","text":"The local-trend version was accepted: $100 became $4,809.00, with a 31.8% drawdown. It uses the same entries and cash rules, but selects the active leader by estimated log-price slope. Nearby filter settings were generally profitable. I’m updating the equations and robustness report to match this entry."},{"t":"2026-10-11T00:59:54.443Z","text":"The new entry stayed profitable at three times costs and still returned $285.75 after removing its twenty best days. I’m testing whether a constrained growth allocation across active trends improves on putting everything in one leader; costs and covariance-estimation noise may erase the diversification gain."},{"t":"2026-10-11T01:02:30.299Z","text":"The growth-allocation tests underperformed, so I’m retaining the single-leader entry. I’m finishing the paper figure and checking that the submitted source, local copy, and writeup all match."},{"t":"2026-10-11T01:07:46.902Z","text":"The figure and writeup are in place. I’m testing one last return idea for this round: a small trend that is strengthening gradually on rising volume, with a quick exit after a large jump. This addresses an earlier failure: buying the apparent leader only after its biggest day."},{"t":"2026-10-11T01:13:15.330Z","text":"The current entry returns 48.09× with a 31.8% drawdown. A final test of short continuation trades during broad uptrends improved both the visible result and the earlier-data result. I’m checking nearby parameters before deciding whether to submit it."},{"t":"2026-10-11T01:15:06.329Z","text":"The continuation idea held up across nearby settings: 76% improved on the current entry, and their median result was $5,695. The best region favors short holds and fewer continuation candidates. I’m narrowing that check and measuring the added cost sensitivity before the final submission."},{"t":"2026-10-11T01:17:09.436Z","text":"The final candidate returned $8,891.70 from $100 with a 30.0% maximum drawdown and passed the causal checks. Submission #007 is now with the referee. The writeup includes every signal equation, all 3,354 tested configurations, cost sensitivity, and the concentration risks in the result."},{"t":"2026-10-11T01:17:39.489Z","text":"Submission #007 was accepted and is now the entry. It ranks second among the 25 agents currently entered. The writeup and figures are complete; the main remaining concern is that most profits came from a few strong moves despite the favorable parameter and cost checks."},{"t":"2026-10-11T01:18:09.246Z","text":"Entry: #007 accepted — strategy.py and writeup.  \nVisible: $100 → **$8,891.70** (88.92×); drawdown **30.0%**; currently **2nd of 25**.  \nNext idea: Volatility-adaptive exits, checked across rolling fresh-account starts."},{"t":"2026-10-11T01:18:16.900Z","text":"I’ll look for a different source of edge while keeping #007 as the entry. I’ll start with the pre-cutoff return data and compare new ideas across several fresh-account windows."},{"t":"2026-10-11T01:20:05.531Z","text":"The event study gives daily lower-wick recoveries a better signal than hourly or 15-minute reversals. The short-bar effects are generally too small to cover costs. I’m testing whether the daily recovery edge survives broader coin selection and a longer hold, while also checking momentum adjusted for BTC exposure."},{"t":"2026-10-11T01:26:17.876Z","text":"The new recovery signal helps as a last-priority trade, but replacing the existing recovery rule made results worse. Its central setting returned $10,858 with a 28.1% drawdown. The BTC-adjusted momentum variants underperformed. I’m also testing sequential trend evidence and a walk-forward classifier for large upside moves."},{"t":"2026-10-11T01:28:51.963Z","text":"#008 was accepted and is now the entry: $10,858.15 from $100, with a 28.1% drawdown. The large-move classifier added no useful edge after costs. The sequential-evidence test has a promising candidate; I’m checking its parameter neighborhood before considering it."},{"t":"2026-10-11T01:56:39.250Z","text":"The sequential-evidence strategy reached $13,679 in its best test, but the improvement was sensitive to its thresholds and came with higher drawdown. I’m keeping the accepted $10,858 entry and updating the writeup with the new signal, all research variants, and the robustness results."},{"t":"2026-10-11T01:57:58.715Z","text":"The writeup now covers the accepted five-rule strategy and all 4,552 tested configurations. It passed the causal checks; doubling trading costs reduced visible wealth to $5,965. I’m using the remaining round time to test breakouts after quiet trading periods, which may offer a different entry signal."},{"t":"2026-10-11T01:59:51.773Z","text":"The compression-breakout test also fell short: its best overlay reached $11,353, but the median fell to $7,167. I’ve retained submission 008 and completed the writeup. The leaderboard now places it eighth of 40, so the next round needs a larger change in approach."},{"t":"2026-10-11T02:00:59.827Z","text":"The accumulation test found a more promising candidate: $16,764 with 29.0% drawdown, and a stronger result in the earlier year. It measures where closes fall within the daily range, weighted by volume, before a modest breakout. I’m checking nearby settings before deciding whether to submit it."},{"t":"2026-10-11T02:02:28.243Z","text":"The accumulation draft passed the referee’s three causal prefix checks and reproduced $16,764. Nearby settings were mixed: five of 27 beat the current entry. I’m leaving submission 008 in place for this round; next I’ll check the draft’s costs and annual concentration before deciding whether to adopt it."},{"t":"2026-10-11T02:03:04.166Z","text":"Entry: submission 008, strategy.py; writeup updated.\nVisible: **$10,858.15 (×108.58)**; maximum drawdown **28.1%**.\nNext idea: validate the accumulation overlay—$16,764 in research, with causal checks passed."},{"t":"2026-10-11T02:03:10.488Z","text":"I’ll finish validating the accumulation overlay, choose the final entry, and leave time for scoring before the workspace seals. The writeup will document the accepted code, results, and limitations."},{"t":"2026-10-11T02:04:39.710Z","text":"I’m submitting the accumulation version. It improved all three fresh annual tests and the consulted earlier year, and retained an advantage with doubled costs: $7,413 versus $5,965. Its neighborhood is uneven, so I’ll document that sensitivity alongside the gains."},{"t":"2026-10-11T02:07:03.417Z","text":"Submission 009 was accepted and is now the entry: $16,764.38 from $100, with 29.0% drawdown. The writeup covers its six rules and results. I’m testing one further hypothesis: whether sustained growth in trading volume can identify emerging trends before they become the largest-volume coins."},{"t":"2026-10-11T02:08:52.304Z","text":"The participation-growth overlay reached $22,513 at its best, but raised drawdown to 40.4% and was unstable across variants. I’ve rejected it. I’m now checking a different opportunity: price discovery soon after a coin first appears in the data, with selection based entirely on age, volume, and price action."},{"t":"2026-10-11T02:11:53.831Z","text":"The early-listing strategy did not improve the entry and suffered larger drawdowns. I also tested causal estimates of each signal’s advantage over the current target; conservative settings made no changes, while more permissive settings added losing trades. That supports keeping the fixed priority order."},{"t":"2026-10-11T02:15:07.900Z","text":"A narrower liquidity neighborhood improved the evidence for the participation rule: 18 of 27 variants beat submission 009. The gain persists across several expansion thresholds and maturity settings, but depends on keeping the universe small. I’m checking costs and staggered two-year starts to weigh the higher drawdown against that gain."},{"t":"2026-10-11T02:17:07.919Z","text":"The participation rule still leads after doubled and tripled costs: $9,469 and $3,983. Its gain comes from 31 additional invested days, while drawdown rises to 40.4%. I’m finishing the two-year comparisons and a tighter-exit test before choosing between it and submission 009."},{"t":"2026-10-11T02:19:27.018Z","text":"I’m keeping submission 009 as the final entry. The higher-dollar participation version lost to it on three of five staggered two-year tests and raised drawdown to 40.4%. Tighter exits and an extra price filter did not resolve that tradeoff. I’ll finish by auditing the accepted file and documenting these rejected alternatives."},{"t":"2026-10-11T02:22:58.027Z","text":"The final writeup now records 4,910 configuration evaluations and equations for all six signals, their filters, target sizing, and cost accounting. Submission 009 remains unchanged and matches the referee’s accepted copy. I’m checking the result tables and saved artifacts for consistency."},{"t":"2026-10-11T02:25:26.060Z","text":"Final entry: **submission 009**, strategy.py, uses daily liquidity-selected trend leaders ranked by a robust slope estimate, with five prioritized recovery and continuation rules when the trend block selects cash. It turned $100 into **$16,764.38 (×167.64)** on the visible years after costs, with **29.0% maximum drawdown**, and passed causal checks. WRITEUP.md is complete with equations, 4,910 configuration evaluations, robustness results, rejected ideas, and overfitting limitations."}],"submissions":[{"n":"001","t":"2026-10-11T00:01:29.873081+00:00","note":"Initial entry: Bitcoin 200-day trend filter; momentum research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:06:42.026295+00:00","note":"Channel-entry trend rotation across BTC, ETH and SOL; causal, fully invested in at most one major.","status":"ok","dev":847.4906239849815,"dd":-0.2868130455694107,"reason":""},{"n":"003","t":"2026-10-11T00:18:24.588051+00:00","note":"Persistent liquidity top five plus BTC/ETH/SOL; 30/7-day channels and six-month relative strength.","status":"ok","dev":1752.9433423982277,"dd":-0.3375052408983199,"reason":""},{"n":"004","t":"2026-10-11T00:25:59.644317+00:00","note":"Trend leadership with guarded RSI rebounds in the most volatile major, only while trend exposure is zero.","status":"ok","dev":3354.4014532938413,"dd":-0.34630436390042385,"reason":""},{"n":"005","t":"2026-10-11T00:37:38.110932+00:00","note":"Three priorities: persistent liquid trends, guarded major rebounds, then one-bar price-recovery candles.","status":"ok","dev":4254.430828567441,"dd":-0.35378684568132557,"reason":""},{"n":"006","t":"2026-10-11T00:52:55.536707+00:00","note":"Robust local-log-price slope replaces fixed momentum rank; channels, liquidity and rebound hierarchy retained.","status":"ok","dev":4809.002824505,"dd":-0.31832197522120287,"reason":""},{"n":"007","t":"2026-10-11T01:16:54.688865+00:00","note":"Robust local trends plus last-priority, market-gated short continuation; neighbors, cost sensitivity and causal checks documented.","status":"ok","dev":8891.69684150335,"dd":-0.3000780437238998,"reason":""},{"n":"008","t":"2026-10-11T01:26:44.743727+00:00","note":"Daily rejected-range recovery pulses as a protected cash fallback; central parameters improve wealth and drawdown, with causal and cost checks.","status":"ok","dev":10858.154793970483,"dd":-0.28053983929774484,"reason":""},{"n":"009","t":"2026-10-11T02:04:43.032710+00:00","note":"Robust trend leaders with causal volume-weighted accumulation as last cash fallback; annual and doubled-cost checks improved; no future data or coin lists.","status":"ok","dev":16764.378497076203,"dd":-0.28992738425064635,"reason":""}],"entry":"009","round_notes":[{"k":1,"text":"Entry: #007 accepted — strategy.py and writeup.  \nVisible: $100 → **$8,891.70** (88.92×); drawdown **30.0%**; currently **2nd of 25**.  \nNext idea: Volatility-adaptive exits, checked across rolling fresh-account starts."},{"k":2,"text":"Entry: submission 008, strategy.py; writeup updated.\nVisible: **$10,858.15 (×108.58)**; maximum drawdown **28.1%**.\nNext idea: validate the accumulation overlay—$16,764 in research, with causal checks passed."},{"k":3,"text":"Final entry: **submission 009**, strategy.py, uses daily liquidity-selected trend leaders ranked by a robust slope estimate, with five prioritized recovery and continuation rules when the trend block selects cash. It turned $100 into **$16,764.38 (×167.64)** on the visible years after costs, with **29.0% maximum drawdown**, and passed causal checks. WRITEUP.md is complete with equations, 4,910 configuration evaluations, robustness results, rejected ideas, and overfitting limitations."}],"title":"Kestrel (a01): writeup","idea":"The accepted entry is submission 009, kestrel_accumulation.py, copied unchanged to strategy.py. It allocates capital to six opportunities in priority order: a persistent trend among a few liquidity leaders; a brief rebound in an extremely oversold major; a one-bar recovery after a large lower wick; and a short, volume-confirmed continuation while BTC is above its long moving average; and a one-bar rejected-range pulse after a volatility-adjusted decline; and an accumulation breakout confirmed by volume-weighted close location. A robust local-log-price filter ranks the core trends by their estimated persistent slope. Each decision targets one coin at 100%, or cash. Trailing median quote volume admits emerging leaders while resisting a single day's spike. The strategy uses daily bars, causal state updates, and no external data, calendar rules, or fixed coin list beyond the permitted BTC, ETH and SOL.","failed":"- Raw broad-universe momentum, even with BTC trend gates, selected many temporary pump leaders just as they reversed. All initial variants were much weaker than the BTC baseline. - Volatility scaling and price-path efficiency did not generally repair broad selection. Lagging the newest returns and adding a breadth gate helped some longer horizons but remained weaker than the small-universe channel rule. - Always-invested cross-sectional reversal paid too much turnover and suffered adverse tails. Absolute extreme-decline statistics were also unstable between the earlier and visible periods; buying the largest declines indiscriminately was particularly poor. - Moving-average and four-hour channel variants often whipsawed or surrendered too much of a move before exit. Four-hour and hourly RSI rebounds did not justify extra activity. - Winsorized momentum, nominal-token-price screens, and broad channel portfolios were inferior. A small nominal unit price was not a sufficient reason to buy a token. - Daily pooled gradient boosting, trained strictly walk-forward, failed after costs. A separate hourly walk-forward model also failed; causal learning alone did not imply an edge. LightGBM could not load because the local OpenMP library was missing, so these experiments used scikit-learn histogram gradient boosting without downloading anything. - Volatility-scaled directional-change entries did not beat channel entries. Ranking coins by the realized profits of their own past channel strategy was promising on majors but less reliable when extended to more coins. - An online, net-cost expert ensemble of channel/momentum horizons underperformed the fixed channel entry. Normalizing away its cash allocation was especially damaging. - Age-aware, shrinkage-adjusted momentum allowed young","holdup":"The rule separates persistent trends, countertrend recovery, and emerging continuation. Established trends can continue when demand persists; channel exits stop waiting for a distant moving average after a breakdown. The local slope estimates the persistent growth component instead of ranking solely on a distant return; its robust innovation cap limits the influence of a single surprising bar. This spends a limited spot budget on a leader. Median trading volume measures sustained participation, so a temporary pump cannot qualify an illiquid name merely through one large volume bar. The tiny core universe was materially more reliable than broad momentum universes in this data. The cash rules address a different situation: a very lopsided sequence of declines, or an intraday rejection of low prices. The first uses only a liquid major, rejects large daily losses, and has a finite signal age. The second demands evidence that buyers actually appeared and is reconsidered after one bar. All signals are scale-free except the explicit USDT liquidity floor, which measures tradability, and every universe is selected causally. The fifth block ranks the size of a rejected excursion rather than ","result":{"dev":{"final":16764.378497076203,"max_drawdown":-0.28992738425064635,"trades":396,"exposure":0.6377737226277372,"assets_traded":45,"fees_paid":2006.2714044787456},"hidden":{"final":113.50838884922607,"max_drawdown":-0.7396222605887279,"trades":275,"exposure":0.6383561643835617,"assets_traded":38,"fees_paid":64.29394984623681},"full":{"final":19028.97593261726,"max_drawdown":-0.7396222605887282,"trades":671,"exposure":0.6380065717415115,"assets_traded":67,"fees_paid":12784.75250742222}},"freq":"1d","equity":[["2021-10-17",108.7152],["2021-10-24",131.9151],["2021-10-31",130.5476],["2021-11-07",172.4002],["2021-11-14",163.8311],["2021-11-21",150.8945],["2021-11-28",150.8945],["2021-12-05",204.1646],["2021-12-12",176.8174],["2021-12-19",183.3565],["2021-12-26",174.3318],["2022-01-02",174.3318],["2022-01-09",155.4222],["2022-01-16",182.825],["2022-01-23",170.0717],["2022-01-30",204.8134],["2022-02-06",218.5286],["2022-02-13",214.3289],["2022-02-20",189.1204],["2022-02-27",205.5382],["2022-03-06",230.3459],["2022-03-13",176.5724],["2022-03-20",189.0074],["2022-03-27",187.4889],["2022-04-03",236.3465],["2022-04-10",193.4882],["2022-04-17",193.4882],["2022-04-24",195.8142],["2022-05-01",181.0901],["2022-05-08",174.135],["2022-05-15",184.8036],["2022-05-22",186.9723],["2022-05-29",186.9723],["2022-06-05",186.9723],["2022-06-12",184.8863],["2022-06-19",172.398],["2022-06-26",183.771],["2022-07-03",175.7426],["2022-07-10",175.7426],["2022-07-17",183.9095],["2022-07-24",189.8241],["2022-07-31",212.1403],["2022-08-07",232.3371],["2022-08-14",239.0448],["2022-08-21",220.8821],["2022-08-28",220.8821],["2022-09-04",218.4107],["2022-09-11",235.6188],["2022-09-18",230.1514],["2022-09-25",224.9903],["2022-10-02",234.8442],["2022-10-09",221.3575],["2022-10-16",221.5746],["2022-10-23",221.815],["2022-10-30",256.8409],["2022-11-06",294.6414],["2022-11-13",224.8052],["2022-11-20",230.1528],["2022-11-27",251.7225],["2022-12-04",252.4567],["2022-12-11",251.7268],["2022-12-18",222.7064],["2022-12-25",225.5843],["2023-01-01",230.2078],["2023-01-08",256.5553],["2023-01-15",334.6398],["2023-01-22",351.2131],["2023-01-29",339.433],["2023-02-05",359.7891],["2023-02-12",333.3238],["2023-02-19",323.4997],["2023-02-26",307.1628],["2023-03-05",305.2231],["2023-03-12",292.2413],["2023-03-19",394.1179],["2023-03-26",390.9533],["2023-04-02",435.9664],["2023-04-09",425.7378],["2023-04-16",452.9702],["2023-04-23",400.1723],["2023-04-30",400.1723],["2023-05-07",391.3061],["2023-05-14",383.8682],["2023-05-21",392.8596],["2023-05-28",598.4231],["2023-06-04",655.106],["2023-06-11",639.8269],["2023-06-18",688.3256],["2023-06-25",762.5977],["2023-07-02",764.0583],["2023-07-09",828.9329],["2023-07-16",976.9671],["2023-07-23",889.2177],["2023-07-30",829.0753],["2023-08-06",900.5163],["2023-08-13",842.8691],["2023-08-20",803.0654],["2023-08-27",829.3752],["2023-09-03",739.737],["2023-09-10",739.737],["2023-09-17",793.151],["2023-09-24",785.2802],["2023-10-01",814.9567],["2023-10-08",885.4515],["2023-10-15",846.7394],["2023-10-22",1035.5574],["2023-10-29",1117.1446],["2023-11-05",1497.885],["2023-11-12",1989.8807],["2023-11-19",2068.2892],["2023-11-26",1924.4516],["2023-12-03",2009.5008],["2023-12-10",2347.9041],["2023-12-17",2394.0179],["2023-12-24",3503.3457],["2023-12-31",3309.1484],["2024-01-07",3375.0908],["2024-01-14",3349.8143],["2024-01-21",3131.0193],["2024-01-28",3194.0088],["2024-02-04",3130.3399],["2024-02-11",3106.709],["2024-02-18",3104.1441],["2024-02-25",3058.2003],["2024-03-03",3768.9086],["2024-03-10",4182.5552],["2024-03-17",5277.1098],["2024-03-24",5004.6306],["2024-03-31",5629.3036],["2024-04-07",5256.7745],["2024-04-14",5218.1177],["2024-04-21",5692.8972],["2024-04-28",5514.4821],["2024-05-05",6112.2356],["2024-05-12",5655.2991],["2024-05-19",6332.3419],["2024-05-26",9936.5995],["2024-06-02",9863.585],["2024-06-09",9184.4423],["2024-06-16",9379.2671],["2024-06-23",8421.3805],["2024-06-30",9528.7624],["2024-07-07",10392.4184],["2024-07-14",11468.153],["2024-07-21",12720.7724],["2024-07-28",13476.9842],["2024-08-04",12304.3309],["2024-08-11",14496.9993],["2024-08-18",14496.9993],["2024-08-25",14722.0655],["2024-09-01",14389.9147],["2024-09-08",14375.5248],["2024-09-15",14131.8542],["2024-09-22",19435.9144],["2024-09-29",22258.3446],["2024-10-06",16719.2583],["2024-10-13",18336.4548],["2024-10-20",18151.03],["2024-10-27",17594.9363],["2024-11-03",17109.9635],["2024-11-10",18007.7195],["2024-11-17",28659.6508],["2024-11-24",27241.6344],["2024-12-01",27456.9955],["2024-12-08",30774.6896],["2024-12-15",24649.4294],["2024-12-22",23105.8845],["2024-12-29",23047.0899],["2025-01-05",24804.8568],["2025-01-12",23948.1684],["2025-01-19",30316.7428],["2025-01-26",28817.4562],["2025-02-02",26683.0359],["2025-02-09",24792.4336],["2025-02-16",26261.2024],["2025-02-23",26475.179],["2025-03-02",27942.5695],["2025-03-09",25933.2774],["2025-03-16",25933.2774],["2025-03-23",25536.6751],["2025-03-30",24634.1003],["2025-04-06",24122.0736],["2025-04-13",23904.849],["2025-04-20",24220.7257],["2025-04-27",26162.7977],["2025-05-04",24423.4868],["2025-05-11",26791.7738],["2025-05-18",23366.4881],["2025-05-25",24331.3744],["2025-06-01",22858.9718],["2025-06-08",23188.7654],["2025-06-15",20816.0649],["2025-06-22",18888.7629],["2025-06-29",21641.7569],["2025-07-06",21641.7569],["2025-07-13",24167.0011],["2025-07-20",30208.9719],["2025-07-27",27905.4323],["2025-08-03",27286.8127],["2025-08-10",29701.7102],["2025-08-17",30826.6556],["2025-08-24",28900.2461],["2025-08-31",26011.4242],["2025-09-07",25685.4794],["2025-09-14",32529.4465],["2025-09-21",29832.1803],["2025-09-28",27019.8238],["2025-10-05",27796.8488],["2025-10-12",24896.359],["2025-10-19",25207.5382],["2025-10-26",25144.5193],["2025-11-02",24121.7427],["2025-11-09",23941.3083],["2025-11-16",23125.9434],["2025-11-23",20162.3484],["2025-11-30",21559.9683],["2025-12-07",20448.5207],["2025-12-14",20382.746],["2025-12-21",19083.7693],["2025-12-28",19083.7693],["2026-01-04",19188.5967],["2026-01-11",17811.5959],["2026-01-18",18226.0927],["2026-01-25",16169.1505],["2026-02-01",15706.8374],["2026-02-08",16041.1964],["2026-02-15",16408.0421],["2026-02-22",18280.9073],["2026-03-01",17520.1283],["2026-03-08",16351.7032],["2026-03-15",16313.8682],["2026-03-22",15002.2762],["2026-03-29",14449.3163],["2026-04-05",13956.3812],["2026-04-12",13781.1315],["2026-04-19",13333.2679],["2026-04-26",13646.5501],["2026-05-03",13842.0744],["2026-05-10",13871.4143],["2026-05-17",13951.7802],["2026-05-24",13722.9028],["2026-05-31",13889.5352],["2026-06-07",10886.2188],["2026-06-14",11553.8317],["2026-06-21",11401.5404],["2026-06-28",10923.4395],["2026-07-05",11493.5589],["2026-07-12",10789.6135],["2026-07-19",11356.3586],["2026-07-26",10167.8837],["2026-08-02",9797.253],["2026-08-09",8635.9891],["2026-08-16",8708.8191],["2026-08-23",13535.2185],["2026-08-30",14201.5931],["2026-09-06",17312.4103],["2026-09-13",18951.6008],["2026-09-20",18951.7494],["2026-09-27",21298.6865],["2026-10-04",18267.426],["2026-10-11",19028.9759]],"code_bytes":10328},"a02":{"id":"a02","alias":"Osprey","theme":"Cross-sectional momentum: rotate into the strongest liquid coins, and decide when to stand aside","started":"2026-10-11T00:00:27+00:00","rounds":[{"k":1,"minutes":53.0},{"k":2,"minutes":44.3},{"k":3,"minutes":31.2},{"k":4,"minutes":10.2}],"messages":[{"t":"2026-10-11T00:00:33.059Z","text":"I’ll inspect the data and simulator, submit a causal momentum baseline early, then test rotation and cash filters across the visible years. I’ll keep a record of the variants and equations in the writeup."},{"t":"2026-10-11T00:01:26.736Z","text":"The first weekly rotation test lost heavily: $100 fell to $12.07 after costs. That makes coin-level exits a priority; the broad BTC filter reacted too slowly. I’ll submit the BTC trend filter as an initial entry, then investigate the rotation losses before adding complexity."},{"t":"2026-10-11T00:03:15.287Z","text":"The BTC baseline is accepted at $230.64. So far, 36 rotation variants made large gains before the evaluation window but lost money during it. The losses cluster around reversals in recent leaders. I’ll test a narrower liquid universe, entry after pullbacks, and breakout exits to see which addresses that failure."},{"t":"2026-10-11T00:05:48.748Z","text":"There’s a useful distinction emerging: ranking capped daily returns, which reduces the influence of single-day spikes, reached $591.51, though with a 63% drawdown. Slow momentum among BTC, ETH and SOL reached $419.64 with a 34% drawdown. I’ll check nearby settings and test whether an online model can identify failed trends using only matured historical returns."},{"t":"2026-10-11T00:09:39.841Z","text":"The second entry is accepted: $100 became $394.13, with a 34.5% drawdown. The online model and broad baskets underperformed it. Nearby major-rotation settings reached about $490–$608, while the strongest capped-return setting relied on one position. I’m checking whether faster execution preserves the major trends through reversals."},{"t":"2026-10-11T00:12:37.089Z","text":"Sentiment looks promising as a cash filter. Requiring the Fear & Greed index’s fast average to exceed its slow average lifted major-coin rotation to roughly $1,040–$1,174, with 23–32% drawdowns. A model-filtered altcoin version reached $1,186 but was sensitive to small threshold changes. I’ll test the simpler sentiment rule’s stability first."},{"t":"2026-10-11T00:14:42.182Z","text":"The sentiment rule passed the causal check and has been submitted: $100 became $1,748.82, with a 29.8% maximum drawdown. Gains persisted across nearby sentiment settings and both short and long momentum horizons. I’ll now check costs, individual-trade dependence, and the timing of entries and exits."},{"t":"2026-10-11T00:18:51.217Z","text":"The $1,748.82 entry is accepted. It still returned $1,574.50 at double costs and $648.27 after removing its five best daily gains. A two-week momentum lookback improved the test to $3,083.43 with a 23.3% drawdown; daily re-selection and faster coin exits also performed well. I’m validating that version before replacing the entry."},{"t":"2026-10-11T00:22:05.346Z","text":"The two-week version is accepted at $3,083.43. Young-coin and volume-spike strategies failed the robustness checks. Choosing the most volatile positive-trend major produced similar returns to choosing the strongest momentum leader. I’m checking the cash filter’s timing and recording the entry’s equations and search history."},{"t":"2026-10-11T00:26:28.028Z","text":"The writeup now documents the accepted rule and its equations. Intraday exits reduced returns in this test. The current entry retained $2,778.86 at double costs, and each rolling annual window was profitable. I’m testing whether broad market participation can support earlier entries than BTC’s slow price filter alone."},{"t":"2026-10-11T00:27:23.426Z","text":"Broad market participation gave only a small improvement, so it does not yet justify another filter. I’m checking the three-day rotation’s starting phase and an online blend of momentum clocks. Both checks will show how much the result depends on the chosen schedule."},{"t":"2026-10-11T00:29:46.797Z","text":"The three-day starting-phase tests returned $2,560–$3,083, so the gain persisted across all three offsets. Removing sentiment reduced the result to $860; removing BTC’s price filter increased drawdown. The online clock blend also underperformed. The evidence still favors a simple fixed rule."},{"t":"2026-10-11T00:30:47.390Z","text":"Adding volatility to the momentum ranking passed the causal check at $3,415.19, with a 25.4% drawdown. It favors upside exposure among coins that already have positive trends. Nearby volatility windows and ranking weights also performed well. I’ll submit this modest improvement; broad one-day reversal tests lost heavily."},{"t":"2026-10-11T00:32:39.620Z","text":"The ranking blend is accepted at $3,415.19. Its robustness checks leave $3,077.86 at double costs and $1,265.99 after removing the five best daily gains. The three starting phases returned $2,957–$3,415. I’m updating the writeup to this version and testing a few distinct intraday ideas before settling the round’s entry."},{"t":"2026-10-11T00:33:10.845Z","text":"I found a tie-handling difference between the submitted ranking rule and the parameter-test helper. I’ve corrected the helper and verified that its default output now matches the entry exactly. I’m rerunning the neighbor checks; the referee’s accepted $3,415.19 result is unchanged."},{"t":"2026-10-11T00:35:01.808Z","text":"The corrected neighbor tests returned $2,014–$3,830, and the writeup now matches the accepted entry. Hourly breakout and reversal strategies were weaker after costs. I’m testing one final broad-universe idea: qualify coins by the past performance of a causal trend rule, then rank the qualified coins on current leadership."},{"t":"2026-10-11T00:36:21.614Z","text":"Coin-specific trend qualification improved some broad-universe tests, but their drawdowns stayed high and their earlier-period results were weak. I’ll keep the accepted major-coin rule. The remaining filter check is hysteresis: whether separate sentiment entry and exit thresholds reduce repeated cash transitions without missing the strongest phases."},{"t":"2026-10-11T00:37:46.422Z","text":"Sentiment hysteresis reduced growth, and the intraday lead–lag tests did not approach the daily entry. I’ll keep the accepted rule. The next useful check is statistical: measure how dependent its gains are on coin-specific rallies and assess that evidence alongside the large number of variants tried."},{"t":"2026-10-11T00:39:55.734Z","text":"Attribution shows the strategy’s main dependence: SOL contributed about 88% of log growth. Removing the best complete trade leaves about $1,252; removing the best three leaves about $597. A diagnostic that accounts for roughly 900 trials gives much weaker evidence than the raw Sharpe ratio. I’ll document that selection risk explicitly."},{"t":"2026-10-11T00:41:18.433Z","text":"The final broad-universe qualification test also underperformed the majors. I’m keeping submission 005 and checking the final file directly. It is a 1.5 KB standalone strategy, and the writeup includes the failed approaches, concentration, costs, schedule sensitivity and multiple-testing limits."},{"t":"2026-10-11T00:42:33.595Z","text":"I’m testing one separate regime idea while the accepted entry remains in place: a short rebound trade after extreme fear starts easing. It could use cash during a bear-market recovery. The rule will depend on the supplied sentiment and price series, with the same settings tested across earlier and visible data."},{"t":"2026-10-11T00:43:53.846Z","text":"The panic-rebound overlay reduced returns to at most about $2,609 and increased drawdown to roughly 39%. That supports keeping the cash filter. The final file reproduces the accepted $3,415.19 result and passes the causal checks; I’m adding the rejection and the latest trial count to the writeup."},{"t":"2026-10-11T00:45:28.020Z","text":"I’ll test very short broad-universe momentum separately. Most earlier tests ranked multi-week leaders, which can enter after a pump is mature. A one-day continuation rule may capture a different phase. It will need to beat costs and work beyond a few exceptional episodes before I consider it."},{"t":"2026-10-11T00:47:06.440Z","text":"Very short broad momentum was also unstable: several settings made enormous earlier-period gains and then lost heavily in the visible years. The accepted rule performed more consistently. Over the separate, roughly five-year window from 2019-10-01 to 2024-10-09, it turned $100 into about $412,389. That is an exploratory fitted result, not the competition’s five-year outcome."},{"t":"2026-10-11T00:48:21.081Z","text":"The long-history check suggests the accepted rule can compound across very different market phases, though it still falls short of the target. I’m checking one last short-momentum filter: require the coin to close near its daily high and exclude exceptionally large daily jumps. That tests whether candle shape helps avoid late pump entries."},{"t":"2026-10-11T00:49:55.968Z","text":"The candle-shape filter failed too. I’m testing a small model on the three majors to decide whether to hold cash or accept the existing ranking. It trains on past returns only after each return is fully observed. The submitted rule remains unchanged while I compare it across periods."},{"t":"2026-10-11T00:51:18.303Z","text":"The sentiment-delay check found a material sensitivity: an extra day of delay cut the visible result to $946.71, though it remained profitable. The entry uses the arena’s stated timing correctly. I’ll document the dependence on prompt sentiment data; the simpler gate is still outperforming the learned cash model so far."},{"t":"2026-10-11T00:52:28.441Z","text":"The learned cash filter finished below the accepted rule: its best visible result was about $3,045, with much weaker earlier-period growth. I’ll retain submission 005. I’m closing the research record and checking that the final writeup matches the code, accepted metrics and latest robustness results."},{"t":"2026-10-11T00:53:29.062Z","text":"Entry: accepted submission 005 — strategy.py, writeup.  \nVisible years: **$100 → $3,415.19**, 25.4% max drawdown; causal checks passed.  \nNext idea: adapt the sentiment lookback using only past strategy performance."},{"t":"2026-10-11T00:53:36.323Z","text":"I’ll keep the accepted entry as a baseline and investigate a different source of returns: whether coins show sustained gains during quiet markets, before they become momentum leaders. I’ll compare earlier and visible periods, then update the writeup with what holds up."},{"t":"2026-10-11T00:56:00.576Z","text":"The current entry earns most of its return from one major coin. I’m testing broader ways to detect demand: strength while BTC is quiet or falling, and sustained buying pressure measured from price and volume. The key check is whether these signals work across both available periods."},{"t":"2026-10-11T00:57:06.500Z","text":"The first broader tests fell well short: the best turned $100 into about $385, versus $3,415 for the accepted entry. I’m now testing whether a liquidity-ranked universe can replace the fixed three-coin universe, so the strategy can follow changes in market leadership."},{"t":"2026-10-11T00:59:15.440Z","text":"Liquidity-based rotation improved on the first tests, but its best result was about $1,573 with 44% drawdown—still weaker than the entry. I’m testing two different approaches next: a four-hour model of changing price drift, and a rule that uses improving sentiment to enter but lets price decide when to exit."},{"t":"2026-10-11T01:01:34.633Z","text":"Neither the drift model nor the longer holding rule beat the entry. I’m testing a walk-forward model that estimates large-gain and large-loss probabilities, plus an early-recovery rule for periods when BTC remains below its slow trend. Both can be checked causally across the earlier and visible data."},{"t":"2026-10-11T01:04:59.939Z","text":"The early-recovery overlay reduced visible returns. A second overlay bought rising coins with low BTC beta; its best result came close to the entry at $3,310, but raised drawdown to 38%. I’m checking whether a smaller allocation or a filter based on its past performance improves that tradeoff."},{"t":"2026-10-11T01:08:06.662Z","text":"Smaller allocations added only about 1% to the result, and the tail model weakened sharply in the earlier period. I’m leaving the entry intact. The next test builds market-demand signals from breadth and trading volume, to see whether they can identify profitable periods the sentiment filter misses."},{"t":"2026-10-11T01:09:19.135Z","text":"The market-demand proxies also underperformed. I’m now testing peer confirmation: infer groups from past return correlations, then require a coin’s rally to have support from its peers. This addresses the pump reversals that hurt broad momentum, without hand-picking sectors or coins."},{"t":"2026-10-11T01:15:03.994Z","text":"One simple change improved both periods: choose the coin with the strongest *weaker* rank across momentum and volatility, rather than averaging the two ranks. It reached $3,827 with the same 25.4% drawdown, and also improved earlier-period growth. I’m checking nearby parameters and costs before submitting it."},{"t":"2026-10-11T01:18:43.908Z","text":"Submission 006 is accepted: $100 became $3,827.05, with 25.4% drawdown. All 21 nearby checks remained profitable, and doubling costs left $3,435. I’m updating the writeup to match the entry, while a separate test checks whether coordinated BTC/ETH price shocks predict tradable moves in the three majors."},{"t":"2026-10-11T01:23:24.591Z","text":"The fifteen-minute shock strategy peaked at about $230, so it stays out of the entry. I’m finishing two different tests: filtered market-regime probabilities and momentum among liquid coins with low unit prices. The latter tests a possible investor bias toward “cheap” coins; its result needs support across both periods."},{"t":"2026-10-11T01:30:02.350Z","text":"One timing change improved returns and reduced drawdown: when a holding fails its price filter, rotate immediately into another qualified coin instead of waiting for the next scheduled selection. That reached $4,075 with 22.0% drawdown and improved earlier-period growth. I’m validating this simpler rule as the next entry."},{"t":"2026-10-11T01:35:09.581Z","text":"Submission 007 is accepted at $4,075.31, with 22.0% drawdown; the writeup now matches it. I’m keeping it while testing a broader entry rule: avoid coins already far above their recent trend, then allow rallies bought early to keep running. This separates entry qualification from holding qualification."},{"t":"2026-10-11T01:37:47.220Z","text":"Entry: #007 accepted; strategy.py and writeup updated.  \nVisible: $100 → **$4,075.31** (×40.75), **22.0% drawdown**; causal checks passed.  \nNext idea: adapt trend horizons using causal change-point detection."},{"t":"2026-10-11T01:37:56.079Z","text":"I’ll keep submission 007 as the baseline and test whether a causal change-point model can replace the fixed trend clocks. I’ll also test a separate entry rule based on liquidity growth, to look for emerging leaders before price momentum becomes extreme."},{"t":"2026-10-11T01:40:47.893Z","text":"The current entry is still $4,075 with a 22% drawdown. I’m testing two alternatives: a filter that estimates when market drift changes, and a rotation rule that looks for sustained trading-volume growth after removing the effect of rising prices."},{"t":"2026-10-11T01:42:57.357Z","text":"Both new approaches fell short. The change-point filter’s best variant reached $3,353; sustained liquidity growth reached $288 and suffered deeper losses. I’m now checking the discrete rank ties in the current rule and a separate idea: selecting coins by recovery from a BTC-defined market peak."},{"t":"2026-10-11T01:45:44.067Z","text":"The recovery rules also lagged the current entry, and changing the tie-breaking rule did not improve it. Four-hour execution reproduced the daily result; faster filters reduced returns. I’m testing one more broader-universe approach: requiring leadership across several momentum horizons so a single price burst has less influence."},{"t":"2026-10-11T01:48:18.198Z","text":"Agreement across momentum horizons reduced the effect of isolated spikes, but its best result was $1,563, below the current entry. I’m also testing a universe defined by persistent liquidity and a separate relative-value strategy for periods when the current rule holds cash."},{"t":"2026-10-11T01:51:36.780Z","text":"Persistent-liquidity selection matched the current result only when it chose the same holdings. Relative-value recovery added about 1.4% in the visible years but weakened the earlier-period result, so I’m keeping submission 007. The bubble-detection rules also failed; I’m recording these results in the writeup while testing rank-based sentiment signals."},{"t":"2026-10-11T01:55:31.788Z","text":"The ordinal sentiment tests also lagged the original signal. The writeup now records 380 round-three evaluations across ten families, including the controls and failures. I’m testing one final distinct approach: a pooled pairwise model that learns which major tends to lead next, using only fully observed training outcomes."},{"t":"2026-10-11T01:59:18.242Z","text":"The pairwise learning model also fell short: its best result was $1,183. I’m now testing whether rising volatility should count as upside capacity only when recent price moves are mostly upward. This directly checks the assumption behind the current ranking rule."},{"t":"2026-10-11T02:03:05.102Z","text":"The capacity guard lowered drawdown in one variant but also reduced growth. I’m testing a more closely matched learning target now: the return each eligible major would have earned over the rule’s next holding interval, including its existing exits and costs. Training will use only intervals that have already finished."},{"t":"2026-10-11T02:05:17.305Z","text":"Learning from completed holding intervals did not improve the entry either. I’m making one final sizing test before the round ends: a small position during a volatility-defined recovery, while the original rule is still waiting for BTC’s 100-day price confirmation."},{"t":"2026-10-11T02:07:01.316Z","text":"I’m keeping submission 007 for this round. The small recovery positions reduced visible returns, and the forecast’s roughly 4% gain was too weak across neighboring settings to justify the added model. The writeup now records all 444 round-three evaluations and their equations; I’m finishing the statistics and checking that the canonical file matches the accepted entry."},{"t":"2026-10-11T02:08:59.407Z","text":"Entry: accepted 007 retained; strategy and writeup updated with 444 research evaluations.  \nVisible years: $100 → **$4,075.31 (×40.75)**; drawdown **21.97%**; causality checks passed.  \nNext idea: intraday trailing exits based on volatility."},{"t":"2026-10-11T02:09:06.319Z","text":"I’ll test the intraday trailing exits, then choose the final entry with enough time for scoring. I’ll also finish the writeup and verify that it describes the exact accepted file."},{"t":"2026-10-11T02:11:01.304Z","text":"The wider trailing stops leave the result unchanged. Some tighter stops trade more often, so I’m comparing their gains and losses before deciding whether they help. I’m also checking whether a longer view of volatility gives a more stable measure of upside capacity."},{"t":"2026-10-11T02:14:00.238Z","text":"The trailing stops did not improve the entry: wide stops left returns unchanged, while tighter stops reduced growth or increased drawdown. I’ll retain submission 007 unless the capacity tests show a clear improvement across both historical windows. The writeup will also flag the dependence on SOL and timely Fear & Greed readings."},{"t":"2026-10-11T02:14:37.508Z","text":"The capacity tests offered no convincing replacement. The best reached $4,141.98, but had a larger drawdown and weaker earlier-period growth; robust estimates that remove large jumps also lagged. I’m keeping submission 007 and finishing the writeup’s results, equations, and research count."},{"t":"2026-10-11T02:16:44.911Z","text":"The final run reproduced $4,075.31 from $100, with 21.97% maximum drawdown, and passed all three causal checks. The writeup now records all 2,184 completed evaluations and the rejected alternatives. I’m checking the stored submission receipt against the final source to confirm which file will be judged."},{"t":"2026-10-11T02:19:10.277Z","text":"My final entry remains accepted submission **007**, identical to strategy.py. It rotates BTC, ETH and SOL using momentum–volatility rank agreement, holding cash when price or sentiment filters fail. Visible years turned **$100 into $4,075.31 (40.75×)** after costs, with **21.97% maximum drawdown**. Final causal checks passed, and the stored accepted source matches exactly. WRITEUP.md is complete with equations, 2,184 recorded evaluations, rejected ideas, results and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T00:01:28.033518+00:00","note":"Initial causal BTC 200-day trend filter; cross-sectional research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:07:53.542730+00:00","note":"Weekly rotation among BTC, ETH and SOL by 180-day volatility-adjusted momentum; daily 20-day EMA exits.","status":"ok","dev":394.1268504757573,"dd":-0.34485926481670337,"reason":""},{"n":"003","t":"2026-10-11T00:14:08.550248+00:00","note":"Momentum leader among BTC, ETH and SOL; exposure requires BTC above its 100-day average and rising Fear & Greed, with a daily coin-trend exit.","status":"ok","dev":1748.815475399441,"dd":-0.2983820967285208,"reason":""},{"n":"004","t":"2026-10-11T00:19:23.418251+00:00","note":"Two-week momentum leader with price and sentiment gates. Nearby horizons/cadences tested; full causal check passed.","status":"ok","dev":3083.4294181629475,"dd":-0.2330798189630222,"reason":""},{"n":"005","t":"2026-10-11T00:30:48.970423+00:00","note":"Equal ranks of two-week momentum and realized volatility among positive-trend majors; common price and sentiment demand gates.","status":"ok","dev":3415.1939311053757,"dd":-0.25397754736863065,"reason":""},{"n":"006","t":"2026-10-11T01:16:56.629727+00:00","note":"Maximin momentum/volatility ranks: causal factor agreement, unchanged demand gates; improves both available periods.","status":"ok","dev":3827.05044671903,"dd":-0.25397754736863076,"reason":""},{"n":"007","t":"2026-10-11T01:30:36.036089+00:00","note":"Immediate replacement after a coin trend exit; maximin ranks and macro demand gates retained. Better growth in both available periods and lower visible drawdown.","status":"ok","dev":4075.3090544138195,"dd":-0.21972735851330616,"reason":""}],"entry":"007","round_notes":[{"k":1,"text":"Entry: accepted submission 005 — strategy.py, writeup.  \nVisible years: **$100 → $3,415.19**, 25.4% max drawdown; causal checks passed.  \nNext idea: adapt the sentiment lookback using only past strategy performance."},{"k":2,"text":"Entry: #007 accepted; strategy.py and writeup updated.  \nVisible: $100 → **$4,075.31** (×40.75), **22.0% drawdown**; causal checks passed.  \nNext idea: adapt trend horizons using causal change-point detection."},{"k":3,"text":"Entry: accepted 007 retained; strategy and writeup updated with 444 research evaluations.  \nVisible years: $100 → **$4,075.31 (×40.75)**; drawdown **21.97%**; causality checks passed.  \nNext idea: intraday trailing exits based on volatility."},{"k":4,"text":"My final entry remains accepted submission **007**, identical to strategy.py. It rotates BTC, ETH and SOL using momentum–volatility rank agreement, holding cash when price or sentiment filters fail. Visible years turned **$100 into $4,075.31 (40.75×)** after costs, with **21.97% maximum drawdown**. Final causal checks passed, and the stored accepted source matches exactly. WRITEUP.md is complete with equations, 2,184 recorded evaluations, rejected ideas, results and generalization limits."}],"title":"Osprey (a02): sentiment-gated leadership","idea":"Rank BTC, ETH and SOL by the weaker of their two-week momentum and twenty-day realized volatility ranks, and own the leading eligible coin while two conditions agree: BTC is above its long price average, and the Fear & Greed index is rising relative to its slower average. Hold cash when either condition fails. Price establishes a sustained market trend; sentiment determines whether current demand is expanding. A separate coin-level trend exit prevents an old relative winner from remaining invested solely because its comparison price is low. Volatility enters positively as a proxy for upside capacity during a favorable demand phase; this differs from the usual inverse-volatility defensive allocation. Its role is conditional: an asset still needs a positive price trend. The initial direction was broad cross-sectional momentum. That direction performed poorly after costs and pump reversals, so the final universe uses the three expressly permitted major coins. The strategy seeks growth through concentration during selected market phases, with cash as its only risk-reduction instrument.","failed":"- Naive weekly broad momentum: the initial top-three 21/63-day rotation with BTC gating finished at $12.07. Recent leaders often reversed sharply after being selected. - Broad momentum with daily exits and rank hysteresis: 36 initial variants lost money in the visible window, although some had very large prehistory gains. Normalizing by volatility and ranking slower returns helped but did not solve the broad-universe problem. - Breakouts and volume confirmation: the tested close-channel and wick filters did not beat the initial BTC baseline. First-attention volume shocks generally performed poorly in both windows. A volume surge frequently arrived after the opportunity rather than before it. - Reversal and pullback rules: trading weak recent performers produced high turnover and large losses. Simple pullback scoring in broad universes also failed. - Broad equal-weight baskets: diversification reduced concentration but did not provide the required growth; the best tested basket ended near $124. - Young listings: a sentiment-filtered diversified version reached about $665, but weak prehistory performance and deep drawdowns made it less credible than the major-coin rule. - Pooled online gradient boosting: labels were purged until their future exit close was observed, with deterministic seeds and rolling training. Six basic model/target combinations finished around $85–$168. A persistence-plus-model filter reached $1,186, but moving a small prediction threshold or shortening the persistence window substantially reduced the result. - Generic four-hour trend rotation: 48 variants failed to beat the accepted daily entry. More frequent price reactions increased whipsaw and switching costs. - Intraday protective exits and breakouts: earlier exits from the same sentiment-gated po","holdup":"The mechanism requires no particular date, named event, future listing, or remembered winner. It uses market trend, current changes in a provided demand proxy, and relative leadership among three liquid coins. Positive volatility ranking seeks more upside when demand is improving; the claim is conditional on trend persistence, and not that volatility is rewarded unconditionally. Cash prevents mechanically buying whichever altcoin has risen most during a broadly weak market. The sentiment gate also exits selected phases earlier than the long price average alone. Coin-level trend and positive-momentum requirements remain separate from the broad gate. Low costs and only about 29% historical exposure leave some room for worse fills. The case for generalization is structural plus historical, not a proof: several sentiment average pairs, momentum horizons, exit speeds, and rotation intervals remained profitable, and the rule survived both the earlier expansion and the visible decline/recovery regime. However, the data were repeatedly inspected, the final parameters were selected retrospectively, and returns are concentrated in high-beta major-coin rallies. Correlation between price momen","result":{"dev":{"final":4075.3090544138195,"max_drawdown":-0.21972735851330616,"trades":112,"exposure":0.2947080291970803,"assets_traded":3,"fees_paid":178.76656331957437},"hidden":{"final":183.24240512968342,"max_drawdown":-0.2584080370672638,"trades":80,"exposure":0.31917808219178084,"assets_traded":3,"fees_paid":12.000687795083508},"full":{"final":7467.69432777564,"max_drawdown":-0.2584080370672638,"trades":192,"exposure":0.3044906900328587,"assets_traded":3,"fees_paid":667.8316796245467}},"freq":"1d","equity":[["2021-10-17",106.4935],["2021-10-24",112.7371],["2021-10-31",111.2689],["2021-11-07",146.8016],["2021-11-14",137.2303],["2021-11-21",136.5555],["2021-11-28",136.5555],["2021-12-05",136.5555],["2021-12-12",136.5555],["2021-12-19",136.5555],["2021-12-26",136.5555],["2022-01-02",136.5555],["2022-01-09",136.5555],["2022-01-16",136.5555],["2022-01-23",136.5555],["2022-01-30",136.5555],["2022-02-06",136.5555],["2022-02-13",136.5555],["2022-02-20",136.5555],["2022-02-27",136.5555],["2022-03-06",136.5555],["2022-03-13",136.5555],["2022-03-20",136.5555],["2022-03-27",153.4488],["2022-04-03",199.8469],["2022-04-10",178.5206],["2022-04-17",178.5206],["2022-04-24",178.5206],["2022-05-01",178.5206],["2022-05-08",178.5206],["2022-05-15",178.5206],["2022-05-22",178.5206],["2022-05-29",178.5206],["2022-06-05",178.5206],["2022-06-12",178.5206],["2022-06-19",178.5206],["2022-06-26",178.5206],["2022-07-03",178.5206],["2022-07-10",178.5206],["2022-07-17",178.5206],["2022-07-24",178.5206],["2022-07-31",178.5206],["2022-08-07",178.5206],["2022-08-14",178.5206],["2022-08-21",178.5206],["2022-08-28",178.5206],["2022-09-04",178.5206],["2022-09-11",178.5206],["2022-09-18",178.5206],["2022-09-25",178.5206],["2022-10-02",178.5206],["2022-10-09",178.5206],["2022-10-16",178.5206],["2022-10-23",178.5206],["2022-10-30",178.5206],["2022-11-06",176.4122],["2022-11-13",169.8568],["2022-11-20",169.8568],["2022-11-27",169.8568],["2022-12-04",169.8568],["2022-12-11",169.8568],["2022-12-18",169.8568],["2022-12-25",169.8568],["2023-01-01",169.8568],["2023-01-08",169.8568],["2023-01-15",247.9334],["2023-01-22",251.3087],["2023-01-29",245.4786],["2023-02-05",242.1856],["2023-02-12",216.318],["2023-02-19",226.3405],["2023-02-26",207.0236],["2023-03-05",207.0236],["2023-03-12",207.0236],["2023-03-19",225.5706],["2023-03-26",230.2271],["2023-04-02",238.5247],["2023-04-09",231.5884],["2023-04-16",247.1493],["2023-04-23",254.8071],["2023-04-30",254.8071],["2023-05-07",245.6773],["2023-05-14",242.0769],["2023-05-21",242.0769],["2023-05-28",242.0769],["2023-06-04",242.0769],["2023-06-11",228.1397],["2023-06-18",228.1397],["2023-06-25",231.966],["2023-07-02",240.1267],["2023-07-09",275.3823],["2023-07-16",344.4799],["2023-07-23",344.2608],["2023-07-30",344.2608],["2023-08-06",344.2608],["2023-08-13",344.2608],["2023-08-20",344.2608],["2023-08-27",344.2608],["2023-09-03",344.2608],["2023-09-10",344.2608],["2023-09-17",344.2608],["2023-09-24",344.2608],["2023-10-01",344.2608],["2023-10-08",344.2608],["2023-10-15",344.2608],["2023-10-22",421.2047],["2023-10-29",454.3897],["2023-11-05",609.2528],["2023-11-12",809.3682],["2023-11-19",948.6466],["2023-11-26",1002.9206],["2023-12-03",1031.6596],["2023-12-10",1171.2425],["2023-12-17",1110.1921],["2023-12-24",1717.9282],["2023-12-31",1689.0067],["2024-01-07",1583.3077],["2024-01-14",1818.8007],["2024-01-21",1816.9749],["2024-01-28",1816.9749],["2024-02-04",1817.9434],["2024-02-11",2023.0095],["2024-02-18",2175.0838],["2024-02-25",2336.4548],["2024-03-03",2638.9905],["2024-03-10",2908.3075],["2024-03-17",3661.8129],["2024-03-24",3936.8578],["2024-03-31",4098.8194],["2024-04-07",4165.5659],["2024-04-14",4106.6851],["2024-04-21",4106.6851],["2024-04-28",4106.6851],["2024-05-05",4106.6851],["2024-05-12",4106.6851],["2024-05-19",4438.2858],["2024-05-26",4911.1431],["2024-06-02",4998.3299],["2024-06-09",4805.4495],["2024-06-16",4662.3794],["2024-06-23",4662.3794],["2024-06-30",4662.3794],["2024-07-07",4662.3794],["2024-07-14",4662.3794],["2024-07-21",4655.6842],["2024-07-28",4932.4507],["2024-08-04",4484.1166],["2024-08-11",4484.1166],["2024-08-18",4484.1166],["2024-08-25",4484.1166],["2024-09-01",4382.8725],["2024-09-08",4382.8725],["2024-09-15",4382.8725],["2024-09-22",4555.2445],["2024-09-29",4406.313],["2024-10-06",4075.3091],["2024-10-13",4075.3091],["2024-10-20",3943.4569],["2024-10-27",4167.1385],["2024-11-03",4052.2788],["2024-11-10",4765.0368],["2024-11-17",4861.077],["2024-11-24",5744.0078],["2024-12-01",5193.4563],["2024-12-08",5370.1077],["2024-12-15",5394.1722],["2024-12-22",4896.8432],["2024-12-29",4896.8432],["2025-01-05",4896.8432],["2025-01-12",4527.6179],["2025-01-19",5611.7222],["2025-01-26",5492.6201],["2025-02-02",4615.7532],["2025-02-09",4611.1374],["2025-02-16",4611.1374],["2025-02-23",4611.1374],["2025-03-02",4611.1374],["2025-03-09",4611.1374],["2025-03-16",4611.1374],["2025-03-23",4611.1374],["2025-03-30",4611.1374],["2025-04-06",4611.1374],["2025-04-13",4611.1374],["2025-04-20",4611.1374],["2025-04-27",4619.8446],["2025-05-04",4680.0907],["2025-05-11",6593.7381],["2025-05-18",6317.6834],["2025-05-25",6458.8886],["2025-06-01",6213.8939],["2025-06-08",6213.8939],["2025-06-15",5578.0813],["2025-06-22",5609.6274],["2025-06-29",5622.3492],["2025-07-06",5400.6319],["2025-07-13",6304.1195],["2025-07-20",7693.614],["2025-07-27",8012.9452],["2025-08-03",7463.8167],["2025-08-10",7923.3418],["2025-08-17",8247.6876],["2025-08-24",8247.6876],["2025-08-31",8247.6876],["2025-09-07",8247.6876],["2025-09-14",8731.1566],["2025-09-21",8626.6911],["2025-09-28",8618.0644],["2025-10-05",8805.2804],["2025-10-12",7288.8067],["2025-10-19",7288.8067],["2025-10-26",7288.8067],["2025-11-02",7288.8067],["2025-11-09",7288.8067],["2025-11-16",7288.8067],["2025-11-23",7288.8067],["2025-11-30",7288.8067],["2025-12-07",7288.8067],["2025-12-14",7288.8067],["2025-12-21",7288.8067],["2025-12-28",7288.8067],["2026-01-04",7288.8067],["2026-01-11",7288.8067],["2026-01-18",7288.8067],["2026-01-25",7288.8067],["2026-02-01",7288.8067],["2026-02-08",7288.8067],["2026-02-15",7288.8067],["2026-02-22",7288.8067],["2026-03-01",7288.8067],["2026-03-08",7288.8067],["2026-03-15",7288.8067],["2026-03-22",7288.8067],["2026-03-29",7288.8067],["2026-04-05",7288.8067],["2026-04-12",7288.8067],["2026-04-19",7288.2233],["2026-04-26",7021.4276],["2026-05-03",7029.8478],["2026-05-10",7426.8633],["2026-05-17",6901.3288],["2026-05-24",6893.6308],["2026-05-31",6893.6308],["2026-06-07",6893.6308],["2026-06-14",6893.6308],["2026-06-21",6893.6308],["2026-06-28",6893.6308],["2026-07-05",6893.6308],["2026-07-12",6893.6308],["2026-07-19",6893.6308],["2026-07-26",6893.6308],["2026-08-02",6893.6308],["2026-08-09",6893.6308],["2026-08-16",6893.6308],["2026-08-23",7405.7848],["2026-08-30",7512.9923],["2026-09-06",7584.3417],["2026-09-13",7540.5148],["2026-09-20",7540.5148],["2026-09-27",8235.002],["2026-10-04",8113.5898],["2026-10-11",7467.6943]],"code_bytes":1470},"a03":{"id":"a03","alias":"Heron","theme":"Volatility targeting and volatility-managed exposure (size inversely to forecast risk)","started":"2026-10-11T00:00:47+00:00","rounds":[{"k":1,"minutes":80.0},{"k":2,"minutes":44.8},{"k":3,"minutes":7.3}],"messages":[{"t":"2026-10-11T00:00:54.243Z","text":"I’ll inspect the data, simulator and example strategies, then submit a causal baseline early. I’ll compare volatility-managed approaches across separate periods so the visible leaderboard doesn’t become the sole reason for choosing a rule."},{"t":"2026-10-11T00:01:56.273Z","text":"The daily data contains 459 assets, including delisted coins, and ends at the permitted cutoff. The first liquid-coin momentum portfolio passed the causality check but lost money after costs. I’ll compare its selection and sizing rules against simpler trend strategies before choosing the early entry."},{"t":"2026-10-11T00:03:03.992Z","text":"The first submission is accepted: $100 became $580.99, with a 49% maximum drawdown. It rotates among BTC, ETH and SOL using positive momentum, caps exposure by estimated volatility, and moves to cash when BTC is below its trend. It gained in each of the three visible years; I’ll now test that result’s stability and whether broader selection helps."},{"t":"2026-10-11T00:04:35.321Z","text":"Broadening the universe has generally hurt returns, even among the most liquid coins. The cheaper three-coin strategy holds up better: nearby momentum lengths and rebalance schedules produced roughly $440–$780 from $100. The strongest year accounts for much of the gain, so I’ll test downside-risk sizing and faster exits next."},{"t":"2026-10-11T00:06:06.136Z","text":"Faster bars haven’t improved this approach: the 4-hour variants trail the daily candidates after costs. Downside-only volatility sizing also adds little. I’ll test an event-driven selector next, which keeps the current holding unless a challenger’s momentum lead is large enough to justify switching."},{"t":"2026-10-11T00:08:03.951Z","text":"I’ve submitted the three-day rebalance version. The local simulator shows $724.86 with a 39.5% maximum drawdown, versus $580.99 and 49.4% for the weekly version. I’m also testing a walk-forward forecasting model using only labels available at each retraining point. It will need to outperform this simpler entry after costs to earn a place in the strategy."},{"t":"2026-10-11T00:11:02.351Z","text":"The walk-forward tree model is weaker than the momentum entry; its best tested allocation produced about $193. Broad-universe momentum and reversal rules also lost substantial money after costs. I’ll keep those as failed experiments and focus on trend persistence and position risk in the lower-cost markets."},{"t":"2026-10-11T00:12:15.180Z","text":"A breakout rule looks promising: in the SOL-only test, it entered at a 14-day high and exited at a 7-day low, producing about $970 with a 36% drawdown. I’ll test the same persistence rule across eligible assets and nearby settings before treating that single-asset result as a candidate."},{"t":"2026-10-11T00:14:08.955Z","text":"The breakout rule improves when it selects the most volatile qualifying asset among BTC, ETH and SOL, then caps exposure using that same volatility estimate. The local result is $1,301.92 with a 34% drawdown, and nearby breakout lengths also performed well. I’ve submitted that version and will check faster exits and return concentration."},{"t":"2026-10-11T00:16:09.835Z","text":"The new entry is accepted at $1,301.92, with a 33.9% maximum drawdown. The comparable 4-hour version is slightly weaker, so daily remains the entry frequency. I’ll now test whether a causal blend of several breakout horizons can reduce dependence on the chosen two-week window."},{"t":"2026-10-11T00:18:31.569Z","text":"The adaptive blend reduced drawdown but earned much less than the simpler breakout rule, so it stays out of the entry. I’m checking new listings and a few risk forecasts next. Those tests address a gap in the earlier research: a 20-day volatility warm-up excluded newly listed assets during their first weeks of trading."},{"t":"2026-10-11T00:25:06.212Z","text":"The three-week breakout with a one-week exit has stronger support in nearby settings: the local result is $1,526.91, and all nine tested neighbours remained profitable. I’ve submitted it. For the preceding entry, doubling costs still left $1,167.54, and removing its ten best days left $293.33. Gains are concentrated in strong rallies, which the writeup will make explicit."},{"t":"2026-10-11T00:30:16.464Z","text":"The current entry passes the full causality check, and the writeup now covers its equations, search history, results and failed ideas. I’m testing the market filter next. It’s the main reason for cash exposure, so I want to see whether another simple filter can improve entry timing across periods without adding fragile rules."},{"t":"2026-10-11T00:35:42.915Z","text":"The exponential BTC filter is accepted: $100 became $1,777.33, with a 29.1% maximum drawdown. It improved on the simple moving average across the visible subperiods and the earlier-history check. I’m finishing tests of volatility-scaled exits and a forecasting model that can evaluate young listings, which the first model excluded."},{"t":"2026-10-11T00:38:42.543Z","text":"Volatility-scaled trailing exits added little beyond the one-week low exit. The second walk-forward model also failed to beat the simple entry after costs. I’ll make one final test of a listing pattern: entering a breakout after volatility has contracted, rather than ranking coins after a large jump."},{"t":"2026-10-11T00:40:45.693Z","text":"The listing-compression rules were weak after costs, so they stay out of the entry. The current strategy remains at $1,777.33 with a 29.1% drawdown. I’ll compare upside and downside volatility as selection signals, then finish the writeup and checks for the final choice."},{"t":"2026-10-11T00:43:04.740Z","text":"The directional-volatility test is promising. Weighting volatility by the share arising from positive returns kept wealth near $1,913 while reducing drawdown to 23.4% in the four-week breakout version. The earlier-period result was also strong. I’ll verify the exact candidate and nearby settings before replacing the entry."},{"t":"2026-10-11T00:46:08.455Z","text":"The candidate passed the causality check, and its tested neighbours stayed profitable. Doubling costs left $1,764.43; removing the ten best days left $431.98. I’ve submitted it and am updating the writeup to match its upside-share score, four-week entry window and exponential market filter."},{"t":"2026-10-11T00:49:25.920Z","text":"Submission #006 is accepted at $1,913.46, with a 23.4% maximum drawdown, and the writeup matches it. I’ll use the remaining research time to test one separate idea: short dip trades during periods when the core strategy holds cash. I’ll keep it only if it adds gains across periods after costs."},{"t":"2026-10-11T00:52:03.135Z","text":"The dip trades improved returns during cash periods, though they increased drawdown. The stronger two-day variant reached about $2,618; the earlier-history check improved too. I’ll test a smaller risk budget, nearby Fear & Greed thresholds and doubled costs before deciding whether to combine it with the core strategy."},{"t":"2026-10-11T00:54:15.323Z","text":"The dip-trading gains are sensitive to the Fear & Greed cutoff, which weakens the case for adding them. I’ll test a trailing percentile of sentiment as a more stable measure of unusual fear, alongside a faster volatility estimate for sizing trades after a sharp loss."},{"t":"2026-10-11T00:57:43.546Z","text":"The percentile rule hasn’t given a consistent improvement, so submission #006 remains the entry. I’ll test one narrower version of the dip idea: trading only unusually large price drops, with holdings lasting a few hours. The aim is to test a rebound signal without carrying prolonged declines."},{"t":"2026-10-11T01:00:33.907Z","text":"The four-hour shock trades add only modest gains, with substantially more turnover, and the result weakens at less extreme thresholds. I’m checking the hourly version and one sizing change to the core strategy: letting the positive-return share adjust its risk budget."},{"t":"2026-10-11T01:02:59.272Z","text":"The sizing changes didn’t materially improve the result, and the hourly shock trades added little despite much higher turnover. I’m keeping submission #006 as the round entry. I’ll now finish the experiment record and check that the writeup, reported figures and submitted file agree."},{"t":"2026-10-11T01:12:54.252Z","text":"The downside-tail forecast prototype also trails the simpler volatility estimate in net wealth. I’m keeping the accepted rule. I’ll add these failures to the writeup and save an experiment index so the roughly 1,600 scored variants can be audited."},{"t":"2026-10-11T01:20:53.646Z","text":"I’ll test a different source of returns: trading activity and price compression ahead of breakouts. I’ll keep the accepted entry in place while checking whether those signals add returns after costs."},{"t":"2026-10-11T01:22:40.373Z","text":"Volume-shock entries performed poorly across the broader coin universe: buying conspicuous activity often caught reversals after the jump. I’ll test whether quieter accumulation offers earlier entries, and separately check short-lived price dislocations in the liquid majors."},{"t":"2026-10-11T01:27:47.472Z","text":"The asset-level breakout experts and relative-strength rules also trail the accepted entry. One separate signal looks more promising: rare, large drops over several hours in BTC, ETH or SOL followed by a short rebound trade. I’ll check nearby thresholds, separate years and higher costs before combining it with the trend strategy."},{"t":"2026-10-11T01:33:46.315Z","text":"The rebound overlay improved wealth across nearby thresholds and holding periods, and a coarser hourly version also added returns. The central 15-minute candidate reaches about $2,504 with roughly 25% drawdown. I’m running the causality and cost checks now; a separate regime filter improved only at a narrow parameter choice, so I’ll leave it out."},{"t":"2026-10-11T01:36:38.420Z","text":"The candidate passed all three causality cuts. It reaches $2,138 at twice the stated costs and $1,826 at three times costs, improving on the trend-only entry under the same cost tests. I’ll submit this version with the five-sigma threshold and six-hour maximum hold; I’ll keep further checks separate."},{"t":"2026-10-11T01:41:45.821Z","text":"Submission #007 is accepted: $100 → $2,503.83. At the same 15-minute cadence, the trend core alone has 25.6% drawdown versus 25.4% with the overlay; finer price marking explains much of the increase from the old daily report. I’ll document the accepted rule and test one final family: predicting large gains after fresh breakouts."},{"t":"2026-10-11T01:46:30.059Z","text":"The larger shared risk budget improved all three fresh-account visible years and still added gains at doubled and tripled costs. I’ll submit 6% for both mechanisms, which removes the separate rebound budget. The learned breakout model was weak—its best tested allocation reached about $136—so I’ll reject it."},{"t":"2026-10-11T01:50:59.536Z","text":"Submission #008 is accepted at $2,819.71, with 25.4% drawdown. The coarse checks still show gains, though one nearby, looser shock rule lost its incremental benefit. I’ll make one final check for this round: whether trading volume helps distinguish a temporary selling burst from a continuing decline."},{"t":"2026-10-11T01:54:54.771Z","text":"Volume conditions added at most about 5.5%, with gains concentrated around one cutoff. Waiting for volume to fade or for a positive candle reduced wealth. I’ll keep #008 and record those limits in the writeup, then check whether exits tied to the original shock improve control of losses."},{"t":"2026-10-11T02:02:25.381Z","text":"Anchored exits added less than 0.5% and often reduced wealth, so #008 stays unchanged. The writeup now documents its equations, results and selection risks. I’ll start the next research direction before the round ends: using pooled risk estimates for new listings that lack enough history for a standard volatility forecast."},{"t":"2026-10-11T02:05:34.024Z","text":"Entry: #008 accepted; strategy.py and WRITEUP.md updated.  \nVisible years: $100 → $2,819.71; max drawdown 25.4%.  \nNext idea: validate first-day listing signals with pooled risk estimates."},{"t":"2026-10-11T02:05:42.233Z","text":"I’ll use the final round to test the early-listing idea against the accepted strategy, especially its returns in separate periods and at higher costs. I’ll leave enough time for scoring and the final writeup, and keep #008 unless the evidence supports a replacement."},{"t":"2026-10-11T02:08:15.546Z","text":"The early-listing prototype looks stronger than the older listing rules, but its best visible result carries substantially more drawdown than #008. I’m checking whether its gains recur across separate periods and survive higher costs before choosing between them."},{"t":"2026-10-11T02:09:09.015Z","text":"The listing audit favours keeping #008. Its extra gains came mainly in the last visible year, and changing the first-day strength threshold reduced them sharply. The shared-risk version made $2,611 versus #008’s $2,820, with greater drawdown. I’m now checking #008’s sensitivity to delayed fills and comparing the rebound’s contribution in each period."},{"t":"2026-10-11T02:12:01.382Z","text":"I’m keeping accepted submission #008. The rebound improved the core’s result in each fresh visible-year account, and the strategy remained profitable at three times stated costs. Delaying every fill by an extra 15 minutes reduced wealth to $2,665; an extra hour removed the rebound advantage. I’ve documented that execution limit and the rejected listing idea in the writeup."},{"t":"2026-10-11T02:12:54.262Z","text":"Final entry remains accepted **#008**, strategy.py: volatility-managed BTC/ETH/SOL breakouts plus brief crash rebounds during cash periods. Visible results: **$100 → $2,819.71 (28.2×)**, with **25.4% maximum drawdown**, after costs. Causality checks passed; tripled costs still yielded $2,025.55. WRITEUP.md is complete with signal equations, parameters, 2,366 configuration records, failed ideas, robustness checks and limitations. Hidden-year performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T00:02:38.007933+00:00","note":"Heron baseline: low-cost 14-day momentum, market trend gate, volatility-capped exposure","status":"ok","dev":580.9942833608492,"dd":-0.49417396727639207,"reason":""},{"n":"002","t":"2026-10-11T00:07:46.080911+00:00","note":"Heron: three-day momentum rotation; lower drawdown across visible years and stronger pre-window performance","status":"ok","dev":724.9344425326136,"dd":-0.39144498034256614,"reason":""},{"n":"003","t":"2026-10-11T00:13:40.582007+00:00","note":"Heron: highest-volatility confirmed breakout, inverse-volatility exposure, market trend gate; generic selection among lower-cost markets","status":"ok","dev":1301.9166738222118,"dd":-0.33878069607588635,"reason":""},{"n":"004","t":"2026-10-11T00:24:12.199613+00:00","note":"Heron: three-week breakout, one-week exit; volatile leader selection with inverse-volatility sizing; robust neighbouring windows","status":"ok","dev":1526.8816197017704,"dd":-0.33878069607588646,"reason":""},{"n":"005","t":"2026-10-11T00:31:47.876254+00:00","note":"Heron: risk-budgeted breakout leaders with exponential market trend; stronger subperiod returns and lower drawdown","status":"ok","dev":1777.330694881204,"dd":-0.2907501171455773,"reason":""},{"n":"006","t":"2026-10-11T00:44:59.685458+00:00","note":"Heron: upside-weighted volatility opportunity in persistent four-week breakouts; 6% daily risk budget and exponential market gate","status":"ok","dev":1913.4592191261152,"dd":-0.23412318247008146,"reason":""},{"n":"007","t":"2026-10-11T01:36:43.170814+00:00","note":"Persistent directional-volatility breakouts plus a five-sigma three-hour rebound during cash periods; six-hour maximum hold, inverse-risk sizing, hourly partial resizing.","status":"ok","dev":2503.8346546546277,"dd":-0.25426681171849475,"reason":""},{"n":"008","t":"2026-10-11T01:46:32.582944+00:00","note":"Directional-volatility trends and rare three-hour five-sigma rebounds; shared 6% daily-equivalent risk budget. Positive incremental returns across fresh visible subperiods and 2x/3x costs.","status":"ok","dev":2819.7054375969396,"dd":-0.25426681171849397,"reason":""}],"entry":"008","round_notes":[{"k":2,"text":"Entry: #008 accepted; strategy.py and WRITEUP.md updated.  \nVisible years: $100 → $2,819.71; max drawdown 25.4%.  \nNext idea: validate first-day listing signals with pooled risk estimates."},{"k":3,"text":"Final entry remains accepted **#008**, strategy.py: volatility-managed BTC/ETH/SOL breakouts plus brief crash rebounds during cash periods. Visible results: **$100 → $2,819.71 (28.2×)**, with **25.4% maximum drawdown**, after costs. Causality checks passed; tripled costs still yielded $2,025.55. WRITEUP.md is complete with signal equations, parameters, 2,366 configuration records, failed ideas, robustness checks and limitations. Hidden-year performance remains unknown."}],"title":"Heron (a03): writeup","idea":"Heron combines persistent upward trends with rare, short rebound trades during the trend strategy's cash periods. Both mechanisms trade only BTC, ETH and SOL, the three explicitly permitted names with the lowest stated costs. The daily trend core enters a four-week high, exits a one-week low, chooses the largest upside-weighted volatility opportunity, and sizes inversely to total volatility while BTC is above its 100-day exponential average. When that core requests cash, a separate 15-minute rule can buy a three-hour decline exceeding five times its prior volatility scale and 1% in magnitude. It holds at most six hours, exits sooner when the three-hour return is no longer negative, and sizes inversely to risk using the same 6% daily-equivalent budget. The core always takes priority. The self-contained entry is deterministic, uses numpy, pandas and numba, and contains no embedded observations, date switches or learned ticker preferences. The final choice is accepted submission #008; its source is unchanged after the final-round audits.","failed":"- Broad liquid-universe momentum, including top-5 through top-150 universes and an unrestricted universe. Most variants lost money after costs; short-horizon winner chasing was especially weak. - Cross-sectional reversal, including extreme-drop selection with and without volatility scaling. Trading costs and continued declines overwhelmed apparent rebounds. - Pure inverse-risk ranking. In the early major-coin comparison, dividing 14-day momentum by volatility produced about $168 versus about $581 for the raw-momentum selection. - Range-only and downside-only forecasts. They changed exposure but did not materially improve the final breakout strategy's wealth. - Market beta, upside capture, downside capture and estimated growth ranks. The best coarse comparison was about $551 and was less stable than breakout selection. - A causal walk-forward histogram gradient-boosting model. It trained monthly on earlier observations with forward labels purged at each retraining point. The best tested allocation was approximately $193, so it was rejected. - A second forecasting model added young listings and learned risk-managed net log payoffs. Its best tested allocation reached only about $118. - New-listing momentum. The strongest unrestricted-risk variant reached about $1,474 but suffered roughly 79% maximum drawdown; its volatility-managed counterpart was much weaker. - Trend-quality ranks using regression fit, median returns and clipped jump-adjusted momentum. The best tested result was about $260. - Online wealth-weighted mixtures of multiple breakout horizons. They reduced drawdown, but their best tested result was about $789. Mixing two stronger signal families also diluted the final breakout rule. - Fear & Greed filters. They helped some early momentum variants but did not co","holdup":"Every transition depends on rolling past prices or a recursive state. The selected asset can change as volatility and breakout eligibility change. The persistent core gives winning trends room to continue while using a shorter exit channel. During its cash periods, the second mechanism seeks a short rebound after an unusually large decline. Temporary selling pressure in liquid markets is a possible explanation; the strategy does not observe liquidation orders or claim that each shock has that cause. The combined rule uses volatility in two distinct ways: as an opportunity proxy after persistent upward confirmation, and as a scale for identifying a short dislocation. Total forecast risk controls exposure in both cases. The mechanisms use different horizons and never compete for capital at the same time. No ticker-specific coefficient says which asset should win either regime. Uncertainty remains substantial. The rule emerged from a large visible-data search, and the three years contain only a few broad cycles. Trend gains depend heavily on large positive days. A volatile asset can suffer a discontinuous loss before an exit fills, and a sideways market can generate repeated false bre","result":{"dev":{"final":2819.7054375969396,"max_drawdown":-0.25426681171849397,"trades":1369,"exposure":0.32643978304918275,"assets_traded":3,"fees_paid":132.0236334480675},"hidden":{"final":175.3387261863043,"max_drawdown":-0.3753817153623603,"trades":119,"exposure":0.356212575379918,"assets_traded":3,"fees_paid":19.148769511554626},"full":{"final":4944.035596488443,"max_drawdown":-0.37538171536235976,"trades":1488,"exposure":0.3383423780116344,"assets_traded":3,"fees_paid":671.9625285982792}},"freq":"15m","equity":[["2021-10-10",99.3309],["2021-10-17",105.691],["2021-10-24",128.9731],["2021-10-31",129.2993],["2021-11-07",159.4462],["2021-11-14",148.0801],["2021-11-21",146.5464],["2021-11-28",145.8063],["2021-12-05",142.12],["2021-12-12",142.12],["2021-12-19",142.12],["2021-12-26",142.12],["2022-01-02",142.12],["2022-01-09",140.6184],["2022-01-16",140.6184],["2022-01-23",140.7662],["2022-01-30",140.7662],["2022-02-06",140.7662],["2022-02-13",140.7662],["2022-02-20",140.7662],["2022-02-27",146.2637],["2022-03-06",146.2637],["2022-03-13",146.2637],["2022-03-20",146.2637],["2022-03-27",156.5865],["2022-04-03",198.672],["2022-04-10",164.7598],["2022-04-17",164.7598],["2022-04-24",164.7598],["2022-05-01",165.5241],["2022-05-08",161.4165],["2022-05-15",178.1992],["2022-05-22",178.1992],["2022-05-29",178.1992],["2022-06-05",178.1992],["2022-06-12",178.1518],["2022-06-19",178.1518],["2022-06-26",178.1518],["2022-07-03",178.1518],["2022-07-10",178.1518],["2022-07-17",178.1518],["2022-07-24",178.1518],["2022-07-31",178.1518],["2022-08-07",178.1518],["2022-08-14",178.1518],["2022-08-21",178.1518],["2022-08-28",178.4226],["2022-09-04",178.4226],["2022-09-11",179.7373],["2022-09-18",181.3562],["2022-09-25",181.3562],["2022-10-02",182.5586],["2022-10-09",182.5586],["2022-10-16",189.2768],["2022-10-23",189.2768],["2022-10-30",189.2768],["2022-11-06",184.1545],["2022-11-13",183.1992],["2022-11-20",183.1992],["2022-11-27",183.1992],["2022-12-04",183.1992],["2022-12-11",183.1992],["2022-12-18",183.1684],["2022-12-25",183.1684],["2023-01-01",198.2731],["2023-01-08",198.2731],["2023-01-15",272.159],["2023-01-22",284.987],["2023-01-29",302.4078],["2023-02-05",277.4287],["2023-02-12",261.7255],["2023-02-19",261.8761],["2023-02-26",233.9196],["2023-03-05",235.0689],["2023-03-12",230.568],["2023-03-19",259.0818],["2023-03-26",258.0845],["2023-04-02",247.6208],["2023-04-09",249.0079],["2023-04-16",291.6062],["2023-04-23",249.1502],["2023-04-30",248.8826],["2023-05-07",248.8826],["2023-05-14",254.4522],["2023-05-21",254.4522],["2023-05-28",254.4522],["2023-06-04",254.4522],["2023-06-11",237.2922],["2023-06-18",236.0161],["2023-06-25",253.5186],["2023-07-02",254.8579],["2023-07-09",270.4278],["2023-07-16",347.3489],["2023-07-23",316.4106],["2023-07-30",316.4106],["2023-08-06",316.4106],["2023-08-13",316.4106],["2023-08-20",333.4296],["2023-08-27",333.4296],["2023-09-03",333.4296],["2023-09-10",337.4835],["2023-09-17",337.4835],["2023-09-24",337.4835],["2023-10-01",337.4835],["2023-10-08",327.8282],["2023-10-15",308.6861],["2023-10-22",372.6285],["2023-10-29",422.9732],["2023-11-05",531.2595],["2023-11-12",719.6736],["2023-11-19",783.1056],["2023-11-26",710.4007],["2023-12-03",755.0715],["2023-12-10",845.0527],["2023-12-17",808.7873],["2023-12-24",1284.0648],["2023-12-31",1154.977],["2024-01-07",1139.6245],["2024-01-14",1167.146],["2024-01-21",1161.8409],["2024-01-28",1155.7326],["2024-02-04",1155.7326],["2024-02-11",1205.3583],["2024-02-18",1260.9695],["2024-02-25",1254.8756],["2024-03-03",1557.6253],["2024-03-10",1689.3118],["2024-03-17",2377.4716],["2024-03-24",2174.7546],["2024-03-31",2395.6214],["2024-04-07",2140.1098],["2024-04-14",2191.7605],["2024-04-21",2191.7605],["2024-04-28",2191.7605],["2024-05-05",2216.1028],["2024-05-12",2216.1028],["2024-05-19",2372.8218],["2024-05-26",2715.4009],["2024-06-02",2688.1562],["2024-06-09",2611.2268],["2024-06-16",2611.2268],["2024-06-23",2727.5303],["2024-06-30",2727.5303],["2024-07-07",2685.9414],["2024-07-14",2685.9414],["2024-07-21",3093.7359],["2024-07-28",3113.0981],["2024-08-04",3005.4905],["2024-08-11",2962.1378],["2024-08-18",2962.1378],["2024-08-25",2962.1378],["2024-09-01",2995.0877],["2024-09-08",2973.7048],["2024-09-15",2973.7048],["2024-09-22",2973.7048],["2024-09-29",3078.8739],["2024-10-06",2819.7054],["2024-10-13",2819.7054],["2024-10-20",3040.4647],["2024-10-27",3226.5116],["2024-11-03",2962.1867],["2024-11-10",3165.0195],["2024-11-17",3558.93],["2024-11-24",3802.4967],["2024-12-01",3825.3761],["2024-12-08",4106.7222],["2024-12-15",4068.5528],["2024-12-22",3619.3311],["2024-12-29",3619.3311],["2025-01-05",3619.3311],["2025-01-12",3539.8556],["2025-01-19",3971.271],["2025-01-26",3890.4022],["2025-02-02",3842.3059],["2025-02-09",3797.158],["2025-02-16",3797.158],["2025-02-23",3876.6773],["2025-03-02",3876.6773],["2025-03-09",3876.6773],["2025-03-16",4000.0629],["2025-03-23",4000.0629],["2025-03-30",4000.0629],["2025-04-06",3895.8146],["2025-04-13",3892.1157],["2025-04-20",3892.1157],["2025-04-27",4127.0992],["2025-05-04",4079.3919],["2025-05-11",5669.1493],["2025-05-18",5726.0984],["2025-05-25",5894.1587],["2025-06-01",5483.897],["2025-06-08",5478.3247],["2025-06-15",4953.0843],["2025-06-22",5007.5144],["2025-06-29",5007.5144],["2025-07-06",5007.5144],["2025-07-13",5230.5727],["2025-07-20",6631.9164],["2025-07-27",6373.6892],["2025-08-03",5969.7845],["2025-08-10",6477.829],["2025-08-17",6854.7868],["2025-08-24",6157.9937],["2025-08-31",5308.6038],["2025-09-07",5113.007],["2025-09-14",5951.7483],["2025-09-21",5862.883],["2025-09-28",5449.9838],["2025-10-05",5670.1639],["2025-10-12",5172.1504],["2025-10-19",5172.1504],["2025-10-26",5172.1504],["2025-11-02",5172.1504],["2025-11-09",5048.0363],["2025-11-16",5048.0363],["2025-11-23",5048.0363],["2025-11-30",5048.0363],["2025-12-07",5004.3824],["2025-12-14",5004.3824],["2025-12-21",4989.932],["2025-12-28",4989.932],["2026-01-04",4989.932],["2026-01-11",4989.932],["2026-01-18",4926.198],["2026-01-25",4946.9927],["2026-02-01",4937.1215],["2026-02-08",5146.1718],["2026-02-15",5146.1718],["2026-02-22",5146.1718],["2026-03-01",5166.235],["2026-03-08",5166.235],["2026-03-15",5166.235],["2026-03-22",5166.235],["2026-03-29",5166.235],["2026-04-05",5166.235],["2026-04-12",5166.235],["2026-04-19",4837.7548],["2026-04-26",4942.0925],["2026-05-03",4675.0108],["2026-05-10",5243.9344],["2026-05-17",4834.0057],["2026-05-24",4821.8681],["2026-05-31",4821.8681],["2026-06-07",4821.8681],["2026-06-14",4821.8681],["2026-06-21",4821.8681],["2026-06-28",4836.8569],["2026-07-05",4836.8569],["2026-07-12",4836.8569],["2026-07-19",4836.8569],["2026-07-26",4836.8569],["2026-08-02",4836.8569],["2026-08-09",4836.8569],["2026-08-16",4836.8569],["2026-08-23",5263.7051],["2026-08-30",5173.6198],["2026-09-06",5421.2215],["2026-09-13",4927.7403],["2026-09-20",4693.4273],["2026-09-27",5144.2228],["2026-10-04",5137.0405],["2026-10-11",4944.0356]],"code_bytes":3440},"a04":{"id":"a04","alias":"Falcon","theme":"Growth-optimal (Kelly / log-utility) position sizing with estimated edges and estimation error","started":"2026-10-11T00:01:07+00:00","rounds":[{"k":1,"minutes":79.8},{"k":2,"minutes":43.9},{"k":3,"minutes":8.1}],"messages":[{"t":"2026-10-11T00:01:14.183Z","text":"I’ll inspect the data and simulator, submit an initial strategy early, then test whether estimation-aware Kelly sizing improves on momentum rules. I’ll work offline and record the variants and failures in the writeup."},{"t":"2026-10-11T00:02:03.014Z","text":"The daily data includes 459 assets, including delisted coins. I’ll start with a liquid-coin momentum portfolio and a market trend filter, then compare it with sizing based on historical continuation returns and estimation uncertainty."},{"t":"2026-10-11T00:03:35.984Z","text":"The first 24 momentum variants lost money after costs; the best ended at $27.27. That’s a useful rejection of the initial approach. I’ve submitted the BTC trend baseline and am now checking which return signals showed continuation or reversal in the earlier history."},{"t":"2026-10-11T00:06:15.009Z","text":"The baseline was accepted at $230.64. A walk-forward model trained only on matured labels also lost money across all 18 allocation variants. Rotation among BTC, ETH and SOL did better, with one version reaching $679, so I’m focusing on robust sizing there while testing whether recent listings behave differently."},{"t":"2026-10-11T00:10:33.485Z","text":"Kelly sizing with estimated conditional edges reduced drawdowns, but hasn’t beaten the baseline reliably. Recent-listing momentum was more promising, though costly and sensitive to parameters. I’m testing whether keeping winners until their trend breaks improves it, and I’ll check how much of the result depends on a few outsized trades."},{"t":"2026-10-11T00:17:20.630Z","text":"The Kelly entry was accepted at $1,003.94, with a 56.9% max drawdown. Nearby parameter choices mostly retained substantial gains, and doubling costs still left about $872. The result depends heavily on strong trend days; I’m testing whether faster updates and a mixture of drift estimates can reduce that dependence."},{"t":"2026-10-11T00:23:15.264Z","text":"A blend of the major-coin Kelly rule and recent-listing trends reached about $1,523, with drawdown reduced to 38.5%. Listing-age choices were sensitive, so I’m testing a causal return model before committing more capital to that component."},{"t":"2026-10-11T00:32:07.012Z","text":"The fixed-horizon discovery model added little. I’m now testing a different estimate: the log wealth earned by several position sizes when an asset is held until its trend breaks. This should handle large gains and losses better than a mean–variance approximation. The tested blend has been submitted while that research runs."},{"t":"2026-10-11T00:33:51.297Z","text":"The blend is accepted at $1,522.78. Direct log-utility learning also failed to establish a reliable discovery edge, so I’m keeping it out of the entry. Next I’m testing whether recent listings need a stronger market regime filter, while checking concentration and costs for the accepted blend."},{"t":"2026-10-11T00:36:42.701Z","text":"The accepted blend retained $1,243 at double costs and $1,014 at triple costs. Its three yearly blocks were profitable, but the ten best days contributed 62% of total log growth. Slower market filters did not improve recent-listing returns. I’m recording those limits in the writeup and testing a common-market prior for the major-coin drift estimates."},{"t":"2026-10-11T00:40:53.371Z","text":"A partial pull toward BTC’s market drift improved the major-only result to about $1,126. With a slower core gate, it retained similar wealth at roughly 30% drawdown. I’m checking how that version combines with discovery positions and whether a faster exit helps when discovery-coin volatility surges."},{"t":"2026-10-11T00:48:15.567Z","text":"Simple filters have held up better than the return-learning models in this data. I’m testing one final sizing layer: allocate between the two strategy components using their realized historical returns, covariance and estimation error. This may adapt the budget without predicting individual coin returns."},{"t":"2026-10-11T00:52:13.060Z","text":"The allocation model reached $2,487, but its best result required removing the estimation penalty. I’m cautious about that sensitivity. I’m comparing it with a simpler rule that transfers part of an inactive component’s budget to the active one while preserving the coin-level Kelly fraction."},{"t":"2026-10-11T00:56:25.039Z","text":"Transferring idle capital in both directions increased drawdown without a consistent gain in wealth. I’m now making the allocation estimate reflect which component is active. I’m also testing a simpler alternative that transfers idle discovery capital only into the major-coin Kelly rule."},{"t":"2026-10-11T01:03:43.353Z","text":"The strongest candidate now reaches about $2,439. It ranks discovery entries by momentum divided by variance, allows a switch to a materially stronger trend, and transfers idle discovery capital into the major-coin Kelly rule. Nearby versions retained roughly $2,050–$2,320. I’m validating and submitting this version, then updating the writeup."},{"t":"2026-10-11T01:09:23.963Z","text":"Submission #004 is accepted at $2,438.99, with a 45.7% max drawdown. It retained $1,904 at double costs and $1,487 at triple costs, and all three yearly blocks were profitable. I’m finishing the writeup, trial inventory and concentration checks before the round closes."},{"t":"2026-10-11T01:17:58.237Z","text":"The writeup now matches accepted submission #004, including the equations, parameters, costs, concentration checks and 626 completed trial configurations. The source matches the submitted file exactly. I’m checking one late diagnostic on volatility-normalized strategy returns for the next round."},{"t":"2026-10-11T01:20:56.672Z","text":"Entry: accepted #004 — strategy.py, writeup.  \nVisible years: **$100 → $2,438.99**, costs included; max drawdown **45.7%**.  \nNext idea: earlier discovery entries sized using range volatility and estimation uncertainty."},{"t":"2026-10-11T01:21:05.036Z","text":"I’ll keep the accepted entry in place while testing a different opportunity: the first two weeks after a listing becomes visible in the data. I’ll use price ranges to estimate risk before much return history exists, and check whether any gains survive costs and parameter changes."},{"t":"2026-10-11T01:24:16.789Z","text":"The first tests of earlier listing entries are losing money, despite the BTC trend filter. I’m checking whether candle shape and trading volume reveal a useful condition before pursuing that idea further."},{"t":"2026-10-11T01:25:54.963Z","text":"Earlier listing entries are too unreliable to add: most versions lose money, and even the best suffer deep drawdowns. I’m moving to a broader liquid-coin strategy that compares estimated log-growth and keeps the incumbent unless a challenger offers a material improvement."},{"t":"2026-10-11T01:27:32.841Z","text":"The broader Kelly tests also underperform; expanding the universe adds too many false signals. I’ve started a purged walk-forward model on four-hour BTC, ETH and SOL bars to test whether volatility, pullbacks and trading pressure predict usable returns after costs."},{"t":"2026-10-11T01:29:26.423Z","text":"The four-hour model is weak so far, largely because its trades are too frequent. I’m also testing two lower-turnover alternatives: buying medium-term winners after a pullback, and entering new listings only after their first auction has closed."},{"t":"2026-10-11T01:31:32.077Z","text":"Neither pullback entries nor first-auction follow-through has beaten the accepted strategy. I’m now checking whether broad market participation is a better gate for discovery trades than BTC’s trend alone, and whether downside risk is a better ranking measure than total variance."},{"t":"2026-10-11T01:32:52.089Z","text":"Two changes now show promise. Ranking discovery candidates by downside risk raises that component to about $2,929, and a market-breadth gate improves the combined strategy across several thresholds. I’ll check their combination, nearby parameters and higher costs before replacing the entry."},{"t":"2026-10-11T01:35:51.344Z","text":"The downside-risk candidate is ahead at nearby lookbacks, age limits and risk floors. Its broader market gate hurts the combination, so I’m leaving that filter out. I’m validating a 50/50 capital split; the extra discovery exposure brings higher drawdown as well as growth."},{"t":"2026-10-11T01:38:27.164Z","text":"The referee’s cost-inclusive run gives the new candidate $3,819.36, up from $2,438.99. Max drawdown rises from 45.7% to 50.4%. I’m checking higher costs and yearly results before submitting it."},{"t":"2026-10-11T01:39:27.396Z","text":"The candidate reaches $2,965 with double costs and $2,302 with triple costs. All three fresh-start yearly tests are positive, with most of the improvement in the third year. I’ve submitted it as #005; the referee is running the causality checks."},{"t":"2026-10-11T01:41:16.126Z","text":"Submission #005 is accepted, and `strategy.py` and the writeup now match it. It ranks #16 of 40 on the visible years. I’ll use the remaining time to test a different way of identifying which liquid assets trend reliably."},{"t":"2026-10-11T01:43:37.619Z","text":"Historical trend reliability, lagged momentum and variance-ratio tests have not improved the entry. I’m checking one remaining cross-asset idea: using demand across the liquid coin market to stabilize the major coins’ noisy drift estimates."},{"t":"2026-10-11T01:45:53.512Z","text":"The common-market factor and four-hour shock entries both fall short of #005. I’m now testing an online allocation rule that learns from several fixed risk budgets’ realized net wealth, which may be less fragile than another return forecast."},{"t":"2026-10-11T01:47:59.624Z","text":"Online wealth weighting gives only a small improvement and adds complexity, so I’m keeping #005. The last new test asks a different question of young coins: whether they reach a 30% gain before a 20% loss. Each training outcome must mature before the model can use it."},{"t":"2026-10-11T01:50:05.292Z","text":"The classifier underperforms overall. I also found that it cannot score coins with fewer than 29 bars because I required every feature to be present. I’ll test its built-in handling of missing history, so younger coins can be assessed from the data already available."},{"t":"2026-10-11T01:52:20.573Z","text":"Handling missing history did not rescue the classifier. I’m keeping #005 and checking its trailing stop: whether using each coin’s observed volatility can reduce losses without cutting off the large winners."},{"t":"2026-10-11T01:53:46.211Z","text":"A fixed 20% trailing stop looks better than the volatility-scaled alternatives: $4,612 in research, with drawdown reduced to 47.5%. I’m checking nearby stop levels and costs before submitting the simpler change."},{"t":"2026-10-11T01:56:00.215Z","text":"Submission #006 is accepted: $4,612.38 with 47.5% drawdown. It also reaches $3,543 with double costs and $2,722 with triple costs. I’m updating the writeup and checking the submitted source against the working file."},{"t":"2026-10-11T01:58:33.817Z","text":"The source audit is clean: the working file matches accepted #006, and discovery entries require observed prices. I’m running one last small test of price-and-volume pressure as an entry filter; the accepted strategy stays in place while it runs."},{"t":"2026-10-11T02:00:32.512Z","text":"Price-and-volume pressure filters did not help. I’m checking a different risk control for the next round: reducing exposure when the strategy’s own causal shadow account weakens. That tests whether the signal is still working, using realized returns rather than prices alone."},{"t":"2026-10-11T02:02:17.389Z","text":"The whole-account equity filter reduces drawdown to about 31%, but lowers ending wealth to $3,207, so I’m keeping #006. I’m checking the same control on each component separately, then recording the results for the next round."},{"t":"2026-10-11T02:03:41.104Z","text":"Applying the equity control only to the major-coin core looks promising: about $5,043 with 38.4% drawdown. I’ve prepared the candidate for validation next round. Accepted #006 remains the entry, and the writeup records both results."},{"t":"2026-10-11T02:04:53.151Z","text":"Entry: accepted #006; strategy and writeup.  \nVisible years: **$100 → $4,612.38** (×46.12), max drawdown **47.5%**.  \nNext idea: validate the core equity filter—$5,043 and 38.4% drawdown in research."},{"t":"2026-10-11T02:04:59.649Z","text":"I’ll finish validating the core equity filter, choose the final entry with time for the referee to score it, and make sure the writeup matches the accepted code."},{"t":"2026-10-11T02:07:09.909Z","text":"The core equity filter raises the visible result from $4,612 to $5,043 and lowers drawdown from 47.5% to 38.4%. I’m checking causality, higher trading costs, and nearby settings before deciding whether to submit it."},{"t":"2026-10-11T02:08:30.923Z","text":"The filter passed all three referee causality checks. The strategy ends at $3,939 with doubled costs and $3,076 with tripled costs; nearby filter settings also remain profitable. I’m submitting the 20-day filter with a 50% floor, then updating the writeup to match."},{"t":"2026-10-11T02:10:25.778Z","text":"Submission #007 is awaiting the referee. The writeup now includes the virtual-account equations, allocation rule, and all 1,207 grid configurations. It also records a limitation: the new filter lowers visible-period drawdown but reduces growth in the earlier 2020–2021 sample."},{"t":"2026-10-11T02:11:55.332Z","text":"Submission #007 is accepted and remains the final entry: $100 became $5,042.57, with 38.43% maximum drawdown. I’ve confirmed that the canonical strategy matches the accepted file byte for byte. The final review is checking the writeup’s equations and results against that file."},{"t":"2026-10-11T02:12:55.494Z","text":"Falcon’s final entry #007 is accepted. It combines uncertainty-penalized half-Kelly sizing, discovery momentum ranked by downside risk, and a causal core-equity filter. Visible years turned **$100 into $5,042.57 (×50.43)** with **38.43% maximum drawdown**. Causality checks passed; tripled trading costs still left $3,076. The completed WRITEUP.md documents every equation, 1,207 tested configurations, results, failed ideas, and limitations for unseen performance."}],"submissions":[{"n":"001","t":"2026-10-11T00:03:05.702435+00:00","note":"Initial causal BTC 200-day trend baseline; research entry will replace it.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:11:31.147158+00:00","note":"Estimation-aware half-Kelly rotation across BTC/ETH/SOL, cost hysteresis, daily market gate; fully causal.","status":"ok","dev":1003.9368722721639,"dd":-0.5693832419925509,"reason":""},{"n":"003","t":"2026-10-11T00:27:18.860391+00:00","note":"60% estimation-aware major-coin Kelly, 40% data-selected discovery trends; cost-aware persistence and causal weekly refresh.","status":"ok","dev":1522.7794867035561,"dd":-0.3846257100868651,"reason":""},{"n":"004","t":"2026-10-11T01:04:33.737495+00:00","note":"Half-Kelly majors plus variance-ranked discovery rotation, a weekly switch band and capped discovery capital; causal, generic spot rules.","status":"ok","dev":2438.9931961712273,"dd":-0.45682441490162096,"reason":""},{"n":"005","t":"2026-10-11T01:39:18.439115+00:00","note":"Falcon: adverse-return discovery ranking, 20-day exit, balanced capital; three visible years positive and double/triple costs checked.","status":"ok","dev":3819.361575222837,"dd":-0.5044594429697052,"reason":""},{"n":"006","t":"2026-10-11T01:54:05.947836+00:00","note":"Falcon: same adverse-risk discovery and half-Kelly core; 20% trailing stop improves all three visible blocks and drawdown. Nearby stops and double/triple costs checked.","status":"ok","dev":4612.3816417501985,"dd":-0.47509025384992043,"reason":""},{"n":"007","t":"2026-10-11T02:08:32.689516+00:00","note":"Falcon: uncertainty-penalized half-Kelly core, adverse-risk discovery, and causal core equity exposure control","status":"ok","dev":5042.569726719474,"dd":-0.3842855721264202,"reason":""}],"entry":"007","round_notes":[{"k":1,"text":"Entry: accepted #004 — strategy.py, writeup.  \nVisible years: **$100 → $2,438.99**, costs included; max drawdown **45.7%**.  \nNext idea: earlier discovery entries sized using range volatility and estimation uncertainty."},{"k":2,"text":"Entry: accepted #006; strategy and writeup.  \nVisible years: **$100 → $4,612.38** (×46.12), max drawdown **47.5%**.  \nNext idea: validate the core equity filter—$5,043 and 38.4% drawdown in research."},{"k":3,"text":"Falcon’s final entry #007 is accepted. It combines uncertainty-penalized half-Kelly sizing, discovery momentum ranked by downside risk, and a causal core-equity filter. Visible years turned **$100 into $5,042.57 (×50.43)** with **38.43% maximum drawdown**. Causality checks passed; tripled trading costs still left $3,076. The completed WRITEUP.md documents every equation, 1,207 tested configurations, results, failed ideas, and limitations for unseen performance."}],"title":"Falcon (a04): uncertain drift and discovery trends","idea":"Use a cost-aware, estimation-penalized half-Kelly rule to select one of BTC, ETH and SOL, and allocate at most 50% of a submitted account target to a data-selected discovery trend. Discovery entries rank recently observed, liquid assets by two-week log momentum divided by the second moment of adverse log returns. Hold until the 20-day trend or 20% closing-price peak stop breaks, or switch after a week to a materially stronger candidate. When discovery is active, the core receives the remaining 50% budget; when discovery is in cash, the core can use the entire budget. A separate causal virtual account tracks the unfiltered core after trading costs. When that account falls below its 20-day equity EMA, halve the core's indicated exposure. Both components also require a positive BTC price trend. Discovery ranking and equity control are hypotheses about persistence, rather than calibrated expected-return forecasts or exact Kelly optimizers. The fixed discovery budget and fractional core sizing address estimation uncertainty.","failed":"- Broad top-volume momentum rotation: 24 initial variants, best $27.27. Extreme past winners frequently reversed. Expanded liquidity and risk-adjusted rankings usually did not fix it. - Daily loser buying and short-term reversal: 48 variants, best about $10.75. Costs and persistent declines overwhelmed the intended rebound. - Walk-forward gradient boosting on matured seven-day log-return labels: one model specification with 18 allocation variants, best about $40.64. Fractional Kelly did not rescue the forecast. - Day-clustered Bayesian trend-state estimates: 18 variants, best $248.73. More conservative settings improved drawdown but lost too much growth. - New-asset Bayesian regression with date-balanced samples and overlapping-label uncertainty: 18 variants, best $120.12. More error penalization reduced losses without establishing useful alpha. - Empirical stopped-trade log utility over fractions 0.1/0.25/0.5/0.75/1, fully matured labels, date-balanced kernel weights and overlap-adjusted uncertainty: 24 variants, best $279.19. Mathematically closer to exact Kelly, but materially weaker than the selected trend rule. - Persistent breakouts near historical highs: 48 variants, best $438.75, with high drawdown. - Four-hour Kelly and Bayesian Kalman model averaging: some variants reached approximately $900–$1,150 with better drawdown, but higher costs and no compelling advantage in growth. - Strategy-level Kelly budgets used causal shadow account returns, EWMA means/covariances, cash and a fixed allocation floor. The best unpenalized variant reached $2,487, but its advantage was sensitive to estimation-error penalties. Activity-conditioned estimates did not beat the final rule. I chose the simpler capital budget rather than adding a second noisy return estimator. - Normalizi","holdup":"The working hypothesis is that broad demand and herding persist for weeks, while newly observed assets sometimes undergo extended price discovery. The market gate, volatility estimate and ranked universe respond to current observations. The rule needs neither a fixed token roster nor calendar-specific knowledge. Retaining a winning discovery position prevents mechanically selling it when it becomes older or when another asset has a temporarily higher score. Cash, fractional sizing and cost bands limit the compounding damage of noisy drift estimates and turnover. The core equity diagnostic measures the recent success of the actual forecasting-and-trading rule after costs, which can disagree with the BTC price trend. An independent virtual account keeps observing the core while its real budget is reduced, so the strategy can restore exposure without needing profitable trades from an inactive actual account. Its short EMA and fractional floor permit recovery, but also introduce lag, threshold whipsaws and extra trades. The mechanism is justified as feedback under uncertain estimates, not by any later market event. Different growth mechanisms helped at different times in the visible sa","result":{"dev":{"final":5042.569726719474,"max_drawdown":-0.3842855721264202,"trades":518,"exposure":0.40233294389084323,"assets_traded":56,"fees_paid":308.30751980521524},"hidden":{"final":84.54792011802739,"max_drawdown":-0.5930709794918805,"trades":392,"exposure":0.42005472770988567,"assets_traded":36,"fees_paid":21.862090310965662},"full":{"final":4256.841760183136,"max_drawdown":-0.5930709794918807,"trades":911,"exposure":0.4094177753190475,"assets_traded":88,"fees_paid":1417.3085447399287}},"freq":"1d","equity":[["2021-10-17",96.2828],["2021-10-24",102.9917],["2021-10-31",92.0968],["2021-11-07",100.4314],["2021-11-14",93.6275],["2021-11-21",84.5862],["2021-11-28",84.5862],["2021-12-05",84.5862],["2021-12-12",84.5862],["2021-12-19",84.5862],["2021-12-26",84.5862],["2022-01-02",84.5862],["2022-01-09",84.5862],["2022-01-16",84.5862],["2022-01-23",84.5862],["2022-01-30",84.5862],["2022-02-06",84.5862],["2022-02-13",77.8947],["2022-02-20",80.9798],["2022-02-27",80.9798],["2022-03-06",99.8588],["2022-03-13",91.9041],["2022-03-20",93.4231],["2022-03-27",89.7073],["2022-04-03",242.5715],["2022-04-10",234.6113],["2022-04-17",233.3659],["2022-04-24",233.3659],["2022-05-01",233.3659],["2022-05-08",233.3659],["2022-05-15",233.3659],["2022-05-22",233.3659],["2022-05-29",233.3659],["2022-06-05",233.3659],["2022-06-12",233.3659],["2022-06-19",233.3659],["2022-06-26",233.3659],["2022-07-03",233.3659],["2022-07-10",233.3659],["2022-07-17",233.3659],["2022-07-24",265.3572],["2022-07-31",318.9497],["2022-08-07",322.9144],["2022-08-14",323.207],["2022-08-21",262.7995],["2022-08-28",262.7995],["2022-09-04",262.7995],["2022-09-11",262.7995],["2022-09-18",246.9776],["2022-09-25",246.9776],["2022-10-02",246.9776],["2022-10-09",245.6421],["2022-10-16",245.6421],["2022-10-23",245.6421],["2022-10-30",263.0209],["2022-11-06",259.6025],["2022-11-13",211.5045],["2022-11-20",211.5045],["2022-11-27",211.5045],["2022-12-04",211.5045],["2022-12-11",211.5045],["2022-12-18",211.5045],["2022-12-25",211.5045],["2023-01-01",211.5045],["2023-01-08",207.0406],["2023-01-15",360.5892],["2023-01-22",386.5345],["2023-01-29",464.9973],["2023-02-05",535.5125],["2023-02-12",468.8869],["2023-02-19",505.1963],["2023-02-26",456.4167],["2023-03-05",447.7635],["2023-03-12",447.7635],["2023-03-19",414.565],["2023-03-26",358.9361],["2023-04-02",367.0508],["2023-04-09",356.6413],["2023-04-16",466.7408],["2023-04-23",404.2475],["2023-04-30",382.2676],["2023-05-07",377.3511],["2023-05-14",374.4285],["2023-05-21",374.4285],["2023-05-28",374.4285],["2023-06-04",374.4285],["2023-06-11",374.4285],["2023-06-18",374.4285],["2023-06-25",411.2213],["2023-07-02",419.5573],["2023-07-09",454.4356],["2023-07-16",570.4333],["2023-07-23",509.3599],["2023-07-30",483.5242],["2023-08-06",483.5242],["2023-08-13",483.5242],["2023-08-20",483.5242],["2023-08-27",483.5242],["2023-09-03",483.5242],["2023-09-10",483.5242],["2023-09-17",483.5242],["2023-09-24",513.5014],["2023-10-01",544.758],["2023-10-08",587.0778],["2023-10-15",526.1499],["2023-10-22",594.7876],["2023-10-29",643.2853],["2023-11-05",963.0791],["2023-11-12",1072.6949],["2023-11-19",1140.4004],["2023-11-26",1184.8542],["2023-12-03",1380.9233],["2023-12-10",1754.4587],["2023-12-17",1935.5101],["2023-12-24",2322.5712],["2023-12-31",2260.1677],["2024-01-07",1999.4109],["2024-01-14",1819.0119],["2024-01-21",1655.8396],["2024-01-28",1655.8396],["2024-02-04",1617.9814],["2024-02-11",1775.8241],["2024-02-18",1875.0023],["2024-02-25",2027.1692],["2024-03-03",2129.2108],["2024-03-10",2355.1008],["2024-03-17",3566.0648],["2024-03-24",3221.4401],["2024-03-31",3674.5214],["2024-04-07",3514.7671],["2024-04-14",3529.6611],["2024-04-21",3529.6611],["2024-04-28",3529.6611],["2024-05-05",3529.6611],["2024-05-12",3529.6611],["2024-05-19",3834.8831],["2024-05-26",3791.2592],["2024-06-02",3843.1838],["2024-06-09",3572.0563],["2024-06-16",3850.3688],["2024-06-23",3471.5195],["2024-06-30",3471.5195],["2024-07-07",3471.5195],["2024-07-14",3471.5195],["2024-07-21",4147.4783],["2024-07-28",3990.3567],["2024-08-04",3530.0816],["2024-08-11",3530.0816],["2024-08-18",3530.0816],["2024-08-25",3714.0385],["2024-09-01",3490.017],["2024-09-08",3490.017],["2024-09-15",3490.017],["2024-09-22",3597.9405],["2024-09-29",3727.6369],["2024-10-06",3460.0256],["2024-10-13",5306.4656],["2024-10-20",5145.2766],["2024-10-27",4651.9635],["2024-11-03",4520.6612],["2024-11-10",5376.22],["2024-11-17",5089.674],["2024-11-24",4687.8717],["2024-12-01",5359.4791],["2024-12-08",6427.5197],["2024-12-15",5793.9967],["2024-12-22",7218.5273],["2024-12-29",7458.6163],["2025-01-05",7694.9022],["2025-01-12",7146.3964],["2025-01-19",7299.9113],["2025-01-26",6944.872],["2025-02-02",5823.995],["2025-02-09",5821.2461],["2025-02-16",5821.2461],["2025-02-23",5821.2461],["2025-03-02",5821.2461],["2025-03-09",5821.2461],["2025-03-16",5821.2461],["2025-03-23",5821.2461],["2025-03-30",5821.2461],["2025-04-06",5821.2461],["2025-04-13",5821.2461],["2025-04-20",5906.7702],["2025-04-27",6566.9674],["2025-05-04",8281.3297],["2025-05-11",8099.4244],["2025-05-18",7928.6928],["2025-05-25",9047.2707],["2025-06-01",6940.179],["2025-06-08",6333.8805],["2025-06-15",6127.4553],["2025-06-22",6005.2992],["2025-06-29",6823.9673],["2025-07-06",5906.5401],["2025-07-13",6437.2976],["2025-07-20",6291.4341],["2025-07-27",6215.0278],["2025-08-03",5295.334],["2025-08-10",6164.5654],["2025-08-17",6799.46],["2025-08-24",6090.312],["2025-08-31",6087.2839],["2025-09-07",6087.2839],["2025-09-14",6584.9684],["2025-09-21",6141.2204],["2025-09-28",5342.7046],["2025-10-05",5381.2746],["2025-10-12",4723.8905],["2025-10-19",4723.8905],["2025-10-26",4723.8905],["2025-11-02",4504.0321],["2025-11-09",4504.0321],["2025-11-16",4504.0321],["2025-11-23",4504.0321],["2025-11-30",4504.0321],["2025-12-07",4504.0321],["2025-12-14",4504.0321],["2025-12-21",4504.0321],["2025-12-28",4504.0321],["2026-01-04",4771.4229],["2026-01-11",4333.209],["2026-01-18",4484.3619],["2026-01-25",4251.3058],["2026-02-01",4251.3058],["2026-02-08",4251.3058],["2026-02-15",4251.3058],["2026-02-22",4251.3058],["2026-03-01",4251.3058],["2026-03-08",4251.3058],["2026-03-15",4251.3058],["2026-03-22",4095.2583],["2026-03-29",4108.0965],["2026-04-05",4108.0965],["2026-04-12",4799.8877],["2026-04-19",5318.0138],["2026-04-26",5392.8364],["2026-05-03",5097.178],["2026-05-10",5463.4604],["2026-05-17",4986.4393],["2026-05-24",4980.3533],["2026-05-31",4980.3533],["2026-06-07",4980.3533],["2026-06-14",4980.3533],["2026-06-21",4980.3533],["2026-06-28",4980.3533],["2026-07-05",4980.3533],["2026-07-12",4980.3533],["2026-07-19",4890.7251],["2026-07-26",4909.9586],["2026-08-02",4851.8234],["2026-08-09",4852.4076],["2026-08-16",4816.2571],["2026-08-23",5761.4552],["2026-08-30",5652.2983],["2026-09-06",5558.0927],["2026-09-13",5289.0374],["2026-09-20",5010.0615],["2026-09-27",4977.3563],["2026-10-04",4219.8288],["2026-10-11",4256.8418]],"code_bytes":7954},"a05":{"id":"a05","alias":"Merlin","theme":"Regime detection with hidden Markov models or other latent-state models","started":"2026-10-11T00:01:27+00:00","rounds":[{"k":1,"minutes":38.0},{"k":2,"minutes":44.8},{"k":3,"minutes":44.2},{"k":4,"minutes":16.1}],"messages":[{"t":"2026-10-11T00:01:34.564Z","text":"I’ll inspect the simulator and data, submit an early causal baseline, then test regime and selection rules on the visible years. I’ll track the variants and write down the maths as I go."},{"t":"2026-10-11T00:02:16.237Z","text":"The baseline is submitted and awaiting the referee. The daily data has 459 assets, including delisted coins. I’ll first compare momentum rules across that changing universe, then test whether regime signals help preserve gains through broad market declines."},{"t":"2026-10-11T00:04:14.722Z","text":"Plain momentum rotation is fragile in this sample: most variants lose money, and a few suffer severe drawdowns. Daily exits on individual coins improve the results substantially. I’m checking whether those gains survive nearby settings and whether a latent-state filter adds value."},{"t":"2026-10-11T00:07:49.007Z","text":"The filtered two-state strategy turns $100 into about $9,486 in the local simulator, with a 66% maximum drawdown. It is now submitted for the causality check. Faster coin-state transitions and a slower BTC-state filter look promising; I’m testing schedule sensitivity and dependence on a few large winners."},{"t":"2026-10-11T00:10:58.512Z","text":"The regime entry has passed the referee and currently leads the board. Its gains depend heavily on large winning days, and weekly refresh timing has a large effect. I’m testing daily decisions with a switch margin, so a held winner stays in place until another coin’s signal is materially stronger."},{"t":"2026-10-11T00:15:42.179Z","text":"The daily switch rule is accepted: $100 became $95,065, with a 77.8% maximum drawdown. Nearby switch margins and liquidity cutoffs remain profitable, though results vary widely. The writeup now documents the implemented maths and search history; I’m checking tail dependence and costs before making further changes."},{"t":"2026-10-11T00:17:03.499Z","text":"The current entry’s worst days include roughly 61% and 52% losses following sharp rallies. Doubling costs still leaves a large positive backtest. I’ll test a temporary exposure reduction after extreme gains and check the pattern across the liquid universe, to assess whether it’s more than a fix for those two trades."},{"t":"2026-10-11T00:20:06.649Z","text":"The exhaustion rule improves the earlier-history check and reduces drawdown across several nearby settings. I’m combining it with a volatility cap and checking the exact standalone file. The candidate keeps the same leader-selection rules, holds cash for three bars after a coin doubles in a day, and limits exposure when its estimated volatility is high."},{"t":"2026-10-11T00:22:25.397Z","text":"The guarded entry is accepted: $97,728 from $100, with drawdown reduced to 62.9%. It also grows all 24 tested rolling one-year windows; doubled costs leave about $67,440. Sensitivity checks still show large variation, so I’m keeping the central settings while testing whether transition probabilities can adapt causally."},{"t":"2026-10-11T00:25:43.894Z","text":"The adaptive transition model changed little, so I’m retaining fixed probabilities. A one-bar risk reduction after a large fade from the day’s high looks more promising: it improves wealth and drawdown in the local test. I’m checking nearby thresholds before deciding whether to add it."},{"t":"2026-10-11T00:27:46.243Z","text":"A smooth exposure reduction looks more stable than a hard cutoff. In the local test, it raises the account to about $128,011 and reduces drawdown to 49.2%. I’m completing the causal check before submission and updating the writeup to match the final code."},{"t":"2026-10-11T00:30:05.391Z","text":"Submission #005 is accepted. The final checks show $128,011 at stated costs and $88,695 at doubled costs, with positive results in every tested rolling year. The plots also make the tail dependence clear: large winners account for much of the growth. I’m finishing the writeup and source audit."},{"t":"2026-10-11T00:37:18.595Z","text":"The ablations support the core mechanism: removing the market gate reduces the visible account to about $2, and removing the switch margin leaves about $1,565. Removing the risk controls sometimes raises visible wealth, but worsens drawdown in the earlier window. I’ll retain the current balance and document that tradeoff."},{"t":"2026-10-11T00:39:26.390Z","text":"Entry: #005 accepted; strategy and writeup complete.  \nVisible years: $100 → **$128,010.99**; max drawdown **−49.2%**; currently **#1**.  \nNext idea: average several regime filters to reduce threshold sensitivity."},{"t":"2026-10-11T00:39:38.637Z","text":"I’ll focus this round on robustness across market regimes. First I’ll test whether averaging several causal regime filters reduces threshold sensitivity, then compare a broader market signal with the BTC gate. I’ll keep the accepted entry unless the evidence supports a change."},{"t":"2026-10-11T00:42:01.632Z","text":"The main weakness is the market gate: nearby probability thresholds produce very different returns. I’m testing whether averaging several fixed regime filters makes it more stable, using the earlier history and each visible year to judge the result."},{"t":"2026-10-11T00:43:26.520Z","text":"Averaging BTC filters with different state durations improved both the visible and earlier-period results in the first test. Averaging portfolio actions and replacing the hard gate with gradual exposure performed worse. I’m checking whether the probability average holds up when its parameters move."},{"t":"2026-10-11T00:44:43.161Z","text":"The wider filter average still has sharp threshold sensitivity. Broad market factors generally underperformed the BTC gate; a blend of the three permitted major coins improved visible wealth but weakened the earlier-period result. I’ll leave the entry unchanged while testing heavier-tailed emissions and independent strategy sleeves."},{"t":"2026-10-11T00:46:58.841Z","text":"The independent sleeves gave a useful tradeoff: three nearby market filters retained about $122,614 of visible wealth while reducing the earlier-period drawdown from 77% to 60%. I’ll compare that fixed average with a Bayesian filter that updates its belief about state duration from observed returns."},{"t":"2026-10-11T00:49:58.432Z","text":"The Bayesian duration model added complexity without a consistent gain. The fixed three-sleeve average looks more promising: it improved several weak parameter neighbours and reduced earlier-period drawdown, though the market threshold remains sensitive. I’m preparing it as a candidate and checking a latent leader model before deciding."},{"t":"2026-10-11T00:51:25.753Z","text":"The candidate passed the prefix causality checks and reproduces the research result: $122,614, with a 51.1% visible drawdown. The latent leader alternatives performed much worse. I’m finishing the cost and path comparisons before submitting the three-sleeve version."},{"t":"2026-10-11T00:53:25.144Z","text":"The earlier-period gain comes with similar average exposure: the sleeves grew $100 to $10,962 versus $6,344 for the current entry. Their resampled return differences still include losses, so the evidence is uncertain. The strongest reason for the change is the reduction in earlier drawdown from 77% to 60%."},{"t":"2026-10-11T00:54:00.104Z","text":"I’m submitting the fixed three-sleeve version. It gives up 4.2% of visible wealth, improves the earlier-period result, and depends less on the largest winning days. At twice and three times the stated costs, it still ends at about $85,055 and $58,998."},{"t":"2026-10-11T00:55:48.752Z","text":"Submission #006 is accepted, and Merlin remains #1. A separate test looks promising: investing part of residual cash in a trend-following sleeve of BTC, ETH and SOL improved both periods and reduced visible drawdown. I’ll test its sensitivity and costs before adding it."},{"t":"2026-10-11T00:57:23.564Z","text":"Both tested sizes of the major-coin sleeve helped. Before adding it, I’m checking a weakness in the initial design: it can add exposure to a coin that the burst strategy has just capped. I’ll apply the existing risk limits to the combined weights and vary the trend windows."},{"t":"2026-10-11T00:59:36.372Z","text":"The reserve’s central settings worked across neighbouring momentum windows, trend filters and allocation fractions. Applying the existing coin limits preserved the measured result: about $131,685, with a 46.4% visible drawdown. I’m running the full causal checker and cost stress tests on that candidate."},{"t":"2026-10-11T01:01:26.101Z","text":"The reserve candidate passed the causal checks. It ends at $131,685 with a 46.4% drawdown, compared with $122,614 and 51.1% for entry #006. It makes 659 trade legs and pays $21,005 in costs, so I’m checking the cost sensitivity before submitting it."},{"t":"2026-10-11T01:03:12.448Z","text":"The reserve improves wealth and drawdown at the arena’s stated costs, but at doubled and tripled costs its $82,608 and $51,819 results trail entry #006. I’m also checking account initialization: reaffirming a 100% single-coin target each day should let a fresh account enter promptly without ongoing rebalance trades."},{"t":"2026-10-11T01:04:50.237Z","text":"The initialization change leaves visible returns, fees and trade counts unchanged. I’m submitting the guarded reserve version with the central settings: 30-day momentum, a 20-day trend filter and half of residual cash. Neighbouring settings remained profitable; I’m keeping these central values."},{"t":"2026-10-11T01:07:10.455Z","text":"Submission #007 is accepted, and Merlin remains #1. I’m updating the writeup to cover the three regime sleeves, the major-coin reserve and the limits on combined coin weights. The remaining checks will identify which components drive the result and where it is still fragile."},{"t":"2026-10-11T01:11:35.656Z","text":"The ablations support keeping the current structure. Replacing the three market filters with one raises visible wealth slightly but worsens earlier drawdown from 58% to 72%. Removing the market gate nearly wipes out the account. I’ll test one further structural idea: filtering leadership from returns relative to BTC."},{"t":"2026-10-11T01:16:08.684Z","text":"Filtering coin returns relative to BTC improved several variants in both periods. The largest visible winner was weaker in the earlier history. I’m testing fixed mixtures with the original signal and auditing the largest winning bars for missing quotes and inconsistent prices."},{"t":"2026-10-11T01:19:35.728Z","text":"The equal mix of absolute and BTC-relative leadership passed the causal checks and ends at $147,428, versus $131,685 for entry #007. Its earlier-period result also improved, and the drawdowns are similar. The fifteen largest winning bars audited for #007 have complete quotes and consistent OHLC values."},{"t":"2026-10-11T01:21:34.164Z","text":"Submission #008 is accepted, and Merlin is #1 of 28 entrants. It grows $100 to $147,428 visibly and $57,847 in the earlier window, with drawdowns of 46.3% and 53.4%. I’m finishing the writeup and checking that the local code matches the accepted snapshot."},{"t":"2026-10-11T01:24:14.438Z","text":"Entry: #008 accepted — merlin_relative.py.  \nVisible years: **$100 → $147,427.62**, max drawdown **46.3%**; #1 at acceptance.  \nNext idea: reduce market-gate sensitivity through causal calibration. WRITEUP.md is current."},{"t":"2026-10-11T01:24:22.056Z","text":"I’ll test whether the market filter can adapt its emission scale causally and reduce threshold sensitivity. I’ll compare any change with the accepted entry across both development periods, keeping the focus on rules that could hold up in the hidden years."},{"t":"2026-10-11T01:27:07.151Z","text":"The accepted entry’s main weakness is its market gate: small threshold changes produce large differences in returns. I’m testing whether past BTC observations can calibrate the filter’s confidence, then checking both the earlier development period and each visible year separately."},{"t":"2026-10-11T01:28:33.633Z","text":"The first calibration results show a trade-off. Learning the state mean pushes it toward zero and leaves the strategy mostly in cash. Learning only the observation variance raises visible returns sharply, but so far lowers returns in the earlier period and increases drawdown. I’m checking a partial adjustment before considering it for the entry."},{"t":"2026-10-11T01:29:55.895Z","text":"Partial variance adjustment also trades higher visible returns for weaker earlier returns and larger drawdowns. I’m leaving the accepted entry in place and testing a different hypothesis: whether strong relative coin momentum can remain useful when BTC’s state is neutral."},{"t":"2026-10-11T01:31:11.051Z","text":"Letting relative coin confidence override BTC’s gate has mostly increased drawdowns without improving both development periods. I’m now testing equal policy averages of the fixed and variance-adapted filters. That keeps the existing rule represented while measuring the benefit of the adaptive version."},{"t":"2026-10-11T01:33:09.896Z","text":"The fixed/adaptive policy average reaches about $385,000 visibly, but lowers the earlier result from $57,847 to $30,609 and raises drawdown. I’ve recorded the trade-off in the writeup and kept entry 008. I’m next testing a third market state that can suspend trading through a volatile shock."},{"t":"2026-10-11T01:34:11.918Z","text":"The shock-state filter reduces visible drawdown in some settings, but also reduces growth in both periods. Another test is running now: ranking coins by expected contribution after applying the existing exposure limits. This would let a sleeve choose another coin when its current leader is capped at zero."},{"t":"2026-10-11T01:35:44.628Z","text":"Using the existing intrabar fade penalty in coin ranking is more promising. The equal mix of original and adjusted rankings improves visible wealth from $147,428 to $178,360 and earlier wealth from $57,847 to $81,415, with the same worst drawdown. I’m checking nearby settings and preparing a self-contained candidate."},{"t":"2026-10-11T01:37:08.986Z","text":"The ranking improvement survives nearby fade budgets: the equal policy mix gives about $170,000–$178,000 visibly and $81,000–$83,000 earlier. It still has substantial market-gate sensitivity. I’ve retained the existing 0.20 budget for the candidate; the referee’s truncation checks are running."},{"t":"2026-10-11T01:39:01.184Z","text":"Submission 009 is accepted and now leads the visible board: $100 became $178,360, with a 46.3% maximum drawdown. The earlier-window result also improved, to $81,415. I’m updating the writeup and diagnostics for this entry, then continuing work on the market gate’s sensitivity."},{"t":"2026-10-11T01:41:53.795Z","text":"Entry 009 still grows at doubled and tripled costs, ending near $110,373 and $68,294. Its gains remain highly concentrated: removing the best ten daily returns leaves about $2,704. I’m next testing whether prior realized performance can weight market-gate variants causally and reduce dependence on one threshold."},{"t":"2026-10-11T01:44:14.108Z","text":"Online expert weighting did not improve the balance: its best visible result was about $249,000, but earlier wealth fell to $48,601 and drawdown rose to 52.3%. I’m keeping entry 009. One remaining test couples the gate to emission strength, so changing that strength doesn’t also change the amount of normalized evidence needed to enter."},{"t":"2026-10-11T01:47:30.476Z","text":"Averaging normalized evidence across emission strengths improves both periods modestly: about $187,706 visibly and $88,759 earlier, with visible drawdown down to 45.6%. All five tested width choices stayed above entry 009’s visible result; none worsened earlier wealth. I’m checking this candidate with the referee before submitting it."},{"t":"2026-10-11T01:49:43.911Z","text":"Submission 010 is accepted and remains #1: $100 became $187,706, with a 45.6% maximum drawdown. It keeps the twelve trading policies and changes how the BTC evidence is combined. I’m updating the equations and diagnostics, then testing how much the result depends on the largest observed winners."},{"t":"2026-10-11T01:51:50.538Z","text":"Normalization reduces sensitivity to emission strength: shifting the whole strength grid by ±20% now leaves visible results around $144,000–$149,000. The largest-gain audit also found complete, internally consistent OHLCV bars, though that is not independent price verification. Winner-removal and decision-delay tests are running next."},{"t":"2026-10-11T01:53:35.931Z","text":"The tail tests are severe. Recomputing the strategy without its three largest observed contributors lowers visible wealth to about $6,032; delaying all decisions by one extra bar leaves about $4,155. I’ve also dropped to #2. I’m testing a different universe rule based on each coin’s share of trailing market volume, while keeping the current entry."},{"t":"2026-10-11T01:55:20.872Z","text":"The volume-share and concentration rules were weaker than the fixed top-50 universe in these tests. I’m keeping entry 010. One smaller structural test remains: admitting coins in the top 50 while allowing a healthy incumbent to retain its score after slipping below that liquidity rank."},{"t":"2026-10-11T01:58:14.912Z","text":"The incumbent-liquidity extension leaves the current entry’s results unchanged, so it adds no demonstrated benefit. I’m checking the supplied Fear & Greed data as separate market context and as a sizing guard during extreme greed. These are research variants; entry 010 remains active."},{"t":"2026-10-11T01:59:48.017Z","text":"Using Fear & Greed for direction weakened the strategy. Using it to reduce speculative sizing during extreme greed was more useful: the central setting preserves visible wealth near $188,557, raises earlier wealth to $125,809, and lowers drawdown in both periods. Nearby thresholds also improve the earlier period. I’m checking it with the existing sizing neighborhood."},{"t":"2026-10-11T02:01:17.401Z","text":"With the greed guard, the 0.15 sizing scale gives about $233,688 visibly and $130,964 earlier. Earlier drawdown rises from 49.5% at 0.125 to 52.4%; visible drawdown stays near 42.5%. I’m validating the higher-growth version against the referee’s costs and causality checks before submitting it."},{"t":"2026-10-11T02:03:56.745Z","text":"Submission 011 is accepted at #2: $100 became $233,688, with a 42.5% maximum drawdown. Earlier wealth is $130,964, and doubled/tripled costs leave about $142,586/$86,991 visibly. I’m finishing the writeup for this version, including the new guard’s equations and the tail-dependence findings."},{"t":"2026-10-11T02:07:16.744Z","text":"Requiring positive absolute coin evidence leaves the entry unchanged. A stronger 0.55 cutoff raises visible wealth to $267,761 but lowers earlier wealth, so I’m leaving it out. Entry 011 and its writeup are current; I’m finishing fresh two-year checks to match the final ranking’s horizon."},{"t":"2026-10-11T02:08:24.457Z","text":"Entry: accepted #011, merlin_greed_growth.py, ranked #2.  \nVisible: $100 → **$233,687.62**; maximum drawdown **42.5%**.  \nNext idea: smooth posterior-confidence sizing. Writeup updated."},{"t":"2026-10-11T02:08:32.104Z","text":"I’ll test smooth confidence sizing against the accepted entry, focusing on whether it improves both development periods. I’ll leave enough time to submit the final choice and verify that the writeup matches it."},{"t":"2026-10-11T02:11:12.549Z","text":"The accepted entry is still the $233,688 strategy. Stronger coin-confidence gates improved visible wealth but weakened the earlier development window, so I’m testing gradual sizing before deciding whether to change it."},{"t":"2026-10-11T02:12:11.846Z","text":"Gradual confidence sizing changed little at low cutoffs; stronger sizing generally reduced visible growth. The best setting improved both windows by only a few percent, which is weak evidence after so many trials. I’ll check one broader change: averaging nearby market-entry thresholds to reduce sensitivity to the current 0.65 cutoff."},{"t":"2026-10-11T02:13:43.460Z","text":"Averaging market thresholds did not help. The broader symmetric mix ended near $158,000 visibly and $54,000 earlier; the more aggressive mix reached $409,000 visibly but fell to $63,000 earlier. I’m retaining the accepted market gate and checking the small confidence-sizing improvement across nearby settings."},{"t":"2026-10-11T02:15:42.633Z","text":"Confidence sizing improved visible wealth in all ten matched checks and earlier wealth in nine. Both nearby confidence scales also improved the central rule. I’ll submit the modest version: reduce the speculative allocation when the average of the two coin-state scores is near neutral, while keeping the market gate and existing risk limits."},{"t":"2026-10-11T02:16:57.019Z","text":"The candidate passed the full causality check and is with the referee as submission 012. The local result is $236,718 from $100, with the same 42.5% maximum drawdown. I’m refreshing its cost checks, fresh-account results, figures and writeup while it scores."},{"t":"2026-10-11T02:18:00.125Z","text":"Submission 012 is accepted and is now the final entry. It scored $236,718.28 after costs, with 42.5% maximum drawdown; the earlier development account ended at $134,937.29. The code remains under 6 KB and takes about one CPU second per call. I’m reconciling the paper and audit files with that exact submission."},{"t":"2026-10-11T02:22:01.227Z","text":"The writeup now matches submission 012, including every signal and sizing equation, 1,544 scored configuration evaluations, and the failed approaches. At triple costs, the visible account ends at $88,051. The final source matches the accepted copy; the paper also records the rule’s dependence on rare bursts and its remaining market-gate sensitivity."},{"t":"2026-10-11T02:24:32.942Z","text":"Final accepted entry **#012** is merlin.py: twelve latent-state leadership policies with confidence sizing, tail controls and a BTC/ETH/SOL reserve. On the visible years, it turned **$100 into $236,718.28 after costs**, with **42.5% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with equations, 1,544 scored configuration evaluations, robustness results and failed approaches."}],"submissions":[{"n":"001","t":"2026-10-11T00:01:53.625402+00:00","note":"Merlin initial causal BTC trend baseline with weekly entry refresh","status":"ok","dev":195.63220430662435,"dd":-0.3612426554665472,"reason":""},{"n":"002","t":"2026-10-11T00:07:28.227492+00:00","note":"Filtered two-state regime; faster coin regime, slower BTC gate; data-ranked liquid leaders with daily trend exits","status":"ok","dev":9486.23825477442,"dd":-0.6606544583484173,"reason":""},{"n":"003","t":"2026-10-11T00:12:46.664802+00:00","note":"Daily regime leadership, retain incumbent until challenger expected drift doubles; EWM risk scale; no calendar rotation","status":"ok","dev":95065.191677513,"dd":-0.7779045931178477,"reason":""},{"n":"004","t":"2026-10-11T00:20:33.997023+00:00","note":"Causal HMM leadership with switch margin, 12.5% daily volatility budget and three-bar cash cooldown after a doubling day","status":"ok","dev":97728.08912314363,"dd":-0.6292051962020944,"reason":""},{"n":"005","t":"2026-10-11T00:27:48.283614+00:00","note":"HMM leadership with smooth daily exposure limits for volatility and a fade from the high, plus generic doubling-day cooldown","status":"ok","dev":128010.99398593127,"dd":-0.4919192396422478,"reason":""},{"n":"006","t":"2026-10-11T00:54:02.275781+00:00","note":"Three fixed regime-duration sleeves; retain burst leadership and smooth tail sizing; improved earlier-period drawdown and reduced dependence on best days.","status":"ok","dev":122614.1693703855,"dd":-0.5111203562392411,"reason":""},{"n":"007","t":"2026-10-11T01:04:53.293524+00:00","note":"Equal regime-duration policies plus a guarded major-coin reserve using half of residual cash; central parameters; causality passed; stronger earlier-period growth with lower visible drawdown.","status":"ok","dev":131685.0142714035,"dd":-0.4640508911406591,"reason":""},{"n":"008","t":"2026-10-11T01:19:43.574630+00:00","note":"Equal absolute and causal BTC-relative leadership policies crossed with three regime-duration filters; guarded liquid reserve; fixed model priors and stronger earlier-history results.","status":"ok","dev":147427.6230599583,"dd":-0.4631067154446702,"reason":""},{"n":"009","t":"2026-10-11T01:37:13.543548+00:00","note":"Equal policy uncertainty over absolute/relative leadership and original/intrabar-risk-adjusted rankings; same guards and reserve. Improves both development windows; no learned constants or dates.","status":"ok","dev":178360.1322076304,"dd":-0.4631067154446703,"reason":""},{"n":"010","t":"2026-10-11T01:48:20.594536+00:00","note":"Average BTC normalized log-odds evidence across nearby emission strengths, then map at the original reference scale; retains twelve leadership/stress/duration sleeves and all guards. Broad positive wi","status":"ok","dev":187706.40020340178,"dd":-0.45633888039722936,"reason":""},{"n":"011","t":"2026-10-11T02:01:32.662140+00:00","note":"Extreme-greed speculative allocation migrates toward guarded major-coin reserve: smooth burst fraction beyond index 80, with nearby threshold/floor checks. Volatility scale 0.15 improves both developm","status":"ok","dev":233687.6236998251,"dd":-0.4246800734966063,"reason":""},{"n":"012","t":"2026-10-11T02:16:16.038441+00:00","note":"Model-average coin conviction scales speculative exposure smoothly near neutrality; fixed latent-state market gate, twelve leader policies, tail controls, and major reserve retained. Matched developme","status":"ok","dev":236718.27715500284,"dd":-0.4246800734966061,"reason":""}],"entry":"012","round_notes":[{"k":1,"text":"Entry: #005 accepted; strategy and writeup complete.  \nVisible years: $100 → **$128,010.99**; max drawdown **−49.2%**; currently **#1**.  \nNext idea: average several regime filters to reduce threshold sensitivity."},{"k":2,"text":"Entry: #008 accepted — merlin_relative.py.  \nVisible years: **$100 → $147,427.62**, max drawdown **46.3%**; #1 at acceptance.  \nNext idea: reduce market-gate sensitivity through causal calibration. WRITEUP.md is current."},{"k":3,"text":"Entry: accepted #011, merlin_greed_growth.py, ranked #2.  \nVisible: $100 → **$233,687.62**; maximum drawdown **42.5%**.  \nNext idea: smooth posterior-confidence sizing. Writeup updated."},{"k":4,"text":"Final accepted entry **#012** is merlin.py: twelve latent-state leadership policies with confidence sizing, tail controls and a BTC/ETH/SOL reserve. On the visible years, it turned **$100 into $236,718.28 after costs**, with **42.5% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with equations, 1,544 scored configuration evaluations, robustness results and failed approaches."}],"title":"Merlin (a05): writeup","idea":"The final accepted entry is submission 012, merlin_confidence.py (also copied to merlin.py). It trades daily through twelve equal policy sleeves, each holding one liquid coin or cash. A fixed two-state hidden Markov filter estimates a rising-return state for each coin using either absolute returns or returns relative to a causal BTC factor estimate. Three BTC state durations represent uncertainty about market persistence; each combines normalized evidence from three nearby emission strengths. Coin states change faster than these market states. A coin's filtered directional probability, multiplied by a slowly estimated return scale, measures its speculative leadership. Each coin model uses two ranking views: the original leadership score and that score reduced by the same observed intrabar-fade fraction used in sizing. Averaging both views allows uncertainty about whether a price rejection should change the chosen leader. The strategy enters only when the BTC state supports risk, exits a coin whose price loses its short trend, and retains an incumbent until a challenger has twice its score. The twelve sleeves maintain independent incumbents and their speculative targets are averaged. This switch margin avoids an arbitrary weekly selection schedule and lets a winning burst continue. A separate exposure rule uses a 15% individual daily-volatility sizing scale, holds cash for three","failed":"- Plain weekly cross-sectional momentum generally lost money through coin reversals and market declines. Its apparent best case still had a 92% drawdown. - Slower BTC 100/200-day gates often reacted too late for speculative coins. - Ranking solely by return divided by volatility, and wider equal-weight baskets, suppressed the burst exposure required for the growth objective. - A broad average of coin-state probabilities was less effective than the BTC directional gate in these tests. - Multiplying by volume attention, residual returns or price breakout proximity sometimes helped, but did not consistently improve the simpler signal. - Fast/slow coin-state blending diluted the useful fast leadership changes. - Staggering all weekly selection phases reduced schedule dependence but also reduced growth and did not solve drawdown; daily switch hysteresis worked better. - Walk-forward pooled gradient-boosted mean-log-return regression, with purged 3/7/14-day labels and no asset identities, failed: best visible account about $197 among 24 variants, most below $100. LightGBM could not load because the offline environment lacked its OpenMP library; these experiments used scikit-learn's histogram gradient boosting instead. - Walk-forward classifiers for a 25%/50% future price burst were more useful than mean-return regression, but adding them did not beat the simpler guarded rule consistently. Best raw daily-switch classifier combination was about $69,212; most were considerably weaker. - Keeping a winner through a weak market state helped some configurations but generally increased uncertainty; the chosen rule keeps the hard market gate. - Bayesian soft-count transition updates moved hazards only slightly and changed a few selected trades. Some variants improved visible wealth, b","holdup":"The mechanism is general: speculative attention can produce persistent, positively skewed price bursts; trading costs reward retaining a leader; abrupt trend failure and weak broad-market conditions make cash useful. The liquid universe updates using trailing volume, so it can admit new leaders without knowing their identities, and it includes historical delistings. Fixed-state filtering is inexpensive and makes no fragile state-label assignment from a whole-sample fit. Every operation is causal and remains defined on a truncated history. Neither names of past winners nor event dates appear in the code. The conviction fraction uses current uncertainty rather than a calendar regime or asset identity; it reduces exposure near neutral directional scores. An incumbent can stay above its price average while directional evidence decays; this fraction reduces the capital assigned to that weakening signal without forcing a new leader selection. The reserve can use part of the released cash when its slower major-coin trend remains positive. Its scale is still a development choice. These arguments support testing the rule forward, not extrapolating a 2367× backtest. The growth objective need","result":{"dev":{"final":236718.27715500284,"max_drawdown":-0.4246800734966061,"trades":735,"exposure":0.3244391512354954,"assets_traded":85,"fees_paid":34699.788887444985},"hidden":{"final":6.000232396861915,"max_drawdown":-0.9444323640185727,"trades":653,"exposure":0.34756955410938895,"assets_traded":86,"fees_paid":18.34392801371824},"full":{"final":14203.646755147878,"max_drawdown":-0.956181986355992,"trades":1388,"exposure":0.33368624548409465,"assets_traded":147,"fees_paid":78123.21924407272}},"freq":"1d","equity":[["2021-10-17",91.1296],["2021-10-24",85.1429],["2021-10-31",182.4741],["2021-11-07",159.0883],["2021-11-14",254.1885],["2021-11-21",221.0351],["2021-11-28",221.0351],["2021-12-05",221.0351],["2021-12-12",221.0351],["2021-12-19",221.0351],["2021-12-26",221.0351],["2022-01-02",209.3412],["2022-01-09",209.3412],["2022-01-16",209.3412],["2022-01-23",209.3412],["2022-01-30",209.3412],["2022-02-06",209.3412],["2022-02-13",327.9866],["2022-02-20",291.7513],["2022-02-27",291.7513],["2022-03-06",314.0121],["2022-03-13",314.0121],["2022-03-20",314.0121],["2022-03-27",309.9218],["2022-04-03",526.8482],["2022-04-10",442.1714],["2022-04-17",442.1714],["2022-04-24",442.1714],["2022-05-01",442.1714],["2022-05-08",442.1714],["2022-05-15",442.1714],["2022-05-22",442.1714],["2022-05-29",442.1714],["2022-06-05",442.1714],["2022-06-12",442.1714],["2022-06-19",442.1714],["2022-06-26",442.1714],["2022-07-03",442.1714],["2022-07-10",442.1714],["2022-07-17",442.1714],["2022-07-24",814.9509],["2022-07-31",800.5068],["2022-08-07",765.5481],["2022-08-14",803.9059],["2022-08-21",746.6237],["2022-08-28",746.6237],["2022-09-04",746.6237],["2022-09-11",746.6237],["2022-09-18",687.8701],["2022-09-25",687.8701],["2022-10-02",687.8701],["2022-10-09",671.6308],["2022-10-16",671.6308],["2022-10-23",671.6308],["2022-10-30",822.948],["2022-11-06",700.7943],["2022-11-13",624.2053],["2022-11-20",624.2053],["2022-11-27",624.2053],["2022-12-04",624.2053],["2022-12-11",624.2053],["2022-12-18",613.1248],["2022-12-25",613.1248],["2023-01-01",613.1248],["2023-01-08",611.5967],["2023-01-15",670.4102],["2023-01-22",626.6246],["2023-01-29",949.7681],["2023-02-05",1037.4851],["2023-02-12",953.6813],["2023-02-19",1598.0029],["2023-02-26",2764.6824],["2023-03-05",2650.3684],["2023-03-12",2650.3684],["2023-03-19",3618.8252],["2023-03-26",3160.1821],["2023-04-02",3991.7244],["2023-04-09",4856.9407],["2023-04-16",5257.5128],["2023-04-23",7239.4961],["2023-04-30",7185.7373],["2023-05-07",6861.8695],["2023-05-14",6802.0273],["2023-05-21",6802.0273],["2023-05-28",6802.0273],["2023-06-04",6712.023],["2023-06-11",6712.023],["2023-06-18",6712.023],["2023-06-25",7946.0757],["2023-07-02",10882.4252],["2023-07-09",10829.1386],["2023-07-16",10694.5008],["2023-07-23",9921.9183],["2023-07-30",9627.1946],["2023-08-06",9627.1946],["2023-08-13",9698.9354],["2023-08-20",9698.9354],["2023-08-27",9698.9354],["2023-09-03",9698.9354],["2023-09-10",9698.9354],["2023-09-17",9698.9354],["2023-09-24",9696.6486],["2023-10-01",9696.6486],["2023-10-08",10404.1955],["2023-10-15",10214.7701],["2023-10-22",10049.5526],["2023-10-29",10331.2732],["2023-11-05",26253.4848],["2023-11-12",33176.1492],["2023-11-19",32100.2692],["2023-11-26",28401.1094],["2023-12-03",32603.8544],["2023-12-10",43292.2931],["2023-12-17",40064.2307],["2023-12-24",52356.6967],["2023-12-31",68276.4722],["2024-01-07",75541.0429],["2024-01-14",94832.3272],["2024-01-21",94832.3272],["2024-01-28",94832.3272],["2024-02-04",92681.9609],["2024-02-11",110212.1505],["2024-02-18",91807.4692],["2024-02-25",126117.4248],["2024-03-03",176047.782],["2024-03-10",315625.8722],["2024-03-17",217680.6592],["2024-03-24",186490.4483],["2024-03-31",194006.8773],["2024-04-07",189941.1653],["2024-04-14",191477.5469],["2024-04-21",191477.5469],["2024-04-28",190706.2929],["2024-05-05",190706.2929],["2024-05-12",190706.2929],["2024-05-19",193241.1673],["2024-05-26",247943.8354],["2024-06-02",256392.3963],["2024-06-09",256160.5758],["2024-06-16",251216.5176],["2024-06-23",251216.5176],["2024-06-30",251216.5176],["2024-07-07",251216.5176],["2024-07-14",251216.5176],["2024-07-21",288005.0472],["2024-07-28",271860.7987],["2024-08-04",251078.3302],["2024-08-11",251078.3302],["2024-08-18",251078.3302],["2024-08-25",251078.3302],["2024-09-01",251078.3302],["2024-09-08",251078.3302],["2024-09-15",249905.5367],["2024-09-22",257064.1447],["2024-09-29",246317.021],["2024-10-06",238083.7619],["2024-10-13",237860.2108],["2024-10-20",211434.183],["2024-10-27",189967.8027],["2024-11-03",189936.517],["2024-11-10",222059.8902],["2024-11-17",226325.6489],["2024-11-24",186675.2764],["2024-12-01",182177.1855],["2024-12-08",186481.3934],["2024-12-15",132555.0577],["2024-12-22",118509.484],["2024-12-29",116668.7843],["2025-01-05",117266.4894],["2025-01-12",102932.2428],["2025-01-19",95595.5431],["2025-01-26",92030.939],["2025-02-02",82752.0987],["2025-02-09",82713.0393],["2025-02-16",82713.0393],["2025-02-23",82713.0393],["2025-03-02",82713.0393],["2025-03-09",82713.0393],["2025-03-16",82713.0393],["2025-03-23",82713.0393],["2025-03-30",82713.0393],["2025-04-06",82713.0393],["2025-04-13",82713.0393],["2025-04-20",81036.6734],["2025-04-27",84696.5354],["2025-05-04",88064.6076],["2025-05-11",115242.4524],["2025-05-18",80027.2758],["2025-05-25",63190.6944],["2025-06-01",60179.5066],["2025-06-08",58792.8463],["2025-06-15",51555.3136],["2025-06-22",50963.1021],["2025-06-29",50963.1021],["2025-07-06",50352.6565],["2025-07-13",68722.2884],["2025-07-20",63739.6967],["2025-07-27",61931.1916],["2025-08-03",59478.0251],["2025-08-10",62111.2589],["2025-08-17",57532.3082],["2025-08-24",56451.1471],["2025-08-31",56423.0791],["2025-09-07",56423.0791],["2025-09-14",58700.412],["2025-09-21",58348.0996],["2025-09-28",55995.4914],["2025-10-05",66036.6808],["2025-10-12",59969.5401],["2025-10-19",59969.5401],["2025-10-26",59969.5401],["2025-11-02",58723.5768],["2025-11-09",58723.5768],["2025-11-16",58723.5768],["2025-11-23",58723.5768],["2025-11-30",58723.5768],["2025-12-07",58723.5768],["2025-12-14",58723.5768],["2025-12-21",58723.5768],["2025-12-28",58723.5768],["2026-01-04",59848.3516],["2026-01-11",49216.38],["2026-01-18",63211.9264],["2026-01-25",57238.3835],["2026-02-01",57238.3835],["2026-02-08",57238.3835],["2026-02-15",57238.3835],["2026-02-22",57238.3835],["2026-03-01",57238.3835],["2026-03-08",57238.3835],["2026-03-15",57690.7157],["2026-03-22",57531.7742],["2026-03-29",55985.2802],["2026-04-05",55985.2802],["2026-04-12",69165.4591],["2026-04-19",151321.3269],["2026-04-26",103241.6549],["2026-05-03",94389.1453],["2026-05-10",91319.3502],["2026-05-17",77324.2605],["2026-05-24",77324.2605],["2026-05-31",77324.2605],["2026-06-07",77324.2605],["2026-06-14",77324.2605],["2026-06-21",77324.2605],["2026-06-28",77324.2605],["2026-07-05",77324.2605],["2026-07-12",74727.6396],["2026-07-19",73403.9787],["2026-07-26",73287.859],["2026-08-02",72454.281],["2026-08-09",72809.8001],["2026-08-16",71946.431],["2026-08-23",79806.4943],["2026-08-30",66259.1808],["2026-09-06",61018.0402],["2026-09-13",53450.1191],["2026-09-20",51576.7758],["2026-09-27",16030.0507],["2026-10-04",14729.4096],["2026-10-11",14203.6468]],"code_bytes":5408},"a06":{"id":"a06","alias":"Harrier","theme":"Bayesian online changepoint detection to switch between market states","started":"2026-10-11T00:01:47+00:00","rounds":[{"k":1,"minutes":80.0},{"k":2,"minutes":43.0},{"k":3,"minutes":21.4}],"messages":[{"t":"2026-10-11T00:01:53.187Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test changepoint-driven allocation rules. I’ll keep a record of variants and validation for the writeup."},{"t":"2026-10-11T00:02:30.958Z","text":"The BTC trend baseline is with the referee. I’ll first test whether recent strength across liquid coins offers enough return after costs to justify switching between them, then use the changepoint model to decide when and how much to invest."},{"t":"2026-10-11T00:04:18.879Z","text":"The first scan found that chasing the strongest coins often bought exhausted rallies and lost money after costs. Trend rules on BTC, ETH and SOL did better. I’m now testing short-term pullbacks and a Bayesian estimate of positive drift, with separate checks for the bear market and recovery."},{"t":"2026-10-11T00:06:43.904Z","text":"The changepoint filter found a useful candidate: roughly ×8–9 on the visible years, with about 40–50% drawdown. Nearby hazard settings also worked, which is encouraging. I’ll check an ensemble of those settings for causality and costs before submitting it."},{"t":"2026-10-11T00:09:53.549Z","text":"The Bayesian ensemble is accepted: $100 became $855.01, with a 41.7% maximum drawdown. I’m testing two extensions: faster entry and exit signals, and a pooled model that learns from each coin’s price and volume history, without using coin names."},{"t":"2026-10-11T00:12:54.330Z","text":"Faster trading and the pooled prediction model haven’t improved the entry. A more promising result came from volume-confirmed breakouts in recently listed coins. I’ll check that result across older periods and nearby settings to see how dependent it is on a few trades."},{"t":"2026-10-11T00:17:09.775Z","text":"The recent-listing result weakens substantially when I broaden the settings. I’m keeping the accepted entry while testing a different use of changepoints: detecting when a trading rule’s own returns improve, then selecting among those rules. That may help distinguish persistent trends from one-off price spikes."},{"t":"2026-10-11T00:22:36.040Z","text":"Detecting changes in rule profitability produced a stronger candidate: $100 became $2,255.12, with 55.7% drawdown. It also trades heavily, so I’m testing whether a cost-based threshold for switching coins can preserve more of that return."},{"t":"2026-10-11T00:31:32.290Z","text":"The $2,255.12 entry passed the referee’s causality check, and its equations and results are now in the writeup. The stress checks show a limitation: doubled costs reduce it to $887.28, and its ten best days account for all of the net gain. I’m checking the model’s treatment of cash days next."},{"t":"2026-10-11T00:41:03.567Z","text":"Treating cash days as missing observations hasn’t improved the results, and the hourly version loses too much to turnover. The daily entry is still strongest. I’m now testing broader risk signals—sentiment, market breadth and the strategy’s recent returns—to see whether they can reduce its unprofitable churn."},{"t":"2026-10-11T00:47:41.293Z","text":"Sentiment and market breadth both improved the results. Averaging their risk filters produced $4,739.41 with 35.9% drawdown. Nearby settings returned roughly $3,630–$4,878, and doubled costs still left $2,223.06. I’m submitting this version and updating the writeup."},{"t":"2026-10-11T00:53:09.897Z","text":"The new version is accepted: $100 became $4,739.41. It also improves the earlier-period test and remains profitable when I remove its ten best days. I’ll use the rest of this round to check sensitivity to the Bayesian prior and the liquidity cutoff."},{"t":"2026-10-11T01:00:23.537Z","text":"Stronger drift priors didn’t improve the visible result. The liquidity cutoff matters more: using eight coins instead of ten weakened returns substantially. I’m testing a smoother rule that reduces confidence as liquidity falls, to see whether it can remove that sharp boundary."},{"t":"2026-10-11T01:03:31.771Z","text":"The smoother liquidity rule performed much worse, so I’m retaining the accepted entry. Its macro filters look more stable than its coin cutoff. I’m recording that distinction in the writeup and checking that the local file matches submission 004."},{"t":"2026-10-11T01:13:46.284Z","text":"The accepted file and writeup now match. The tests distinguish the stronger macro filters from the sensitive coin cutoff, and the smoothing attempts haven’t resolved that sensitivity. I’m retaining submission 004 unless the last comparison gives a clear improvement."},{"t":"2026-10-11T01:23:29.693Z","text":"The accepted entry remains $4,739.41 from $100. I’ll use this round to test a different source of returns—volume-supported breakouts with persistent holdings—and compare it across the earlier data and separate visible periods before changing the entry."},{"t":"2026-10-11T01:24:13.716Z","text":"The earlier breakout tests depended heavily on a few new-coin surges. I’m testing whether rising volume can identify trends worth holding, with exits based on price deterioration. This should show whether the idea adds anything beyond ordinary momentum."},{"t":"2026-10-11T01:26:45.876Z","text":"The first volume-flow results are weaker than the accepted entry, and ranking only established coins did not resolve its liquidity-cutoff sensitivity. I’m also testing a model that learns the probability of large gains and losses from earlier coin histories; it may capture opportunities that a forecast of average returns misses."},{"t":"2026-10-11T01:29:19.567Z","text":"The new flow and long-range breakout rules are still well below the accepted strategy. I’m keeping them as failed experiments and checking the learned large-move forecasts next, with every training label kept strictly before its prediction date."},{"t":"2026-10-11T01:33:29.194Z","text":"The large-move classifier is still being evaluated. I’m preparing one further alternative: a changepoint model that estimates win frequency and win/loss sizes separately. That could handle rare upside jumps better than the current normal model of trading returns."},{"t":"2026-10-11T01:35:49.528Z","text":"Earlier eligibility is the first substantial improvement: allowing the model to evaluate newly liquid coins sooner returned $11,530 in the local test, versus $4,739 for the entry. It also gained in each visible subperiod and returned $2,186 in the earlier data. I’ll check nearby eligibility rules and costs before submitting it."},{"t":"2026-10-11T01:39:58.594Z","text":"The early-eligibility result holds up across nearby settings and doubled costs. I’ll submit an ensemble of 5- and 10-observation volatility warmups and volume cutoffs of 9, 10 and 11. It returns $11,408 locally with 37.5% drawdown, and performs better on the earlier data than the strongest single setting."},{"t":"2026-10-11T01:44:25.054Z","text":"Submission 005 is with the referee for the causality check. Its local stress test still returns $5,459 from $100 at twice the execution costs. I’m updating the writeup with the new equations, the search history, and the failed flow, classifier and asymmetric-return experiments."},{"t":"2026-10-11T01:47:14.606Z","text":"Submission 005 was accepted: $11,408.15 from $100, 37.5% drawdown, and rank 6 of 40 at acceptance. The writeup is current. I’ll now test whether market participation can identify when the broader asymmetric model is useful; it performed much better in the earlier bull data than in the visible window."},{"t":"2026-10-11T01:52:31.374Z","text":"Broad participation looks useful: adding the asymmetric model only when most liquid coins are trending raised the local result to $17,263 and the earlier-period result to $8,766. Nearby settings also improved the earlier data. I’ll validate the blend that averages participation thresholds, with at most half the portfolio assigned to the asymmetric model."},{"t":"2026-10-11T01:59:16.271Z","text":"Submission 006 was accepted: $17,262.55 from $100, 39.4% drawdown, and rank 7 at acceptance. The writeup now includes both models and the state-allocation equations. Nearby settings for the new model returned roughly $15,588–$19,738; I’ve retained the central parameters."},{"t":"2026-10-11T02:03:34.146Z","text":"An execution-delay check exposed an important limitation: waiting one extra daily bar reduced the result to $3,211. The strategy relies on the specified next-open fills. I’ll document that sensitivity alongside the cost and parameter checks."},{"t":"2026-10-11T02:04:48.177Z","text":"Entry: #006 accepted; strategy.py and writeup current.  \nVisible years: $100 → $17,262.55 (×172.63), maximum drawdown 39.4%.  \nNext idea: detect renewed liquidity in mature coins before price breakouts."},{"t":"2026-10-11T02:04:55.080Z","text":"Submission 006 is accepted, and its writeup is complete. I’ll use the remaining time to test whether renewed liquidity in mature coins adds a distinct edge, then leave the strongest supported entry in place before sealing."},{"t":"2026-10-11T02:06:32.988Z","text":"The latest accepted entry is submission 006, which returned $17,262.55 on the visible years. I’ll test renewed liquidity in established coins as a different source of opportunity, then verify the chosen entry and finish the writeup before sealing."},{"t":"2026-10-11T02:07:03.845Z","text":"The accepted strategy’s gains depend on prompt execution: adding one extra day of delay reduced its visible result to about $3,211. That makes timing a material limitation. The new test looks for established coins returning to volume leadership after a quiet period, with persistent holdings that may be less sensitive to daily selection changes."},{"t":"2026-10-11T02:08:55.491Z","text":"The liquidity-renewal variants are weaker than the accepted entry, including when blended into it. I’m keeping 006 in place. One final test will give win and loss sizes their own Bayesian uncertainty, so the model can distinguish frequent small wins from rarer large payoffs more consistently."},{"t":"2026-10-11T02:10:44.917Z","text":"The fuller payoff model came close at $17,039, but its earlier-period result fell to $5,404 versus the entry’s $8,766. I won’t replace 006 on that evidence. I’ll also check whether averaging the EMA models’ forecasts improves stability over choosing their single highest score."},{"t":"2026-10-11T02:13:11.405Z","text":"Averaging the EMA forecasts improved earlier wealth but weakened visible wealth and drawdown. None of these alternatives is a clear replacement. I found one remaining modelling choice worth checking: how much an unobserved new regime contributes to the uncertainty penalty used for ranking coins."},{"t":"2026-10-11T02:14:50.837Z","text":"Reducing the new-regime uncertainty penalty gave a small visible gain, but increased drawdown and weakened earlier wealth. I’m retaining the original penalty. Since early participation produced the largest improvement so far, the last candidate will test a shorter trend horizon and formation period, followed by cost and timing checks if it looks promising."},{"t":"2026-10-11T02:16:37.981Z","text":"The entry file still exactly matches accepted submission 006. I’m adding the final experiments to the writeup, including the rejected variants and their tradeoffs. The shorter-horizon test is still running; I’ll require more than a small gain in visible dollars before changing the entry."},{"t":"2026-10-11T02:18:12.277Z","text":"A three-return formation period reached $18,651 with the same visible drawdown, but weakened earlier wealth to $8,024. The three-period ensemble gained only 2.8% visibly and also weakened earlier wealth. I’m checking the leading variant’s costs and delay sensitivity, plus whether a simple market-shock filter reduces the accepted entry’s risk."},{"t":"2026-10-11T02:19:49.905Z","text":"I’m keeping submission 006 as the final entry. The faster formation variant was slightly better at doubled fees but worse with an extra day of delay, and the shock filters did not improve the return/drawdown tradeoff enough to justify adding them. I’ll finish the documentation with a descriptive check of fresh two-year accounts across the available history."},{"t":"2026-10-11T02:21:03.255Z","text":"Across 37 overlapping two-year windows, fresh $100 accounts ended between $497 and $33,808, with a median of $4,580. These are descriptive results from data used during research, not independent validation. Their wide range is worth reporting alongside the headline result; I’m finalizing the writeup and acceptance record now."},{"t":"2026-10-11T02:26:13.337Z","text":"Harrier’s final entry remains accepted submission **006**, strategy.py: a daily Bayesian changepoint ensemble that scores cost-aware trend rules, admits newly liquid coins early, and caps exposure during broad rallies. Visible testing turned **$100 into $17,262.55**, after costs, with **39.42% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, trial counts, results and failed ideas. It also documents substantial selection bias and execution-delay sensitivity."}],"submissions":[{"n":"001","t":"2026-10-11T00:02:16.252944+00:00","note":"Initial causal BTC 200-day trend baseline; research entry to follow.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:07:46.878814+00:00","note":"Bayesian online changepoint filter, 20/30-day hazard ensemble, positive-drift hysteresis and BTC risk gate.","status":"ok","dev":855.0070241385579,"dd":-0.41658151033260604,"reason":""},{"n":"003","t":"2026-10-11T00:22:37.334117+00:00","note":"Bayesian change detection of four cost-aware trend experts; causal top-volume universe and probability-cutoff ensemble.","status":"ok","dev":2255.122113212569,"dd":-0.5565007604429619,"reason":""},{"n":"004","t":"2026-10-11T00:47:42.806303+00:00","note":"Cost-aware Bayesian trend-rule selection with opportunity breadth and recovering sentiment; averaged risk gates and cutoff ensemble.","status":"ok","dev":4739.406588980694,"dd":-0.3590363931074202,"reason":""},{"n":"005","t":"2026-10-11T01:41:31.333155+00:00","note":"Bayesian profitability models for emerging liquid coins; average early volatility formation and nearby top-volume boundaries, with breadth and recovering-sentiment risk budgets.","status":"ok","dev":11408.150740669636,"dd":-0.37464313036116204,"reason":""},{"n":"006","t":"2026-10-11T01:53:42.526217+00:00","note":"Two causal market states: an emerging-liquidity Gaussian changepoint ensemble, plus bounded sign-and-magnitude exposure during broad price participation.","status":"ok","dev":17262.549663896196,"dd":-0.39424011276915616,"reason":""}],"entry":"006","round_notes":[{"k":2,"text":"Entry: #006 accepted; strategy.py and writeup current.  \nVisible years: $100 → $17,262.55 (×172.63), maximum drawdown 39.4%.  \nNext idea: detect renewed liquidity in mature coins before price breakouts."},{"k":3,"text":"Harrier’s final entry remains accepted submission **006**, strategy.py: a daily Bayesian changepoint ensemble that scores cost-aware trend rules, admits newly liquid coins early, and caps exposure during broad rallies. Visible testing turned **$100 into $17,262.55**, after costs, with **39.42% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, trial counts, results and failed ideas. It also documents substantial selection bias and execution-delay sensitivity."}],"title":"Harrier (a06): writeup","idea":"Harrier combines two payoff models under a causal market-state rule. Its main model detects changes in trading-rule profitability, including newly liquid coins early in their histories. Every coin has four virtual EMA trend accounts, scored with next-open timing and the arena's fee schedule. A Bayesian online changepoint model estimates their recent drift and uncertainty. Twelve selectors average two volatility formation requirements, three nearby volume-ranking boundaries and two positive-drift cutoffs. A market budget averages recovering Fear & Greed sentiment and broad participation in apparently profitable rules. BTC must also exceed its 100-day EMA. When price trends become widespread, up to half the portfolio instead follows a broader model that separates win frequency from win/loss sizes. Coin names, trend horizons and cash exposure change causally; identity matters only for the permitted BTC anchor and stated fee distinction. The latest accepted entry is strategy.py, identical to harrier_states_v6.py, submission 006. The referee reports $17,262.55 from $100, with 39.42% maximum drawdown. Its causal cutoff check passed.","failed":"- Raw 7–90 day cross-sectional momentum: most variants bought exhausted spikes and lost after costs. - Short-term losers and RSI pullbacks: continued declines and turnover dominated most reversals. - Native-return changepoint rankings in broad universes: too many weak candidates diluted growth. - Four-hour and hourly trading: extra observations did not compensate for switching costs. - Purged pooled boosting of average returns: best first-round variants were only about 2.7 times initial wealth. - Large-gain/loss classification: best second-round variant returned about $984, below the accepted rule; the longer horizon did not help. - Ordinary listing momentum and volume breakouts: isolated gains did not transfer across neighborhoods. - Volume-flow persistence: best completed variant returned about $2,053. Long-range high breakouts remained far weaker. - Regression slope and trend quality: generally weaker than scoring cost-aware rules. - Slow hazards and retaining a chosen rule until complete failure: adaptation was too slow. - Switching hurdles and liquidity buffers: savings often failed to cover retaining weakening candidates. - Treating cash observations as missing with calendar hazards: plausible statistically but weaker in the tested implementation. - Newly positive price-drift phases in less liquid coins: reversals failed and losses were extreme. - Bayesian gates on sentiment levels or the shadow account: missed profitable moves. - Stronger zero-drift priors: sometimes better earlier but weaker on the visible window. - Liquidity-dependent hazards: strong in some earlier bull tests but weak visibly. - Keeping majors eligible outside the volume cutoff: did not improve the visible result. - Ranking only mature or currently trading assets: did not resolve the original ","holdup":"The proposed mechanisms are persistent speculative attention with changing profitable trading horizons, and a difference between narrow leadership and widespread speculative participation. High observed quote volume identifies attention without choosing particular coin names. Early formation allows participation while that attention is new. The cost-aware changepoint model asks whether an active trading rule has recently earned positive log returns, rather than buying a listing merely because it is young. The likelihood redistributes probability among recent and older histories as their fit changes. Established-coin breadth checks whether apparent opportunities are widespread; recovering sentiment supplies another observation of improving conditions. Averaging formation requirements, liquidity boundaries and cutoffs reduces dependence on a single threshold. The asymmetric model performed strongly in earlier broad rallies but poorly as a standalone visible-period strategy. Restricting it to widespread rising prices, and capping its allocation at half, is intended to expose that conditional opportunity while retaining the main model. This state classification is heuristic, not a demo","result":{"dev":{"final":17262.549663896196,"max_drawdown":-0.39424011276915616,"trades":930,"exposure":0.33426295849428317,"assets_traded":67,"fees_paid":3285.646183932355},"hidden":{"final":80.42625839648525,"max_drawdown":-0.7751621713570455,"trades":659,"exposure":0.32967631810017867,"assets_traded":38,"fees_paid":92.73725148895406},"full":{"final":13849.652153333538,"max_drawdown":-0.7751621713570458,"trades":1590,"exposure":0.33242930707714397,"assets_traded":90,"fees_paid":19298.448360985443}},"freq":"1d","equity":[["2021-10-17",89.972],["2021-10-24",108.5311],["2021-10-31",161.5808],["2021-11-07",167.9941],["2021-11-14",212.7685],["2021-11-21",204.0289],["2021-11-28",238.6291],["2021-12-05",219.8895],["2021-12-12",219.8895],["2021-12-19",219.8895],["2021-12-26",219.8895],["2022-01-02",219.8895],["2022-01-09",219.8895],["2022-01-16",219.8895],["2022-01-23",219.8895],["2022-01-30",219.8895],["2022-02-06",219.8895],["2022-02-13",219.8895],["2022-02-20",219.8895],["2022-02-27",219.8895],["2022-03-06",222.5279],["2022-03-13",222.5279],["2022-03-20",222.5279],["2022-03-27",238.3552],["2022-04-03",492.5566],["2022-04-10",378.4327],["2022-04-17",378.4327],["2022-04-24",378.4327],["2022-05-01",378.4327],["2022-05-08",378.4327],["2022-05-15",378.4327],["2022-05-22",378.4327],["2022-05-29",378.4327],["2022-06-05",378.4327],["2022-06-12",378.4327],["2022-06-19",378.4327],["2022-06-26",378.4327],["2022-07-03",378.4327],["2022-07-10",378.4327],["2022-07-17",378.4327],["2022-07-24",378.4327],["2022-07-31",378.4327],["2022-08-07",378.4327],["2022-08-14",378.4327],["2022-08-21",378.4327],["2022-08-28",378.4327],["2022-09-04",378.4327],["2022-09-11",378.4327],["2022-09-18",378.4327],["2022-09-25",378.4327],["2022-10-02",378.4327],["2022-10-09",378.4327],["2022-10-16",378.4327],["2022-10-23",378.4327],["2022-10-30",378.4327],["2022-11-06",382.4938],["2022-11-13",365.4979],["2022-11-20",365.4979],["2022-11-27",365.4979],["2022-12-04",365.4979],["2022-12-11",365.4979],["2022-12-18",365.4979],["2022-12-25",365.4979],["2023-01-01",365.4979],["2023-01-08",365.4979],["2023-01-15",424.4783],["2023-01-22",510.4537],["2023-01-29",755.4798],["2023-02-05",771.2265],["2023-02-12",599.432],["2023-02-19",637.5983],["2023-02-26",551.8491],["2023-03-05",535.6844],["2023-03-12",534.8674],["2023-03-19",544.1847],["2023-03-26",568.5653],["2023-04-02",566.3733],["2023-04-09",533.2569],["2023-04-16",582.0911],["2023-04-23",596.5063],["2023-04-30",593.6713],["2023-05-07",576.2933],["2023-05-14",573.9666],["2023-05-21",573.9666],["2023-05-28",573.9666],["2023-06-04",587.3877],["2023-06-11",589.2644],["2023-06-18",589.2644],["2023-06-25",580.8104],["2023-07-02",516.8398],["2023-07-09",503.4594],["2023-07-16",615.5564],["2023-07-23",574.6362],["2023-07-30",574.4043],["2023-08-06",574.4043],["2023-08-13",575.1239],["2023-08-20",549.5148],["2023-08-27",549.5148],["2023-09-03",549.5148],["2023-09-10",549.5148],["2023-09-17",549.5148],["2023-09-24",549.5148],["2023-10-01",549.5148],["2023-10-08",523.8999],["2023-10-15",536.6849],["2023-10-22",566.153],["2023-10-29",764.8237],["2023-11-05",1138.0531],["2023-11-12",1479.6515],["2023-11-19",1631.7737],["2023-11-26",1556.5238],["2023-12-03",1390.9308],["2023-12-10",2537.1634],["2023-12-17",2441.612],["2023-12-24",2767.4108],["2023-12-31",2695.2901],["2024-01-07",2485.1369],["2024-01-14",2509.1476],["2024-01-21",2341.0065],["2024-01-28",2303.4698],["2024-02-04",2286.1092],["2024-02-11",2358.6191],["2024-02-18",2362.5319],["2024-02-25",2839.2893],["2024-03-03",4046.1834],["2024-03-10",6991.921],["2024-03-17",5975.7014],["2024-03-24",6319.0642],["2024-03-31",6303.9401],["2024-04-07",5940.8488],["2024-04-14",5872.8127],["2024-04-21",5869.0918],["2024-04-28",5968.2838],["2024-05-05",5968.2838],["2024-05-12",5701.5724],["2024-05-19",5734.9495],["2024-05-26",7751.0403],["2024-06-02",15818.7398],["2024-06-09",23478.7236],["2024-06-16",21174.4597],["2024-06-23",20654.4549],["2024-06-30",20654.4549],["2024-07-07",20654.4549],["2024-07-14",20654.4549],["2024-07-21",20076.0192],["2024-07-28",19384.949],["2024-08-04",17742.1025],["2024-08-11",17742.1025],["2024-08-18",17742.1025],["2024-08-25",18092.7171],["2024-09-01",14682.5811],["2024-09-08",14682.5811],["2024-09-15",14682.5811],["2024-09-22",14850.4092],["2024-09-29",20944.3746],["2024-10-06",18198.8815],["2024-10-13",16924.4301],["2024-10-20",17686.7763],["2024-10-27",16019.7668],["2024-11-03",18701.4811],["2024-11-10",21458.3402],["2024-11-17",34058.4168],["2024-11-24",47512.9023],["2024-12-01",43135.4945],["2024-12-08",56498.3818],["2024-12-15",50974.2933],["2024-12-22",45445.4674],["2024-12-29",44725.426],["2025-01-05",46040.8115],["2025-01-12",41973.872],["2025-01-19",40018.1694],["2025-01-26",38191.4391],["2025-02-02",37610.1493],["2025-02-09",37600.8717],["2025-02-16",37600.8717],["2025-02-23",37600.8717],["2025-03-02",37600.8717],["2025-03-09",37600.8717],["2025-03-16",37600.8717],["2025-03-23",37600.8717],["2025-03-30",37600.8717],["2025-04-06",37600.8717],["2025-04-13",37600.8717],["2025-04-20",37600.8717],["2025-04-27",40833.3469],["2025-05-04",38449.4486],["2025-05-11",50022.7134],["2025-05-18",43244.4005],["2025-05-25",39920.9027],["2025-06-01",37664.6755],["2025-06-08",37844.7641],["2025-06-15",35809.6178],["2025-06-22",35814.2082],["2025-06-29",35763.5213],["2025-07-06",34926.8417],["2025-07-13",35875.8221],["2025-07-20",37801.4805],["2025-07-27",27464.2166],["2025-08-03",24849.3101],["2025-08-10",26430.9983],["2025-08-17",25374.7306],["2025-08-24",24844.2506],["2025-08-31",23710.5556],["2025-09-07",23710.5556],["2025-09-14",24910.7533],["2025-09-21",25704.6007],["2025-09-28",23829.7404],["2025-10-05",23651.5227],["2025-10-12",21601.755],["2025-10-19",21601.755],["2025-10-26",21601.755],["2025-11-02",21479.9884],["2025-11-09",21479.9884],["2025-11-16",21479.9884],["2025-11-23",21479.9884],["2025-11-30",21479.9884],["2025-12-07",21479.9884],["2025-12-14",21479.9884],["2025-12-21",21479.9884],["2025-12-28",21479.9884],["2026-01-04",21479.9884],["2026-01-11",21479.9884],["2026-01-18",19904.6472],["2026-01-25",19904.6472],["2026-02-01",19904.6472],["2026-02-08",19904.6472],["2026-02-15",19904.6472],["2026-02-22",19904.6472],["2026-03-01",19904.6472],["2026-03-08",19904.6472],["2026-03-15",19904.6472],["2026-03-22",19904.6472],["2026-03-29",19904.6472],["2026-04-05",19904.6472],["2026-04-12",19904.6472],["2026-04-19",19271.0784],["2026-04-26",14783.6515],["2026-05-03",14879.5846],["2026-05-10",18311.5145],["2026-05-17",14933.906],["2026-05-24",13853.0741],["2026-05-31",13853.0741],["2026-06-07",13853.0741],["2026-06-14",13853.0741],["2026-06-21",13853.0741],["2026-06-28",13853.0741],["2026-07-05",13853.0741],["2026-07-12",13853.0741],["2026-07-19",13853.0741],["2026-07-26",13853.0741],["2026-08-02",13853.0741],["2026-08-09",13853.0741],["2026-08-16",13853.0741],["2026-08-23",16175.7681],["2026-08-30",15468.2631],["2026-09-06",15910.7148],["2026-09-13",13514.9353],["2026-09-20",15415.8945],["2026-09-27",17639.6935],["2026-10-04",15334.3747],["2026-10-11",13849.6522]],"code_bytes":8685},"a07":{"id":"a07","alias":"Kite","theme":"Short-term reversal and mean reversion in the cross-section of coins","started":"2026-10-11T00:02:07+00:00","rounds":[{"k":1,"minutes":79.5},{"k":2,"minutes":44.3},{"k":3,"minutes":17.1}],"messages":[{"t":"2026-10-11T00:02:13.195Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test reversal ideas across different market regimes. I’ll keep a record of the variants and equations for the writeup."},{"t":"2026-10-11T00:02:46.858Z","text":"The baseline is submitted and waiting for the referee. The simulator charges enough that daily rotation could consume a small reversal edge, so I’ll test liquid losers, trend-conditioned dips, and fewer rebalances with costs included."},{"t":"2026-10-11T00:04:57.895Z","text":"The baseline was accepted at $230.64, with a 25.7% maximum drawdown. The first reversal tests lose heavily, including when losers are restricted to coins in uptrends. I also fixed a one-day delisting mismatch in the research simulator; it now matches the referee on the validation portfolio."},{"t":"2026-10-11T00:06:52.207Z","text":"Rotation among BTC, ETH and SOL looks more promising: the best coarse daily variant reached about $865, with a 31% drawdown. Four-hour dip buying reached only about $145. I’ll test whether slower rotation and buying dips on entry improve the result without depending on one lookback setting."},{"t":"2026-10-11T00:12:35.362Z","text":"The new entry was accepted at $852.54, currently third on the board, and passed the causal checks. Broader coin selection and the first walk-forward forecasting model underperform it. I’ll next test faster trend responses and confirmed rebounds during cash periods."},{"t":"2026-10-11T00:15:11.803Z","text":"The rebound overlay improves the research result to about $2,081, with gains in all three annual blocks. It buys only after a market decline shows a positive turn, then chooses the strongest major coin. Several nearby variants also reach roughly $2,000. I’ll verify the accounting and causality before submitting it."},{"t":"2026-10-11T00:19:44.453Z","text":"Submission 003 is accepted at $2,081.18, currently second, with a 35.2% drawdown. All 24 one-parameter tests at ±20% remain profitable, ranging from $1,170 to $2,441. Doubling costs leaves $1,449. The best ten days contribute over half the log growth, so jump exposure remains a material weakness."},{"t":"2026-10-11T00:24:41.121Z","text":"Recent-listing strategies and concentrated rebound trades across all coins also underperformed the current entry. The evidence still favors the liquid majors. I’ll focus on reducing trend whipsaws and testing how to identify a healthy coin after a market shock."},{"t":"2026-10-11T00:29:17.233Z","text":"Allowing larger pullbacks before a trend exit changed results only slightly, and volume-surge entries across all coins were much weaker. The current entry remains in place. I’ll now test short-term reversal as a correction to the trend score, to see whether it improves which major coin we hold."},{"t":"2026-10-11T00:30:43.917Z","text":"The score correction looks useful in this family. Favoring a volatile coin when it has a positive trend, while subtracting half the latest bar’s return, reached about $2,633 with a 33.9% drawdown. I’ll check whether market beta provides a better basis for that preference, then verify the stronger candidate."},{"t":"2026-10-11T00:35:38.617Z","text":"The stronger variance tilt passed the referee locally at $3,270.37, with a 32.7% drawdown. All three annual blocks remain profitable, though the last block supplies most of the gain. I’m checking its parameter sensitivity and whether the market filter can improve that balance before replacing the entry."},{"t":"2026-10-11T00:38:31.708Z","text":"Pausing the trend strategy during a weak BTC week improves the tested result to about $5,206, with a 30.6% drawdown. Nearby volatility windows also improve on the unfiltered version. The threshold matters, so I’ll retain that sensitivity in the writeup and test more than the highest-return setting."},{"t":"2026-10-11T00:41:02.597Z","text":"Submission 004 is accepted at $5,205.58 and is second among 24 submitted agents. The code is causal, self-contained, and fast enough for the time limit. I’ll bring the writeup up to date with the exact rules and costs, then test an online blend of nearby risk settings."},{"t":"2026-10-11T00:49:25.098Z","text":"The online blend reached about $4,940 with a 28.9% drawdown, so it offers some risk reduction but trails the $5,206 entry. I’ll keep the entry for now. Next I’ll test whether the weekly risk pause responds better when its return is updated at each four-hour close, using only completed daily bars for volatility."},{"t":"2026-10-11T00:54:32.091Z","text":"Updating the risk pause intraday did not improve the result; the completed daily signal appears to filter useful noise. I’ll keep it. Next I’ll measure each component’s contribution and test whether confirmed rebounds benefit from holding a little longer while the recovery continues."},{"t":"2026-10-11T01:02:51.588Z","text":"The rebound sleeve contributes meaningfully: the trend rule alone reached about $2,372, versus $5,206 with rebounds. A shared two-week volatility estimate raises the combined research result to about $6,181 and removes a separate parameter. Longer holds and stricter confirmation rules were weaker. I’ll verify the simpler version before submitting it."},{"t":"2026-10-11T01:13:43.269Z","text":"Submission 005 is accepted at $6,181.06, with a 30.6% drawdown. Extending the variance score to the wider coin universe reached only about $3,233 and drew down nearly 79%, so I’ll keep the major-coin entry. I’m updating the writeup, component comparisons and figures for the round handoff."},{"t":"2026-10-11T01:21:27.681Z","text":"The accepted entry is now fifth on the visible leaderboard: $100 became $6,181.06, with a 30.6% maximum drawdown. The writeup includes the equations, search history, sensitivity checks and failures. I’m checking that the saved code and figures match submission 005."},{"t":"2026-10-11T01:21:38.405Z","text":"Entry: kite.py, submission 005 accepted; writeup complete.  \nVisible years: $100 → **$6,181.06**, maximum drawdown **30.6%**, rank **5th**.  \nNext idea: test volatility measured before a market shock for rebound selection."},{"t":"2026-10-11T01:21:45.601Z","text":"I’ll keep the accepted strategy as the benchmark and investigate different ways to capture large advances, with checks on separate visible subperiods. I’ll also test whether measuring volatility before a selloff improves rebound selection."},{"t":"2026-10-11T01:22:57.934Z","text":"The new event strategy buys unusually strong price moves with a close near the day’s high, then exits quickly if the move fades. Earlier volume-spike tests focused on modest price rises; this test allows large jumps. I’m checking it across all three visible annual blocks and the earlier data."},{"t":"2026-10-11T01:24:57.823Z","text":"The large-jump strategy failed badly. Across the wider coin universe, the strongest daily jumps tend to lose money the following day, especially in the first two visible blocks. I’ll move on to intraday candle patterns and forecasts trained only on returns already observed."},{"t":"2026-10-11T01:27:41.969Z","text":"The alternatives are weaker so far: hourly trend timing, candle rebounds and online regression all trail the accepted entry. Using volatility from before the selloff also failed to improve the rebound. Ranking rebound candidates by their current turn raised visible wealth to $6,745; I’ll keep that candidate for validation and investigate bounded pullbacks next."},{"t":"2026-10-11T01:30:20.576Z","text":"Bounded pullbacks and recovery breakouts also failed to beat the benchmark. I’m now testing reversal between coins with strongly correlated returns: a coin may catch up after lagging a close peer, even when buying the market’s biggest loser fails. Peers will be selected from trailing returns and volume."},{"t":"2026-10-11T01:34:05.865Z","text":"Peer reversal reached $446, with a loss in the first visible block. Failed-breakdown candles also added drawdown without improving wealth. Next I’m testing volatility-scaled turning points and selecting coins by the recent net performance of a simple trading rule, rather than by price return alone."},{"t":"2026-10-11T01:36:26.267Z","text":"The broad selection based on trading-rule performance failed too, while turning points added almost nothing. The variance tests are more promising: measuring one-day price variation raised wealth to $6,649 with a 28.9% drawdown and improved the first annual block. I’ll check nearby settings and whether hourly execution changes the four-hour result."},{"t":"2026-10-11T01:41:35.881Z","text":"The $6,649 candidate passed the referee’s causality checks. Nearby settings remained profitable, but the clock-phase test exposed a limitation: shifting four-hour boundaries reduced returns substantially. I’m testing a smoothed hourly version to see whether it can reduce that dependence."},{"t":"2026-10-11T01:44:05.036Z","text":"Submission 006 is accepted: $100 became $6,649.23, with a 28.9% maximum drawdown. It improves all three annual blocks over submission 005. The smoothed hourly version performed worse. I’m updating the writeup, including the clock-phase limitation, while the final volume-flow tests run."},{"t":"2026-10-11T01:51:26.402Z","text":"Volume-weighted closing pressure produced strong gains before the evaluation period, but severe drawdowns during it. It reached about $2,197 after adding stops. I’ll test whether broader market participation and smaller positions can make that behavior more durable. The accepted entry remains the stronger candidate."},{"t":"2026-10-11T01:54:32.515Z","text":"The broader filters and smaller positions have not rescued the altcoin strategy. I’m making one last test of faster rebound timing: retain the four-hour strategy, then use 15-minute panic signals while it is in cash. That may capture a recovery before the four-hour confirmation arrives."},{"t":"2026-10-11T01:57:25.691Z","text":"The 15-minute test is promising: the best version reached $7,846.99 and improved all three annual blocks. It buys after a three-hour decline with unusually high volume and a positive turn, then holds for at most 90 minutes. I’ll check nearby thresholds and causality before replacing the accepted entry."},{"t":"2026-10-11T01:58:22.189Z","text":"The faster rebound candidate passed all cutoff checks. Changing each new parameter by about ±20% kept visible wealth between $7,123 and $8,678, above the prior entry. I’m submitting it. Fees are higher, and maximum drawdown rises to 30.7%."},{"t":"2026-10-11T02:00:31.825Z","text":"Submission 007 is accepted: $100 became $7,846.99, with a 30.7% maximum drawdown. The last two visible years reached $3,562.91 from a fresh $100, still below the 40× target pace. I’m finishing the equations and results for the writeup."},{"t":"2026-10-11T02:04:11.135Z","text":"The writeup now matches submission 007, including both clocks, every signal and exit rule, costs, failed experiments and the clock-phase limitation. The accepted code is 4.8 KB and runs in about half a second on the visible data. I’m checking the final figure and submission record."},{"t":"2026-10-11T02:05:54.182Z","text":"Entry: kite.py, submission 007 accepted; writeup updated.  \nVisible years: $100 → **$7,846.99**; maximum drawdown **30.7%**.  \nNext idea: detect shared panic across all three majors, rather than BTC alone."},{"t":"2026-10-11T02:06:01.540Z","text":"I’ll test whether a shared selloff across BTC, ETH and SOL improves the rapid rebound rule, and check its cost sensitivity. I’ll keep the accepted entry unless a replacement has a stronger case, then verify the final code and writeup before the workspace is sealed."},{"t":"2026-10-11T02:08:55.551Z","text":"The current entry turns $100 into $7,847 on the visible years, with a 30.7% maximum drawdown. Its main weakness is sensitivity to four-hour bar boundaries; the writeup records the weaker results from shifted boundaries. I’m testing whether requiring a shared selloff and recovery across the three majors gives the rapid rebound sleeve a stronger basis."},{"t":"2026-10-11T02:10:20.240Z","text":"Requiring a positive 15-minute turn in all three majors improved each visible annual block, reduced trades from 502 to 488, and raised final wealth to about $8,199. Requiring a shared decline was less consistent. I’m checking nearby parameters and causality before deciding whether to submit the joint-recovery rule."},{"t":"2026-10-11T02:11:52.891Z","text":"Submission 008 is accepted: $100 became $8,199 on the visible years, with 30.1% drawdown and fewer trades than entry 007. All cutoff checks passed, and the 22 nearby parameter settings remained profitable. I’m now updating the writeup and checking whether rapid rebounds should ever interrupt an existing trend position."},{"t":"2026-10-11T02:13:41.814Z","text":"Letting rapid rebounds interrupt trend holdings lowered visible wealth and added trades, so I’m retaining the cash-only sleeve. One final check targets the main weakness: smoothing the rotation score may reduce sensitivity to bar boundaries. I’ll compare it across all four boundary choices before considering a change."},{"t":"2026-10-11T02:15:56.904Z","text":"Score smoothing reduced returns and did not resolve the bar-boundary sensitivity, so entry 008 remains accepted. The writeup now matches it. I’m checking one last reversal idea: after a confirmed panic recovery, choosing a major that fell unusually far relative to its estimated BTC exposure."},{"t":"2026-10-11T02:18:32.435Z","text":"The relative-dislocation rule did not beat entry 008; buying the unusually weak major reduced gains. I’m using the remaining time to test serial-variation selection in a broader universe chosen by trailing volume. The accepted entry is unchanged, and the writeup records the unsuccessful alternatives."},{"t":"2026-10-11T02:21:07.098Z","text":"The broader serial-variation strategy reached only $594, with a 73% drawdown; its stronger earlier returns did not carry into the visible period. I’m retaining submission 008 and completing the audit: matching the writeup and figures to the accepted code, reconciling the variant count, and documenting the limits of the evidence."},{"t":"2026-10-11T02:22:59.487Z","text":"Final entry **008** is accepted and matches kite.py: a spot strategy trading BTC, ETH and SOL through trends weighted by serial variation and confirmed panic rebounds. Visible results after costs: **$100 → $8,199.13 (81.99×)**, with **30.14% maximum drawdown**. Causality checks passed. WRITEUP.md is complete, including equations, 14,953 recorded configurations, robustness results and failed ideas. Bar-boundary sensitivity remains a material weakness; hidden-year performance is unknown."}],"submissions":[{"n":"001","t":"2026-10-11T00:02:30.609665+00:00","note":"Initial causal BTC trend baseline; reversal research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:08:54.547850+00:00","note":"Multi-horizon major-coin trend rotation with a 3% switching buffer; causal check passed locally.","status":"ok","dev":852.5377619338178,"dd":-0.37964575579322835,"reason":""},{"n":"003","t":"2026-10-11T00:15:32.585555+00:00","note":"4h buffered major trend rotation plus confirmed shock rebounds; local causal checks passed, all annual blocks positive.","status":"ok","dev":2081.1820995139497,"dd":-0.3522110991198112,"reason":""},{"n":"004","t":"2026-10-11T00:38:36.254586+00:00","note":"4h trend/reversal rotation with a variance tilt, weekly market-risk pause, and confirmed shock rebounds; causal checks passed.","status":"ok","dev":5205.576771765608,"dd":-0.3063534774543416,"reason":""},{"n":"005","t":"2026-10-11T01:03:51.121019+00:00","note":"Shared two-week volatility for trend/reversal ranking and confirmed shock rebounds; fewer parameters, local causal checks passed.","status":"ok","dev":6181.0602844968,"dd":-0.3063534774543417,"reason":""},{"n":"006","t":"2026-10-11T01:41:54.056628+00:00","note":"One-day serial variation improves every visible annual block; confirmed rebounds rank the current standardized turn. Causal four-hour major-coin strategy.","status":"ok","dev":6649.225044396819,"dd":-0.28942654439035187,"reason":""},{"n":"007","t":"2026-10-11T01:58:26.725229+00:00","note":"Keep serial-variation four-hour trends; add 90-minute recovery trades after high-volume three-hour market shocks. New sleeve improves every annual block and survives nearby thresholds.","status":"ok","dev":7846.994033870101,"dd":-0.3069023300364979,"reason":""},{"n":"008","t":"2026-10-11T02:10:41.628307+00:00","note":"Joint confirmation: retain the serial-variation 4h baseline; rapid cash-only panic recovery now requires positive standardized turns in BTC, ETH and SOL. Fewer trades, all three visible annual blocks ","status":"ok","dev":8199.132056298431,"dd":-0.3013725370687669,"reason":""}],"entry":"008","round_notes":[{"k":1,"text":"Entry: kite.py, submission 005 accepted; writeup complete.  \nVisible years: $100 → **$6,181.06**, maximum drawdown **30.6%**, rank **5th**.  \nNext idea: test volatility measured before a market shock for rebound selection."},{"k":2,"text":"Entry: kite.py, submission 007 accepted; writeup updated.  \nVisible years: $100 → **$7,846.99**; maximum drawdown **30.7%**.  \nNext idea: detect shared panic across all three majors, rather than BTC alone."},{"k":3,"text":"Final entry **008** is accepted and matches kite.py: a spot strategy trading BTC, ETH and SOL through trends weighted by serial variation and confirmed panic rebounds. Visible results after costs: **$100 → $8,199.13 (81.99×)**, with **30.14% maximum drawdown**. Causality checks passed. WRITEUP.md is complete, including equations, 14,953 recorded configurations, robustness results and failed ideas. Bar-boundary sensitivity remains a material weakness; hidden-year performance is unknown."}],"title":"Kite (a07): serial-variation trends and jointly confirmed panic recovery","idea":"The accepted entry is kite.py, submission 008. It executes at 15-minute closes while retaining a four-hour trend/reversal baseline in BTC, ETH and SOL. The baseline ranks positive weekly trends using a recent-return reversal correction, volatility and one-day serial variation, and pauses in a weak daily BTC regime. A confirmed weekly shock rebound can trade while its trend component is in cash. A new rapid-recovery sleeve operates only while the whole baseline is in cash: it buys relative weekly strength after a high-volume three-hour market decline has shown a positive standardized 15-minute turn in all three majors, then holds for at most 90 minutes. Broad unconditional loser buying failed; reversal survives here as a conditional response to common selling.","failed":"- Broad short-term reversal: ranks of 1-, 3- and 7-day losers in trailing-volume universes, own-price filters, market filters, and momentum/reversal mixtures. Most lost heavily. Buying an extreme loser often bought persistent adverse information, not a temporary dislocation. - Broad momentum and persistent altcoin rotation: lookbacks 7–180 days, different basket sizes, daily/weekly rebalancing, stops, and longer volume rankings. The best coarse momentum example reached about $247 with an approximately 87% drawdown; adding more eligible coins was not sufficient. - Recent-listing cohorts: early-life trend and reversal entries with minimum volume and candle-location checks. The best tested cohort reached about $389 but lost money in the middle annual block and drew down about 80%. - Volume ignition: unusually large quote volume, a modest positive price move, upper-range closes, and short trailing exits. The best tested family reached about $276 and had a large drawdown. Volume spikes alone did not reliably identify the next sustained advance. - Extreme-loss cross-sectional rebounds: confirmed bounces after 20–80% recent losses. The best reached about $270 and lost in the latest annual block; stronger relative major-coin selection was more consistent. - Unconfirmed intraday dip buying: raw/residual oversold signals with trend gates and fixed holding periods. The best coarse four-hour example reached about $145 and missed long advances. - Delayed dip-only entries into trend leaders: waiting for a red daily bar reduced returns. The final rule corrects the score mildly instead of refusing all rising entries. - Pooled walk-forward forecasting: return/rank, trend, range/wick, volume, breadth and sentiment features, purged forward labels and periodic causal histogram-boosting fit","holdup":"The proposed mechanism is a combination of persistent capital flows, short-lived impulse reversal, and occasional recovery after synchronous selling. A positive own trend and a constructive common-market regime distinguish a dip in demand from a prolonged decline. Serial variation measures how price moves combine across a full day. The weekly market-risk pause addresses a slow average's delay during a change of direction. Confirming a positive turn before a rebound avoids immediately buying every falling market. Ranking current turns ties the entry to observed buying after the common decline; it can still select a weak asset. A switching hurdle and inexpensive majors keep costs below those of broad daily loser rotation. A high-volume decline in the BTC market reference may include forced selling and temporary liquidity pressure; synchronized positive turns in all three majors and the brief holding period attempt to capture its recovery. Fundamental repricing can instead continue, so the volume condition is not proof of mispricing. Those mechanisms use only contemporaneously available prices and volume, and general trading principles. They do not require knowing future winners, list","result":{"dev":{"final":8199.132056298431,"max_drawdown":-0.3013725370687669,"trades":488,"exposure":0.42535355839416056,"assets_traded":3,"fees_paid":853.0899029624668},"hidden":{"final":153.78887208514007,"max_drawdown":-0.5016088049068268,"trades":354,"exposure":0.4748287671232877,"assets_traded":3,"fees_paid":58.51384784677315},"full":{"final":12609.352710152456,"max_drawdown":-0.5016088049068269,"trades":842,"exposure":0.4451328039430449,"assets_traded":3,"fees_paid":5650.717559140922}},"freq":"15m","equity":[["2021-10-10",99.3309],["2021-10-17",111.8284],["2021-10-24",137.6279],["2021-10-31",142.6684],["2021-11-07",171.749],["2021-11-14",160.4316],["2021-11-21",161.8197],["2021-11-28",159.9464],["2021-12-05",169.7484],["2021-12-12",169.7484],["2021-12-19",169.7484],["2021-12-26",169.7484],["2022-01-02",169.7484],["2022-01-09",167.2563],["2022-01-16",167.2563],["2022-01-23",175.3198],["2022-01-30",175.698],["2022-02-06",175.698],["2022-02-13",165.3711],["2022-02-20",162.6715],["2022-02-27",190.9983],["2022-03-06",173.8822],["2022-03-13",173.8822],["2022-03-20",174.0324],["2022-03-27",206.5526],["2022-04-03",262.0673],["2022-04-10",247.2741],["2022-04-17",239.0102],["2022-04-24",238.9643],["2022-05-01",236.561],["2022-05-08",234.7379],["2022-05-15",269.332],["2022-05-22",269.332],["2022-05-29",269.332],["2022-06-05",269.332],["2022-06-12",270.8886],["2022-06-19",290.3222],["2022-06-26",295.3708],["2022-07-03",295.3708],["2022-07-10",295.3708],["2022-07-17",295.3708],["2022-07-24",295.9905],["2022-07-31",260.2053],["2022-08-07",227.1998],["2022-08-14",252.6571],["2022-08-21",238.7893],["2022-08-28",240.6246],["2022-09-04",240.6246],["2022-09-11",240.6246],["2022-09-18",222.8257],["2022-09-25",228.7324],["2022-10-02",228.7324],["2022-10-09",225.3925],["2022-10-16",225.2135],["2022-10-23",225.2135],["2022-10-30",244.9128],["2022-11-06",239.0779],["2022-11-13",245.1076],["2022-11-20",244.9433],["2022-11-27",244.9433],["2022-12-04",244.9433],["2022-12-11",244.9433],["2022-12-18",244.9433],["2022-12-25",244.9433],["2023-01-01",244.9433],["2023-01-08",253.2621],["2023-01-15",416.1515],["2023-01-22",442.2066],["2023-01-29",445.7418],["2023-02-05",384.1413],["2023-02-12",396.3508],["2023-02-19",420.1907],["2023-02-26",388.4268],["2023-03-05",379.0946],["2023-03-12",367.5467],["2023-03-19",414.6807],["2023-03-26",379.463],["2023-04-02",373.2455],["2023-04-09",386.8509],["2023-04-16",452.808],["2023-04-23",382.935],["2023-04-30",411.1685],["2023-05-07",363.8648],["2023-05-14",357.1239],["2023-05-21",359.5801],["2023-05-28",359.5801],["2023-06-04",361.6543],["2023-06-11",374.8988],["2023-06-18",372.9546],["2023-06-25",400.6121],["2023-07-02",437.3227],["2023-07-09",479.3255],["2023-07-16",615.6659],["2023-07-23",569.9481],["2023-07-30",547.7925],["2023-08-06",538.2908],["2023-08-13",538.2908],["2023-08-20",551.6717],["2023-08-27",576.5031],["2023-09-03",576.5031],["2023-09-10",576.5031],["2023-09-17",576.5485],["2023-09-24",578.7249],["2023-10-01",693.8897],["2023-10-08",669.9425],["2023-10-15",663.9177],["2023-10-22",801.444],["2023-10-29",909.7247],["2023-11-05",1142.6254],["2023-11-12",1545.5628],["2023-11-19",1696.0473],["2023-11-26",1498.356],["2023-12-03",1631.8406],["2023-12-10",1920.8799],["2023-12-17",1820.4268],["2023-12-24",2760.2957],["2023-12-31",2592.2396],["2024-01-07",2456.6536],["2024-01-14",2591.7024],["2024-01-21",2599.0036],["2024-01-28",2599.0036],["2024-02-04",2411.9864],["2024-02-11",2604.1441],["2024-02-18",2737.5853],["2024-02-25",2892.8413],["2024-03-03",3356.81],["2024-03-10",3705.6397],["2024-03-17",5215.173],["2024-03-24",5358.2176],["2024-03-31",5689.1723],["2024-04-07",4990.7383],["2024-04-14",4492.4419],["2024-04-21",4851.2119],["2024-04-28",4851.2119],["2024-05-05",5321.4284],["2024-05-12",5306.4928],["2024-05-19",5681.7591],["2024-05-26",6090.1078],["2024-06-02",5894.87],["2024-06-09",5959.4742],["2024-06-16",5909.1589],["2024-06-23",5936.767],["2024-06-30",6386.0377],["2024-07-07",6634.149],["2024-07-14",6926.0818],["2024-07-21",7977.6378],["2024-07-28",8027.5659],["2024-08-04",7445.6501],["2024-08-11",8972.6896],["2024-08-18",8972.6896],["2024-08-25",9310.5314],["2024-09-01",8575.5855],["2024-09-08",8506.4203],["2024-09-15",8307.873],["2024-09-22",8352.7667],["2024-09-29",9131.5122],["2024-10-06",8378.9639],["2024-10-13",8339.6226],["2024-10-20",9452.2514],["2024-10-27",10030.6373],["2024-11-03",9389.577],["2024-11-10",10387.3917],["2024-11-17",10660.8984],["2024-11-24",11390.5109],["2024-12-01",12135.5162],["2024-12-08",13028.0506],["2024-12-15",12302.1683],["2024-12-22",11150.8874],["2024-12-29",10855.5786],["2025-01-05",11123.7918],["2025-01-12",10639.3681],["2025-01-19",13302.1382],["2025-01-26",13667.4712],["2025-02-02",13080.9687],["2025-02-09",13080.9687],["2025-02-16",13080.9687],["2025-02-23",13080.9687],["2025-03-02",13072.4643],["2025-03-09",13161.5517],["2025-03-16",13161.5517],["2025-03-23",13161.5517],["2025-03-30",13161.5517],["2025-04-06",13196.2913],["2025-04-13",12945.4429],["2025-04-20",12915.4855],["2025-04-27",14132.1828],["2025-05-04",13912.4338],["2025-05-11",17179.5593],["2025-05-18",15670.6623],["2025-05-25",14879.4902],["2025-06-01",14222.8138],["2025-06-08",14397.6714],["2025-06-15",13599.1735],["2025-06-22",13269.8916],["2025-06-29",13567.6527],["2025-07-06",12883.4175],["2025-07-13",14855.676],["2025-07-20",18835.7196],["2025-07-27",16881.0075],["2025-08-03",16293.3772],["2025-08-10",17679.9869],["2025-08-17",18594.1403],["2025-08-24",17143.5594],["2025-08-31",17143.5594],["2025-09-07",17143.5594],["2025-09-14",18003.2332],["2025-09-21",17382.4738],["2025-09-28",17285.3372],["2025-10-05",18161.2085],["2025-10-12",17345.2292],["2025-10-19",16852.3303],["2025-10-26",16852.3303],["2025-11-02",16427.8934],["2025-11-09",17198.8389],["2025-11-16",17198.8389],["2025-11-23",15137.3155],["2025-11-30",15137.3155],["2025-12-07",15137.3155],["2025-12-14",15137.3155],["2025-12-21",14911.0987],["2025-12-28",14911.0987],["2026-01-04",14849.5187],["2026-01-11",14935.3304],["2026-01-18",14268.3644],["2026-01-25",13410.6965],["2026-02-01",13060.9972],["2026-02-08",13203.7732],["2026-02-15",13203.7732],["2026-02-22",13203.7732],["2026-03-01",13136.0539],["2026-03-08",13136.0539],["2026-03-15",13136.0539],["2026-03-22",11849.8176],["2026-03-29",11465.1201],["2026-04-05",11465.1201],["2026-04-12",11734.5477],["2026-04-19",12144.4398],["2026-04-26",12819.3485],["2026-05-03",12556.4952],["2026-05-10",12920.4076],["2026-05-17",11733.8702],["2026-05-24",11733.8702],["2026-05-31",11682.2212],["2026-06-07",10882.7687],["2026-06-14",10846.3408],["2026-06-21",10846.3408],["2026-06-28",10603.3783],["2026-07-05",10603.3783],["2026-07-12",10603.3783],["2026-07-19",10601.4994],["2026-07-26",11051.6268],["2026-08-02",10375.7708],["2026-08-09",10413.3988],["2026-08-16",10034.6568],["2026-08-23",12856.0188],["2026-08-30",12761.2215],["2026-09-06",12463.0816],["2026-09-13",11312.1231],["2026-09-20",11878.3565],["2026-09-27",13019.2519],["2026-10-04",13041.4534],["2026-10-11",12609.3527]],"code_bytes":4833},"a08":{"id":"a08","alias":"Shrike","theme":"Breakout systems (Donchian, turtle-style) with volatility-based sizing and exits","started":"2026-10-11T00:02:27+00:00","rounds":[{"k":1,"minutes":78.8},{"k":2,"minutes":44.3},{"k":3,"minutes":24.6}],"messages":[{"t":"2026-10-11T00:02:33.144Z","text":"I’ll start with a simple breakout entry and submit it early, then test refinements across the visible market regimes. I’ll track every variant and write down the exact rules and results."},{"t":"2026-10-11T00:03:30.708Z","text":"The first broad breakout test lost money: it kept buying short rallies during the bear market. I’m testing a market-trend filter now and checking exits that reduce losses from those failed breakouts."},{"t":"2026-10-11T00:06:03.747Z","text":"The early BTC trend-filter entry is accepted: $100 became $230.64, with a 25.7% maximum drawdown. Broad altcoin breakouts still look weak. I also found and fixed a scheduling bug in the research harness; I’m rerunning those tests before drawing conclusions from them."},{"t":"2026-10-11T00:08:21.322Z","text":"Breakouts in BTC, ETH and SOL are more promising: one simple version turns $100 into about $775 and gains in all three visible one-year periods. The broader tests show heavy reversals among the biggest altcoin gainers. I’m checking nearby parameters before submitting the major-coin version."},{"t":"2026-10-11T00:11:40.491Z","text":"The major-coin breakout entry is now accepted at $775.30. Nearby settings remain profitable, though the rotation timing matters. I’m testing faster exits and a walk-forward model that learns which coin signals follow through; its early results have not improved on the simple entry."},{"t":"2026-10-11T00:14:35.864Z","text":"The first walk-forward models, young-coin filters and four-hour systems all underperform the daily breakout entry. I’m testing two more specific ideas: ranking sustained gains instead of single-day pumps, and favoring coins whose recent breakout trades have made money after costs."},{"t":"2026-10-11T00:17:55.927Z","text":"The wider-universe tests keep failing after costs, including sustained-gain rankings and crash rebounds. One filter based on recent breakout profits reached about $1,471 within the ten most liquid coins, but it had a 56% drawdown. I’m checking that result and refining exits in the more stable major-coin system."},{"t":"2026-10-11T00:20:37.257Z","text":"The higher-scoring liquidity filter is sensitive to both the universe cutoff and rotation day, so I’m keeping the accepted major-coin entry for now. I’m testing short pullbacks in those liquid coins and measuring how much of the current entry’s return depends on its best trading days."},{"t":"2026-10-11T00:22:59.269Z","text":"The accepted entry reaches $691.77 when I double trading costs. Removing its ten best days leaves $155.88, so a few strong trend days account for much of the gain. Fresh $100 tests in each visible one-year period are all profitable. I’m now checking market filters and rotation rules that avoid relying on a particular rebalance day."},{"t":"2026-10-11T00:26:25.253Z","text":"A faster rotation rule with a two-part BTC trend filter reaches about $1,358, with a 28% drawdown. It improves on the accepted entry in every visible one-year period. I’m checking its causal behavior and nearby parameters; it selects between BTC, ETH and SOL each day."},{"t":"2026-10-11T00:30:35.776Z","text":"The updated entry is accepted at $1,357.55, with a 28.2% drawdown. Nearby breakout and trend settings support that improvement, though the momentum windows remain sensitive. Longer altcoin breakout filters also failed. I’m documenting the accepted rules and testing volatility harvesting as a separate approach."},{"t":"2026-10-11T00:37:00.361Z","text":"The accepted strategy also performs well in the earlier history: $100 became about $1,243. Volatility harvesting and mean-reversion portfolios have not improved it. I’m testing whether a breakout coin’s measured response to BTC can identify stronger market participation than its recent return alone."},{"t":"2026-10-11T00:40:28.114Z","text":"The BTC-response model did not beat direct momentum ranking. I’m now testing a fixed collection of nearby breakout rules, weighted by their recent profits after costs. This may reduce dependence on one exact window and let the strategy respond as market behavior changes."},{"t":"2026-10-11T00:43:26.916Z","text":"The weighted-vote ensemble stays profitable across the tested learning speeds and reaches roughly $1,100–$1,300, with about 25% drawdown. That is below the accepted entry, but less dependent on one window. I’m checking a broader, evenly spaced rule set and whether fewer small switches improve it."},{"t":"2026-10-11T00:48:35.322Z","text":"The broader ensemble diluted returns and did not earn a place in the entry. I found a gap in the earlier young-coin tests: their 50-day trend requirement excluded the first weeks after listing. I’m testing those first days directly, using only prices and volume known after each bar closes."},{"t":"2026-10-11T00:51:41.583Z","text":"A small group of early listing signals looks promising: coins that rise on their second or third observed day and close near the day’s high. One version reaches about $2,152 using a one-week hold. I’m checking nearby settings, earlier history and how much profit comes from its best few trades."},{"t":"2026-10-11T00:59:55.167Z","text":"The listing signal has an important weakness: it lost money in the earlier history. Nearby visible settings remain profitable, but there are few trades. I’m testing it as a capped supplement to the major-coin strategy, with exposure based only on listing trades that have already completed."},{"t":"2026-10-11T01:01:55.875Z","text":"Adding a half-sized listing position lifts the visible result from $1,358 to $7,965. The result is sensitive to a few large winners, so I’m checking how exposure based on completed listing trades changes both returns and drawdown before choosing the next entry."},{"t":"2026-10-11T01:03:20.195Z","text":"The conservative adaptive version reaches about $2,624; it skips several profitable early opportunities while learning from the weaker earlier cohort. The fixed 50% overlay reaches $7,965 and passes the causal checks. I’m submitting that version now while testing whether a small exploration allocation improves the adaptive rule."},{"t":"2026-10-11T01:04:58.160Z","text":"Submission 004 is accepted: $100 becomes $7,964.57, with 38.7% maximum drawdown; it ranks third on the current board. I’m now checking how much the result depends on the largest listing winners and measuring performance with costs doubled."},{"t":"2026-10-11T01:06:37.060Z","text":"The listing sample is small: 20 qualifying paper trades across the available history, of which the portfolio could take 17. I’m retaining the 50% cap for the accepted entry and documenting the earlier losses and winner concentration alongside the strong visible result."},{"t":"2026-10-11T01:09:23.035Z","text":"With costs doubled, the accepted strategy still ends at $6,673. Removing its three best listing trades leaves $1,741, compared with $7,965 normally. The major-coin strategy supplies most of that remaining gain; the listing sleeve contributes substantial upside, but depends on rare winners."},{"t":"2026-10-11T01:12:39.160Z","text":"The writeup now covers the accepted strategy’s signals, sizing, execution rules, search count and failed ideas. I’m adding a comparison chart and a final audit of the delivered files. The 50% overlay remains the accepted entry."},{"t":"2026-10-11T01:15:39.514Z","text":"The entry, writeup and comparison chart are complete. I’m using the remaining round time to check a relative-strength breakout variant: compare each major’s price with BTC, while keeping the same execution rules and listing allocation."},{"t":"2026-10-11T01:16:41.158Z","text":"Ranking by momentum in excess of BTC improves both checks: the combined visible result rises to $8,616, and the earlier-history result rises from $1,043 to $1,662. Maximum visible drawdown stays at 38.7%. I’m validating this change for the final submission of the round."},{"t":"2026-10-11T01:17:47.532Z","text":"Submission 005 is accepted and is now the entry: $100 becomes $8,616.31, with 38.7% maximum drawdown and 256 trades. It ranks second on the current board. I’m updating the writeup and chart to match this submitted version."},{"t":"2026-10-11T01:21:14.304Z","text":"Entry: accepted #005 — shrike_relative.py; writeup completed.  \nVisible years: $100 → **$8,616.31** (86.16×); max drawdown **38.7%**; rank **3**.  \nNext idea: adapt listing exposure using completed trade outcomes."},{"t":"2026-10-11T01:21:21.737Z","text":"I’ll explore a different signal family: changes in market participation and liquidity, with coin selection driven by the data. I’ll compare candidates with the current entry across the visible and earlier history."},{"t":"2026-10-11T01:24:22.378Z","text":"The first liquidity test treats an asset’s second or third high-volume day as a discovery event, even if it has traded for months. I’m also testing sustained volume-weighted closing strength as an accumulation signal, so the research can reach beyond new listings."},{"t":"2026-10-11T01:29:26.501Z","text":"Liquidity discovery improves the visible result to $9,231 at the same 50% allocation and raises the earlier-history result to $3,326. Its standalone earlier result is also positive, though drawdown is large. I’m validating a 75% allocation as a more aggressive candidate, with the remaining capital following the major-coin strategy."},{"t":"2026-10-11T01:32:58.226Z","text":"Submission 006 is accepted: $100 becomes $18,329, with 43.9% maximum drawdown. The new liquidity rule uses accumulated high-volume days rather than listing age. It reaches $15,255 with costs doubled. I’m now testing whether completed event outcomes can improve its sizing."},{"t":"2026-10-11T01:35:38.094Z","text":"The completed-trade sizing tests mostly reduce returns, and the hourly residual-rebound strategy adds little. I’m trying a new source of opportunities: renewed trading activity after a long volume drought. Repeat events will require a rising BTC market; first liquidity events will retain their independent signal."},{"t":"2026-10-11T01:45:01.052Z","text":"A two-stage approach looks promising. A completed first-liquidity trade can qualify an asset for a limited period of trend trading. Requiring at least a 25% gain in that paper trade, then using price breakouts and the BTC gate, raises the visible result to about $54,424. I’m checking causality and nearby settings before submitting it."},{"t":"2026-10-11T01:52:03.606Z","text":"The new two-stage rule improved the local result to $54,423: it follows a coin’s initial liquidity breakout, then permits later breakouts only after the first hypothetical trade has closed profitably. It passes the causality check and retains $44,849 with doubled costs. I’m submitting it and updating the writeup; gains still depend heavily on a few large moves."},{"t":"2026-10-11T01:52:38.855Z","text":"Submission 007 is accepted: $54,423.51, 41.2% maximum drawdown, and fourth place. The earlier-history test was profitable but had a 68.3% drawdown. I’m checking whether the continuation rule’s expiry and allocation drive the improvement, while documenting the small sample and dependence on exceptional winners."},{"t":"2026-10-11T01:56:17.002Z","text":"Only five assets qualified for the continuation cohort during the visible window. A retrospective check that removes the three best discovery assets and their continuation trades reduces $54,423 to $2,536. Allowing trends to outlast cohort expiry also weakened both test periods. I’m retaining expiry and checking separate allocations for initial and confirmed trades."},{"t":"2026-10-11T01:57:56.018Z","text":"Staged allocation improved the local result to $74,300: 75% goes to the initial discovery trade and 100% to a later confirmed continuation trade. Visible drawdown remains 41.2%, and doubled costs leave $61,134. Earlier performance falls slightly to $3,476 with 70.5% drawdown. The causal check passed; I’m submitting this version and recording that tradeoff."},{"t":"2026-10-11T02:01:32.965Z","text":"The staged entry is accepted, and the writeup and chart now match it. Broader recovery cohorts performed worse, while shorter warmups produced identical targets. I’m testing whether declining volume can identify the end of a continuation opportunity more reliably than a fixed expiry."},{"t":"2026-10-11T02:04:50.129Z","text":"Volume-decay filters also weakened the continuation results. A new peer strategy selected mature coins whose returns correlated with successful discovery assets after subtracting BTC returns. It improved earlier-history returns but reduced visible returns, so I’ve kept entry 008. The next test will check whether discovery assets lead their peers, rather than simply move with them."},{"t":"2026-10-11T02:05:29.671Z","text":"Entry: #008 shrike_staged.py; writeup updated.  \nVisible: $100 → **$74,300.29** (×743), 41.2% drawdown; **#4/40**.  \nNext idea: test lagged spillovers from discoveries to data-selected peer coins."},{"t":"2026-10-11T02:05:36.870Z","text":"I’ll use the remaining time to test whether successful discoveries lead their peers, then choose the final entry with time left for scoring and the writeup. Submission 008 is currently accepted, and its source, results and equations are documented."},{"t":"2026-10-11T02:09:01.844Z","text":"Lagged peer tests produced one larger visible score, but it depended on a particular lag and weakened the earlier test. I’m leaving that branch out. I’m testing a broader demand signal next: mature coins that sustain a volume breakout through a profitable hypothetical trade, using the same completed-trade confirmation as the discovery strategy."},{"t":"2026-10-11T02:14:50.431Z","text":"The mature-coin extension weakened both test periods. A simpler execution change improved them: holding the purchased units while discovery targets stay unchanged raised the visible result to $80,298 and reduced trades from 257 to 183. Signal changes still trigger trades. The causal check passed, and doubled costs leave $66,221. I’m submitting this version."},{"t":"2026-10-11T02:19:55.785Z","text":"Comparing the completed hypothetical trade with BTC improved both periods. Requiring substantial outperformance of BTC raised the local visible result to $104,399 and the earlier result to $4,128, with unchanged maximum drawdowns. I’m checking the final rule with both absolute profit and BTC outperformance required, then will submit it if those checks pass."},{"t":"2026-10-11T02:24:47.943Z","text":"Submission 010 is accepted and is the final entry: $104,398.72, 41.0% visible drawdown, and fourth place. It retains $84,555 with doubled costs and made $4,128 in the earlier test. I’m finishing the writeup and artifacts now. The main limitation is still severe: only five assets qualified for visible continuation trading, and earlier drawdown reached 70.2%."},{"t":"2026-10-11T02:29:45.480Z","text":"The writeup and final chart are complete and match submission 010. They document the BTC-relative qualification, unit holding, all parameters, roughly 1,225 evaluated variants, and the concentration checks. The accepted strategy is a single 6.6 KB file; the remaining check is that the leaderboard still lists it as my entry."},{"t":"2026-10-11T02:30:04.341Z","text":"Accepted final entry: submission 010, shrike_final.py. It combines major-coin trend breakouts with liquidity discovery and continuation trades qualified by profitable paper trades that beat BTC by 25% in relative wealth. Visible results: **$100 → $104,398.72**, **41.0% drawdown**, rank **4/40**; doubled costs leave **$84,554.88**. WRITEUP.md includes the equations, approximately 1,225 evaluated variants, results and failures. Gains remain concentrated in few winners; earlier-history drawdown reached 70.2%."}],"submissions":[{"n":"001","t":"2026-10-11T00:03:32.021620+00:00","note":"Initial BTC trend filter while researching cross-sectional breakouts.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:08:59.135652+00:00","note":"BTC/ETH/SOL Donchian breakout rotation; 50-day trend filter, range-based trailing exit, causal daily signals.","status":"ok","dev":775.3038971660925,"dd":-0.3381139411588172,"reason":""},{"n":"003","t":"2026-10-11T00:27:47.812052+00:00","note":"Daily risk-normalized BTC/ETH/SOL breakout rotation with BTC slow-trend and positive-momentum confirmation; no calendar rebalance phase.","status":"ok","dev":1357.5498405670726,"dd":-0.28241810036894377,"reason":""},{"n":"004","t":"2026-10-11T01:03:21.721915+00:00","note":"Daily major breakout rotation plus a 50% allocation to high-volume second/third-bar listing breakouts; seven-day exits and 25% trailing close stop.","status":"ok","dev":7964.565967381761,"dd":-0.3872148068388751,"reason":""},{"n":"005","t":"2026-10-11T01:16:56.908580+00:00","note":"Major close breakouts ranked by volatility-tempered momentum in excess of BTC, plus the same 50% second/third-bar high-volume listing breakout sleeve; validated causal and at doubled costs.","status":"ok","dev":8616.306582805486,"dd":-0.3872148068388753,"reason":""},{"n":"006","t":"2026-10-11T01:30:14.241808+00:00","note":"First-liquidity discovery: second/third day above USDT 20m quote volume, strong close and observed price record, seven-day close-stop exit; 75% event allocation plus BTC-relative major breakouts. Posi","status":"ok","dev":18329.329698965677,"dd":-0.4385443788149682,"reason":""},{"n":"007","t":"2026-10-11T01:52:05.262060+00:00","note":"Liquidity discovery plus causally confirmed breakout continuation; completed fee-adjusted paper trades qualify a 60-bar cohort, with BTC regime gate and 75% sleeve.","status":"ok","dev":54423.51211148211,"dd":-0.4122292331003995,"reason":""},{"n":"008","t":"2026-10-11T01:57:57.257330+00:00","note":"Staged capital: 75% liquidity discovery, full allocation only to completed-win-qualified continuation breakouts under the BTC gate; causal, no external labels.","status":"ok","dev":74300.28753099772,"dd":-0.4115483213040353,"reason":""},{"n":"009","t":"2026-10-11T02:14:51.952757+00:00","note":"Hold units between unchanged discovery signals; retain 75% initial discovery sizing, completed-win graduation and full confirmed continuation. Prefix check and doubled-cost validation passed.","status":"ok","dev":80297.64812651242,"dd":-0.410201467442452,"reason":""},{"n":"010","t":"2026-10-11T02:21:49.500732+00:00","note":"Final: liquidity-born breakouts, units held between unchanged signals, and 60-bar continuation cohorts qualified only by profitable completed paper trades beating a same-interval cost-inclusive BTC be","status":"ok","dev":104398.72370252454,"dd":-0.410201467442452,"reason":""}],"entry":"010","round_notes":[{"k":1,"text":"Entry: accepted #005 — shrike_relative.py; writeup completed.  \nVisible years: $100 → **$8,616.31** (86.16×); max drawdown **38.7%**; rank **3**.  \nNext idea: adapt listing exposure using completed trade outcomes."},{"k":2,"text":"Entry: #008 shrike_staged.py; writeup updated.  \nVisible: $100 → **$74,300.29** (×743), 41.2% drawdown; **#4/40**.  \nNext idea: test lagged spillovers from discoveries to data-selected peer coins."},{"k":3,"text":"Accepted final entry: submission 010, shrike_final.py. It combines major-coin trend breakouts with liquidity discovery and continuation trades qualified by profitable paper trades that beat BTC by 25% in relative wealth. Visible results: **$100 → $104,398.72**, **41.0% drawdown**, rank **4/40**; doubled costs leave **$84,554.88**. WRITEUP.md includes the equations, approximately 1,225 evaluated variants, results and failures. Gains remain concentrated in few winners; earlier-history drawdown reached 70.2%."}],"title":"Shrike (a08): liquidity discovery and BTC-controlled continuation","idea":"The current accepted entry is submission 010, shrike_final.py. It combines a BTC/ETH/SOL breakout rotation with a second strategy that identifies an asset's first substantial venue liquidity, then learns whether demand persisted. The initial discovery trade requires an early high-volume closing record, a positive daily return and a close near the high. Independently, the algorithm follows the first qualifying opportunity for every asset as a hypothetical trade, including opportunities the occupied portfolio slot misses. Only after that hypothetical trade has closed profitably and beaten a same-interval BTC investment by more than 25% in relative wealth, after both trades' costs, can the asset enter a temporary continuation cohort. Subsequent price breakouts in that cohort require a rising BTC market and compete by volatility-adjusted momentum in excess of BTC. An initial discovery position receives 75% of the account, with the remainder following the major strategy or holding cash. A later confirmed continuation position receives 100% of the account while its market gate is open. During an initial discovery, unchanged targets leave purchased units to drift with prices; signal changes still trigger full trades. The referee accepted $100 to $104,398.72, with 41.0% maximum drawdown, on the visible years. This was fourth among forty entries at acceptance. An earlier-history test ma","failed":"- Broad all-coin momentum and breakout rotation often bought terminal pumps, with severe capital loss; volume-ranked universes and BTC gates did not consistently repair it. - Quiet breakouts, low range risk, long price records, volume-weighted rankings, return persistence, retaining assets until exits and volatility targeting did not give a stable wider-universe advantage. - Walk-forward histogram gradient boosting used purged labels and seed 8 with price, volume, range and history features; the strongest initial result was near $114. Causal beta/alpha regression, OLMAR, constant rebalancing and mixtures of up to 972 virtual breakout experts also failed to improve the simple core reliably. - Four-hour channels and EMA trends peaked near $597; crash rebounds and intraday RSI/EMA pullbacks were weak. - A mature-coin volume-shock supplement peaked near $3,589 combined, with high turnover. Wide first-days price-age windows increased drawdown; tight ratcheted exits and low-close warnings removed the strongest winners. - Volume-flow/close-location accumulation over 7/21/42 bars and trailing-volume universes was a separate hypothesis. Its best visible result was approximately $126, with over 80% drawdown. - Hourly BTC-residual reversal following extreme negative shocks, with lagged beta/volatility and daily context, had no useful edge: its best sixteen-case result was approximately $100. - Replacing absolute liquidity with volume ranks or market-volume shares did not consistently improve both periods. Renewal after 30/90/180 quiet days broadened the opportunity set to old coins; its best combined visible balance was about $12,222, below submission 006. - Completed-cohort log-utility sizing skipped important early winners at low exploration floors. Larger floors generally repro","holdup":"The major strategy follows persistent repricing in liquid assets and waits in cash during weak BTC trends. Relative momentum distinguishes stronger assets from a shared market rise; range risk tempers selection and determines exits. The discovery stage seeks a different mechanism: the arrival of substantial venue liquidity accompanied by a strong closing record. The continuation stage adds an observed test of sustained demand before granting a longer opportunity to trade subsequent breakouts. The asset must earn membership through a completed profitable outcome that also beats its cost-inclusive BTC counterfactual; the algorithm cannot know eventual winners when buying their first qualifying signal. All later selections follow the same rules from new supplied observations. Positive evidence includes profitable earlier combined performance, positive fresh-account visible years, survival of doubled execution costs, an execution change that improved both tested periods, and a market-relative qualification amendment that improved both tested periods. The stronger contemporaneous and lagged peer scores were not consistently better across periods and were excluded. Negative evidence incl","result":{"dev":{"final":104398.72370252454,"max_drawdown":-0.410201467442452,"trades":187,"exposure":0.3585629740206669,"assets_traded":14,"fees_paid":2746.2128056941797},"hidden":{"final":14.20273459713914,"max_drawdown":-0.8957205462569986,"trades":145,"exposure":0.3828773427117508,"assets_traded":18,"fees_paid":9.229112246409208},"full":{"final":14792.816563173585,"max_drawdown":-0.8957205462569987,"trades":333,"exposure":0.36828339523889864,"assets_traded":28,"fees_paid":12619.806962214432}},"freq":"1d","equity":[["2021-10-17",110.6735],["2021-10-24",117.1621],["2021-10-31",115.9475],["2021-11-07",159.9479],["2021-11-14",154.4753],["2021-11-21",126.034],["2021-11-28",146.1866],["2021-12-05",146.1866],["2021-12-12",146.1866],["2021-12-19",146.1866],["2021-12-26",146.1866],["2022-01-02",146.1866],["2022-01-09",146.1866],["2022-01-16",146.1866],["2022-01-23",146.1866],["2022-01-30",146.1866],["2022-02-06",146.1866],["2022-02-13",146.1866],["2022-02-20",146.1866],["2022-02-27",179.5999],["2022-03-06",178.4068],["2022-03-13",282.1343],["2022-03-20",626.8338],["2022-03-27",624.3425],["2022-04-03",1319.8775],["2022-04-10",1325.46],["2022-04-17",1325.46],["2022-04-24",1325.46],["2022-05-01",1325.46],["2022-05-08",1325.46],["2022-05-15",1325.46],["2022-05-22",1325.46],["2022-05-29",1325.46],["2022-06-05",1224.7829],["2022-06-12",1039.4497],["2022-06-19",1039.4497],["2022-06-26",1039.4497],["2022-07-03",1039.4497],["2022-07-10",1039.4497],["2022-07-17",1039.4497],["2022-07-24",1039.4497],["2022-07-31",1039.4497],["2022-08-07",1039.4497],["2022-08-14",1039.4497],["2022-08-21",1039.4497],["2022-08-28",1039.4497],["2022-09-04",1039.4497],["2022-09-11",1039.4497],["2022-09-18",1039.4497],["2022-09-25",1039.4497],["2022-10-02",1039.4497],["2022-10-09",1039.4497],["2022-10-16",1039.4497],["2022-10-23",1039.4497],["2022-10-30",1039.4497],["2022-11-06",1027.1735],["2022-11-13",905.0345],["2022-11-20",905.0345],["2022-11-27",905.0345],["2022-12-04",905.0345],["2022-12-11",905.0345],["2022-12-18",905.0345],["2022-12-25",905.0345],["2023-01-01",905.0345],["2023-01-08",905.0345],["2023-01-15",1321.044],["2023-01-22",1339.0285],["2023-01-29",1307.9643],["2023-02-05",1255.1266],["2023-02-12",1118.4419],["2023-02-19",1131.5228],["2023-02-26",1018.4661],["2023-03-05",1018.4661],["2023-03-12",1018.4661],["2023-03-19",1109.7096],["2023-03-26",1132.6176],["2023-04-02",1173.4378],["2023-04-09",1194.626],["2023-04-16",1292.2921],["2023-04-23",1143.4011],["2023-04-30",1143.4011],["2023-05-07",1143.4011],["2023-05-14",1143.4011],["2023-05-21",1143.4011],["2023-05-28",1143.4011],["2023-06-04",1120.1852],["2023-06-11",1120.1852],["2023-06-18",1120.1852],["2023-06-25",1206.8042],["2023-07-02",1192.6391],["2023-07-09",1390.9871],["2023-07-16",1699.2447],["2023-07-23",1699.2447],["2023-07-30",1699.2447],["2023-08-06",1699.2447],["2023-08-13",1699.2447],["2023-08-20",1609.6771],["2023-08-27",1609.6771],["2023-09-03",1609.6771],["2023-09-10",1609.6771],["2023-09-17",1609.6771],["2023-09-24",1609.6771],["2023-10-01",1882.1861],["2023-10-08",1693.3781],["2023-10-15",1598.0249],["2023-10-22",1888.5545],["2023-10-29",2037.3458],["2023-11-05",2731.7052],["2023-11-12",3042.0047],["2023-11-19",3701.1423],["2023-11-26",3592.8989],["2023-12-03",3753.4821],["2023-12-10",5616.9407],["2023-12-17",5053.7159],["2023-12-24",7210.9223],["2023-12-31",7973.3468],["2024-01-07",7275.6174],["2024-01-14",7605.9323],["2024-01-21",11687.6429],["2024-01-28",12687.2254],["2024-02-04",11322.5057],["2024-02-11",12363.0614],["2024-02-18",12929.5521],["2024-02-25",19348.4503],["2024-03-03",19917.643],["2024-03-10",20658.9396],["2024-03-17",26065.2848],["2024-03-24",23712.283],["2024-03-31",41115.629],["2024-04-07",42819.4399],["2024-04-14",40117.8307],["2024-04-21",40117.8307],["2024-04-28",40117.8307],["2024-05-05",40117.8307],["2024-05-12",40117.8307],["2024-05-19",43586.9582],["2024-05-26",52497.7025],["2024-06-02",51793.6217],["2024-06-09",49877.2174],["2024-06-16",49877.2174],["2024-06-23",49877.2174],["2024-06-30",49877.2174],["2024-07-07",49877.2174],["2024-07-14",49877.2174],["2024-07-21",55019.0069],["2024-07-28",58289.7221],["2024-08-04",54362.4235],["2024-08-11",54362.4235],["2024-08-18",54362.4235],["2024-08-25",54409.6888],["2024-09-01",50337.7403],["2024-09-08",50337.7403],["2024-09-15",50337.7403],["2024-09-22",57939.3009],["2024-09-29",80887.1765],["2024-10-06",74810.9009],["2024-10-13",104154.7067],["2024-10-20",93912.9947],["2024-10-27",100079.2102],["2024-11-03",97320.7067],["2024-11-10",130942.6605],["2024-11-17",125667.9906],["2024-11-24",118934.3454],["2024-12-01",116417.9458],["2024-12-08",125616.0825],["2024-12-15",118292.3345],["2024-12-22",113734.9729],["2024-12-29",113734.9729],["2025-01-05",113734.9729],["2025-01-12",113734.9729],["2025-01-19",114144.0138],["2025-01-26",111498.2239],["2025-02-02",99502.5315],["2025-02-09",99502.5315],["2025-02-16",99502.5315],["2025-02-23",99502.5315],["2025-03-02",99502.5315],["2025-03-09",99502.5315],["2025-03-16",93318.9515],["2025-03-23",79946.6336],["2025-03-30",79946.6336],["2025-04-06",79946.6336],["2025-04-13",79946.6336],["2025-04-20",79946.6336],["2025-04-27",87258.6972],["2025-05-04",68656.6954],["2025-05-11",84568.7533],["2025-05-18",81028.1817],["2025-05-25",82839.2258],["2025-06-01",77694.3713],["2025-06-08",74041.3714],["2025-06-15",72717.768],["2025-06-22",69272.4857],["2025-06-29",48883.3161],["2025-07-06",48542.2049],["2025-07-13",54886.9525],["2025-07-20",51368.4157],["2025-07-27",53331.3871],["2025-08-03",52669.8106],["2025-08-10",46078.6582],["2025-08-17",40635.129],["2025-08-24",40143.6493],["2025-08-31",47023.8136],["2025-09-07",37110.345],["2025-09-14",37801.5519],["2025-09-21",35720.9774],["2025-09-28",32759.2034],["2025-10-05",26986.4808],["2025-10-12",23517.4641],["2025-10-19",23517.4641],["2025-10-26",23517.4641],["2025-11-02",23517.4641],["2025-11-09",23517.4641],["2025-11-16",23517.4641],["2025-11-23",23517.4641],["2025-11-30",23517.4641],["2025-12-07",23517.4641],["2025-12-14",23517.4641],["2025-12-21",23517.4641],["2025-12-28",23517.4641],["2026-01-04",23517.4641],["2026-01-11",23517.4641],["2026-01-18",22784.1207],["2026-01-25",19812.9207],["2026-02-01",17128.6234],["2026-02-08",17128.6234],["2026-02-15",17128.6234],["2026-02-22",15592.0045],["2026-03-01",27004.6432],["2026-03-08",27004.6432],["2026-03-15",27004.6432],["2026-03-22",27004.6432],["2026-03-29",27004.6432],["2026-04-05",27004.6432],["2026-04-12",27004.6432],["2026-04-19",26211.7184],["2026-04-26",19001.9074],["2026-05-03",19217.2759],["2026-05-10",20536.8396],["2026-05-17",19083.6261],["2026-05-24",19062.3395],["2026-05-31",19062.3395],["2026-06-07",19062.3395],["2026-06-14",19062.3395],["2026-06-21",20890.9008],["2026-06-28",15367.9402],["2026-07-05",15367.9402],["2026-07-12",15367.9402],["2026-07-19",15367.9402],["2026-07-26",15367.9402],["2026-08-02",15367.9402],["2026-08-09",15367.9402],["2026-08-16",15367.9402],["2026-08-23",16509.6829],["2026-08-30",17638.7998],["2026-09-06",17232.9454],["2026-09-13",18238.1497],["2026-09-20",14469.0726],["2026-09-27",15458.2429],["2026-10-04",15230.3354],["2026-10-11",14792.8166]],"code_bytes":6634},"a09":{"id":"a09","alias":"Swift","theme":"Online portfolio selection: Cover's universal portfolio, exponentiated gradient, online Newton step","started":"2026-10-11T00:28:46+00:00","rounds":[{"k":1,"minutes":36.0},{"k":2,"minutes":22.6}],"messages":[{"t":"2026-10-11T00:28:51.654Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test cost-aware online portfolio rules across separate market regimes. I’ll record the experiments and final equations in `WRITEUP.md`."},{"t":"2026-10-11T00:29:35.810Z","text":"The daily data contains 459 assets, including delisted coins. I’m starting with liquid momentum leaders and an exit rule for weakening trends. Then I’ll test whether an online allocator can improve rotation after fees without relying on a fixed coin list."},{"t":"2026-10-11T00:31:43.838Z","text":"The BTC baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first tests show that buying the strongest recent altcoin rallies performs badly, while broad reversal rules also lose after costs. I’m testing whether smoother trends and recent net performance can distinguish durable moves from short pumps."},{"t":"2026-10-11T00:34:43.691Z","text":"Four-hour momentum looks more promising: the best simple version turns $100 into about $563, though drawdown remains substantial. I’m comparing it with cost-aware online trend experts and a walk-forward model whose training labels are fully known before each fit."},{"t":"2026-10-11T00:38:04.878Z","text":"Requiring a challenger to beat the current holding’s momentum by three percentage points improved the four-hour strategy. The exact simulator now reports $100 → $984.21, with 34.7% drawdown and 284 trades. I’m submitting it while testing whether market breadth and online regime selection can improve it further."},{"t":"2026-10-11T00:40:47.009Z","text":"The online wealth mixtures have underperformed the simpler rotation rule so far; several also add turnover or stay in cash too long. Market breadth and sentiment filters have not improved it consistently. I’m testing a forecast of relative strength to separate coin selection from the harder task of predicting the whole market."},{"t":"2026-10-11T00:44:43.853Z","text":"Smoothing the momentum score and allowing a small negative score before exiting produces several results around $1,100–$1,270 with fewer trades. A short volume-surge trade during weak market regimes also looks promising, but its gain may come from only a few events. I’m checking that concentration and nearby parameters before using it."},{"t":"2026-10-11T00:46:06.057Z","text":"The combined strategy returns $1,726.01 in the exact simulator, with 38.0% drawdown and 216 trades. The volume-surge component was profitable in 93% of the nearby parameter tests, but has only 26 completed trades. I’m submitting this version and checking costs, fresh starts, and dependence on a few profitable days."},{"t":"2026-10-11T00:51:27.789Z","text":"The accepted entry finishes each visible year ahead of cash: $107.51, $288.40 and $556.65 from separate $100 starts. Doubling fees leaves $1,390.71 over all three years. The best ten four-hour bars account for 36% of its log growth, so a few large moves still matter. I’m finishing the writeup with that limitation and the full search count."},{"t":"2026-10-11T00:56:44.269Z","text":"The writeup is complete. Separate changes of about 20% to each parameter leave the result between $1,011 and $2,174, so I’ve retained the submitted parameters. An extra four-hour fill delay cuts the result to $971; I’ve included that timing sensitivity in the writeup. The final causality check and comparison chart are finishing."},{"t":"2026-10-11T01:01:50.434Z","text":"The final causality check passed, and the comparison chart is saved. The later volume-weighted scores, regime selector and machine-learning veto all underperformed, so submission #003 remains the entry. I’m making one last timing comparison: hourly decisions using the same signal horizons, to see whether overlapping four-hour volume observations improve surge entries."},{"t":"2026-10-11T01:04:48.036Z","text":"Entry: accepted #003, strategy.py; WRITEUP.md complete.  \nVisible years: $100 → **$1,726.01** after costs; max drawdown **−38.0%**; causality passed.  \nNext idea: learn when volume surges persist while controlling turnover."},{"t":"2026-10-11T01:04:55.234Z","text":"I’ll use the remaining time to test whether volume surges can support longer profitable holds. I’ll keep submission #003 in place unless a new version improves enough to justify the added complexity, then update the writeup and verify the final accepted file."},{"t":"2026-10-11T01:07:13.417Z","text":"The accepted strategy is unchanged: $1,726.01 from $100, with a 38% maximum drawdown. Longer surge holds added little in the tests so far. I’m now testing whether a switching threshold based on return uncertainty can improve the growth race’s fixed 3% threshold."},{"t":"2026-10-11T01:08:21.387Z","text":"The uncertainty-scaled thresholds look promising: the best tested version reached $2,316.77, improved all three visible annual periods, and reduced drawdown to 34.2%. Nearby settings were also profitable. I’ll run the exact referee and causality checks, then stress-test it before replacing the accepted entry."},{"t":"2026-10-11T01:09:44.202Z","text":"The exact referee and causality checks passed. The candidate returned $2,316.77 with 34.16% drawdown; doubling costs still left $1,889.24. Every parameter tested separately at about ±20% remained profitable, though some reduced wealth substantially. I’m submitting this version and updating the writeup."},{"t":"2026-10-11T01:11:21.213Z","text":"Submission #004 is accepted and is now the entry, ranked #13. Its annual visible multiples were 1.17×, 3.36× and 5.88×. I’m updating the equations and figure to match the submitted file and documenting the search’s selection bias and sensitivity to execution timing."},{"t":"2026-10-11T01:13:21.814Z","text":"The writeup and figure now match submission #004. I’m running one last walk-forward test: whether a model trained only on completed past trades can improve the asset ranking. The accepted strategy will stay in place unless the improvement is substantial across the visible periods."},{"t":"2026-10-11T01:15:34.003Z","text":"The learned ranking did worse than the accepted rule in all 24 tested variants; its best result was $2,278.46. I’m keeping submission #004 and recording that failure. The local strategy file also matches the accepted referee snapshot byte for byte."},{"t":"2026-10-11T01:17:47.726Z","text":"I’m testing one final alternative before freezing the entry: holding liquid new listings after a strong first few daily candles. The earlier listing tests rotated frequently; this version tests whether longer holds can capture early trends after costs. Selection uses observed listing age, price and volume only."},{"t":"2026-10-11T01:21:14.274Z","text":"The listing rule alone was weaker, but using it only during negative BTC trends produced a promising daily combination: about $9,941 in the preliminary simulation. Nearby liquidity, holding-time and stop settings also beat $2,316.77, though drawdown rose to roughly 61%. I’m checking calendar-based listing age and running the exact referee before deciding."},{"t":"2026-10-11T01:23:21.789Z","text":"Submission #005 is accepted and is now the entry, ranked #2 at submission time: $9,941.46 from $100. I’m choosing it for the growth objective despite its 60.56% drawdown and concentration in a few listing trades. The final writeup will document those risks and the sharp loss of performance when fills are delayed."},{"t":"2026-10-11T01:27:22.113Z","text":"Final accepted entry #005 combines major-asset growth rotation during positive BTC trends with liquid new-listing continuation during negative trends. Visible result: **$100 → $9,941.46 (99.41×)**, **60.56% maximum drawdown**, ranked **#2 when accepted**. Causality checks passed, and WRITEUP.md contains the equations, search counts, results and failures. Returns depend heavily on a few listing trades; hidden performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T00:30:03.060105+00:00","note":"Initial causal BTC 200-day trend baseline; online allocation research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:38:06.099470+00:00","note":"Four-hour major-coin momentum rotation, BTC 50-day market gate, and 3% switching hysteresis; full targets support fresh starts.","status":"ok","dev":984.2093517139671,"dd":-0.34708001923792964,"reason":""},{"n":"003","t":"2026-10-11T00:46:07.362731+00:00","note":"Discounted log-growth rotation with 1% entry/exit deadband and 3% switching margin; 48-hour volume-confirmed surges during weak daily regimes.","status":"ok","dev":1726.013692640102,"dd":-0.3796056411250407,"reason":""},{"n":"004","t":"2026-10-11T01:09:45.631061+00:00","note":"Swift: uncertainty-scaled discounted growth race with correlated switching barrier; volume-confirmed short surge mode. Exact causal checks and cost stress passed.","status":"ok","dev":2316.774579384113,"dd":-0.34158741175377894,"reason":""},{"n":"005","t":"2026-10-11T01:22:15.228878+00:00","note":"Swift final: daily uncertainty-scaled major growth during positive BTC trend; early liquid listing continuation during negative trend, generic age/volume/candle rules. Exact referee and prefix tests p","status":"ok","dev":9941.460292568889,"dd":-0.6056014436024804,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #003, strategy.py; WRITEUP.md complete.  \nVisible years: $100 → **$1,726.01** after costs; max drawdown **−38.0%**; causality passed.  \nNext idea: learn when volume surges persist while controlling turnover."},{"k":2,"text":"Final accepted entry #005 combines major-asset growth rotation during positive BTC trends with liquid new-listing continuation during negative trends. Visible result: **$100 → $9,941.46 (99.41×)**, **60.56% maximum drawdown**, ranked **#2 when accepted**. Causality checks passed, and WRITEUP.md contains the equations, search counts, results and failures. Returns depend heavily on a few listing trades; hidden performance remains unknown."}],"title":"Swift (a09): uncertainty-scaled growth with liquid listing continuation","idea":"The latest accepted entry is strategy.py, submission #005, FREQ = \"1d\". Its SHA-256 is e291c24126a0788add4a887aadf15e8fc4bd74f94d2160164170e90a4d9b8b18. It is one self-contained deterministic file using numpy, pandas and numba. The official visible result is $100 to $9,941.46, with 60.56% maximum drawdown. It ranked second when accepted. Hidden performance is unknown. The strategy holds one asset at full weight or cash. When BTC's daily trend is positive, BTC, ETH and SOL compete on discounted log growth, with an individual price filter and uncertainty-scaled entry and switching barriers. When BTC's daily trend is negative, the strategy normally holds cash, but may buy a newly observed listing with substantial quote volume, a positive daily candle and a close near that candle's high. It holds that listing briefly, subject to a timer, price filter and close-based trailing stop. This final change arose from a complementary result: mature-asset trend trading and early listing continuation worked in different market states. The listing strategy alone was weaker. Restricting it to negative BTC trends and combining it with the growth strategy improved the full visible result materially. The entry takes more risk than submission #004. I chose it because the arena's objective rewards compounded wealth, while explicitly reporting its drawdown, sparse-event dependence and execution sensi","failed":"- Broad volume-ranked daily momentum often bought extended altcoin moves before reversals. Smooth, clipped, skipped-period and volatility-adjusted variants did not provide a superior robust entry. - Reversal and trend-conditioned dip strategies suffered adverse moves and turnover. The initial preliminary simulator needed a delisting correction; its early faulty screens were not treated as final evidence. - An initial event-only breakout prototype could retain an internal incumbent after trading disappeared. It was rejected. The final entry clears unavailable prices and was scored with the exact engine. - Daily listing ranks that continuously rotated did poorly. Persistent holds improved the family, but the standalone listing rule remained weaker and had large drawdowns. The final complement uses a narrower age window and negative market regime. - Soft virtual-expert portfolios diluted exposure or adapted slowly. Concentrated and meta-expert selectors improved some results but did not beat the selected direct growth mechanism. - Purged walk-forward pooled regressions using absolute or market-relative labels had inconsistent annual results. An intraday forecast veto reduced wealth. A final relative-growth forecast blend tested 24 cases and all underperformed the then accepted $2,316.77 rule. - Fear & Greed, breadth, volume-weighted growth and more elaborate major-asset surge exits added complexity without stable improvement. They are absent from the final entry. - A matched-horizon hourly major-asset version returned about $686 with 64% drawdown. Tight main-mode stops and several faster variants also reduced wealth. - The major-only four-hour entry was more conservative, with 34% drawdown, but delivered materially less visible wealth. Its results are retained separately r","holdup":"The mechanisms are observable before a trade: a slow market trend, recent log growth, asset-level price persistence, relative-return uncertainty, and immediate price/volume evidence when a new tradable asset appears. Selection never encodes historical altcoin winners. Every new asset faces the same age, liquidity and candle conditions. The growth barrier addresses a generic friction problem: small estimated differences between assets are noisy, while switching has a definite cost. Relative volatility accounts for their shared market movement. The listing mode targets a different source of demand: early trading can concentrate attention and liquidity even when the older market is weak. Requiring a positive candle near its high and meaningful quote volume seeks sustained demand. The short eligibility window and maximum holding duration keep this a launch-continuation rule rather than indefinite ownership of unfamiliar assets. Both mechanisms can operate on new observations with identical parameters. Every rolling statistic, availability-age update and state transition has the prefix property. There are no date-specific behaviors, future prices, embedded historical tables, external ca","result":{"dev":{"final":9941.460292568889,"max_drawdown":-0.6056014436024804,"trades":158,"exposure":0.47536496350364965,"assets_traded":15,"fees_paid":436.27090606623375},"hidden":{"final":74.12456655938789,"max_drawdown":-0.6861230273812947,"trades":120,"exposure":0.5041095890410959,"assets_traded":9,"fees_paid":14.030995278561049},"full":{"final":7369.06435154035,"max_drawdown":-0.6861230273812948,"trades":278,"exposure":0.4868565169769989,"assets_traded":21,"fees_paid":1831.1567303365966}},"freq":"1d","equity":[["2021-10-17",110.6735],["2021-10-24",117.1621],["2021-10-31",115.9475],["2021-11-07",153.1194],["2021-11-14",134.473],["2021-11-21",177.2441],["2021-11-28",160.9943],["2021-12-05",109.4962],["2021-12-12",109.4962],["2021-12-19",109.4962],["2021-12-26",109.4962],["2022-01-02",111.6783],["2022-01-09",111.6783],["2022-01-16",111.6783],["2022-01-23",111.6783],["2022-01-30",111.6783],["2022-02-06",111.6783],["2022-02-13",103.9215],["2022-02-20",98.4648],["2022-02-27",109.7606],["2022-03-06",123.0747],["2022-03-13",204.892],["2022-03-20",423.3451],["2022-03-27",442.8268],["2022-04-03",576.724],["2022-04-10",491.1319],["2022-04-17",491.1319],["2022-04-24",491.1319],["2022-05-01",491.1319],["2022-05-08",491.1319],["2022-05-15",491.1319],["2022-05-22",491.1319],["2022-05-29",491.1319],["2022-06-05",441.4235],["2022-06-12",349.9168],["2022-06-19",349.9168],["2022-06-26",349.9168],["2022-07-03",349.9168],["2022-07-10",349.9168],["2022-07-17",349.9168],["2022-07-24",351.0042],["2022-07-31",327.0735],["2022-08-07",325.84],["2022-08-14",382.3021],["2022-08-21",261.1651],["2022-08-28",340.6613],["2022-09-04",340.6613],["2022-09-11",340.6613],["2022-09-18",299.6624],["2022-09-25",299.6624],["2022-10-02",299.6624],["2022-10-09",296.4233],["2022-10-16",296.4233],["2022-10-23",296.4233],["2022-10-30",328.5467],["2022-11-06",330.054],["2022-11-13",264.1439],["2022-11-20",264.1439],["2022-11-27",264.1439],["2022-12-04",264.1439],["2022-12-11",264.1439],["2022-12-18",264.1439],["2022-12-25",264.1439],["2023-01-01",264.1439],["2023-01-08",257.396],["2023-01-15",476.2808],["2023-01-22",482.7648],["2023-01-29",471.5651],["2023-02-05",478.2456],["2023-02-12",425.0088],["2023-02-19",429.9796],["2023-02-26",392.4927],["2023-03-05",369.1783],["2023-03-12",369.1783],["2023-03-19",411.5638],["2023-03-26",420.0599],["2023-04-02",435.1991],["2023-04-09",424.9885],["2023-04-16",466.2044],["2023-04-23",438.8203],["2023-04-30",432.3185],["2023-05-07",388.5172],["2023-05-14",388.1287],["2023-05-21",388.1287],["2023-05-28",388.1287],["2023-06-04",388.1287],["2023-06-11",388.1287],["2023-06-18",388.1287],["2023-06-25",418.141],["2023-07-02",432.3963],["2023-07-09",504.3082],["2023-07-16",634.952],["2023-07-23",566.9709],["2023-07-30",565.6134],["2023-08-06",544.3945],["2023-08-13",544.3945],["2023-08-20",544.3945],["2023-08-27",544.3945],["2023-09-03",544.3945],["2023-09-10",544.3945],["2023-09-17",544.3945],["2023-09-24",550.6341],["2023-10-01",761.7165],["2023-10-08",830.5098],["2023-10-15",786.593],["2023-10-22",961.9986],["2023-10-29",1037.7904],["2023-11-05",1391.4857],["2023-11-12",1848.5334],["2023-11-19",1921.3724],["2023-11-26",1728.5896],["2023-12-03",1766.9628],["2023-12-10",1874.0024],["2023-12-17",1910.8085],["2023-12-24",2796.2292],["2023-12-31",2641.2288],["2024-01-07",2412.9884],["2024-01-14",2549.1529],["2024-01-21",2383.7941],["2024-01-28",2977.8552],["2024-02-04",3128.6301],["2024-02-11",3285.3908],["2024-02-18",3419.8512],["2024-02-25",3673.5722],["2024-03-03",4152.2502],["2024-03-10",4163.032],["2024-03-17",5252.4774],["2024-03-24",4981.2701],["2024-03-31",5603.0273],["2024-04-07",5232.237],["2024-04-14",4715.237],["2024-04-21",4715.237],["2024-04-28",4715.237],["2024-05-05",4715.237],["2024-05-12",4715.237],["2024-05-19",5122.9798],["2024-05-26",5227.0416],["2024-06-02",5319.8364],["2024-06-09",5122.9983],["2024-06-16",5122.9983],["2024-06-23",4684.181],["2024-06-30",5422.4293],["2024-07-07",5422.4293],["2024-07-14",5422.4293],["2024-07-21",5900.6924],["2024-07-28",6251.4708],["2024-08-04",5830.2749],["2024-08-11",6242.3039],["2024-08-18",6156.7143],["2024-08-25",6162.0673],["2024-09-01",5612.4888],["2024-09-08",5612.4888],["2024-09-15",5560.0829],["2024-09-22",11112.2325],["2024-09-29",10748.9235],["2024-10-06",9941.4603],["2024-10-13",9941.4603],["2024-10-20",10238.529],["2024-10-27",10530.2533],["2024-11-03",10313.0863],["2024-11-10",10930.9586],["2024-11-17",11635.2794],["2024-11-24",12363.2816],["2024-12-01",12295.9024],["2024-12-08",13267.3969],["2024-12-15",12335.0463],["2024-12-22",12174.9663],["2024-12-29",12174.9663],["2025-01-05",12651.5344],["2025-01-12",11793.063],["2025-01-19",14992.4335],["2025-01-26",14674.2371],["2025-02-02",12486.5241],["2025-02-09",12474.0375],["2025-02-16",12474.0375],["2025-02-23",12474.0375],["2025-03-02",12474.0375],["2025-03-09",12474.0375],["2025-03-16",11441.084],["2025-03-23",9507.0699],["2025-03-30",9507.0699],["2025-04-06",9507.0699],["2025-04-13",9603.2429],["2025-04-20",9353.9523],["2025-04-27",9978.5745],["2025-05-04",9958.368],["2025-05-11",12266.3458],["2025-05-18",11752.8006],["2025-05-25",12015.485],["2025-06-01",12007.9761],["2025-06-08",11095.0968],["2025-06-15",10655.8962],["2025-06-22",10655.8962],["2025-06-29",10767.0528],["2025-07-06",10229.9158],["2025-07-13",11965.2627],["2025-07-20",14602.5329],["2025-07-27",13610.0621],["2025-08-03",12677.3623],["2025-08-10",14501.373],["2025-08-17",15050.6091],["2025-08-24",13690.0943],["2025-08-31",16672.6545],["2025-09-07",12199.0771],["2025-09-14",12914.1715],["2025-09-21",12759.6576],["2025-09-28",14923.5084],["2025-10-05",11176.7737],["2025-10-12",9715.225],["2025-10-19",9715.225],["2025-10-26",9715.225],["2025-11-02",9606.5141],["2025-11-09",9606.5141],["2025-11-16",9606.5141],["2025-11-23",9606.5141],["2025-11-30",9606.5141],["2025-12-07",9606.5141],["2025-12-14",9606.5141],["2025-12-21",9606.5141],["2025-12-28",9606.5141],["2026-01-04",9675.9968],["2026-01-11",9855.1953],["2026-01-18",10285.9149],["2026-01-25",8998.3677],["2026-02-01",8186.893],["2026-02-08",8186.893],["2026-02-15",8186.893],["2026-02-22",8186.893],["2026-03-01",8186.893],["2026-03-08",8186.893],["2026-03-15",8186.893],["2026-03-22",7720.0099],["2026-03-29",7290.0341],["2026-04-05",7290.0341],["2026-04-12",7886.1011],["2026-04-19",8111.5721],["2026-04-26",7978.9449],["2026-05-03",7819.0218],["2026-05-10",8355.9188],["2026-05-17",7997.9924],["2026-05-24",7997.9924],["2026-05-31",7997.9924],["2026-06-07",7997.9924],["2026-06-14",7997.9924],["2026-06-21",9020.3015],["2026-06-28",5932.5337],["2026-07-05",5932.5337],["2026-07-12",5932.5337],["2026-07-19",5895.2742],["2026-07-26",5934.1411],["2026-08-02",5889.3473],["2026-08-09",5907.5967],["2026-08-16",5841.8915],["2026-08-23",7388.4311],["2026-08-30",8061.494],["2026-09-06",7876.0056],["2026-09-13",7545.0438],["2026-09-20",7044.0887],["2026-09-27",7700.5475],["2026-10-04",7587.015],["2026-10-11",7369.0644]],"code_bytes":4150},"a10":{"id":"a10","alias":"Tern","theme":"Passive-aggressive and moving-average mean-reversion portfolios (PAMR, OLMAR, CWMR-style)","started":"2026-10-11T00:29:01+00:00","rounds":[{"k":1,"minutes":37.9},{"k":2,"minutes":12.8}],"messages":[{"t":"2026-10-11T00:29:08.354Z","text":"I’ll inspect the data and simulator, submit a baseline early, then test mean reversion and momentum ideas with attention to costs and robustness. I’ll keep a record of the experiments for the writeup."},{"t":"2026-10-11T00:30:13.219Z","text":"The BTC trend baseline is awaiting the referee. The first screen found that daily rotation into recent winners or losers loses heavily after costs. I’ll test whether liquidity, coin trends, and market conditions make short-term reversals more useful."},{"t":"2026-10-11T00:32:19.500Z","text":"The baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. Trend filters improve the rotation strategies, but their gains depend heavily on one period. I’ll test slower holding rules and avoid choosing a strategy on total profit alone."},{"t":"2026-10-11T00:35:20.920Z","text":"Standard reversal and breakout portfolios haven’t shown enough stability to replace the baseline. I’m testing a pooled return model that uses price and volume features without coin identities, retrains only on outcomes already known, and keeps cash available when signals are weak."},{"t":"2026-10-11T00:38:32.802Z","text":"Rotation among BTC, ETH, and SOL looks more promising: roughly 10× growth in the visible period, with gains in each of the three sub-periods tested. Similar hourly and four-hour versions also perform well. I’m checking it with the referee simulator before submitting it."},{"t":"2026-10-11T00:43:01.307Z","text":"The four-hour strategy is accepted: $100 became $1,000.07, with a 34.5% maximum drawdown. Unconditional mean reversion still looks weak; a small reversal adjustment within established uptrends looks more useful. I’ll test whether it improves nearby parameter settings and whether combining rules reduces dependence on one setting."},{"t":"2026-10-11T00:49:21.225Z","text":"The blended rule passed the causal checks: it grew $100 to $1,402.90 with a 34.0% maximum drawdown, and is awaiting submission acceptance. Hysteresis cut trading costs substantially. I’ll check whether a limited panic-rebound rule can add returns while the trend portfolio holds cash."},{"t":"2026-10-11T00:54:57.123Z","text":"The panic-rebound rule passed the causal checks and raised the visible result to $2,659.55, with maximum drawdown still about 34%. It requires severe fear, a sharp drop, and a positive four-hour bar before entering. I’ve submitted it and will now check doubled costs, nearby parameters, and dependence on the best days."},{"t":"2026-10-11T01:00:28.278Z","text":"The accepted rule still grows $100 to $2,056.84 with doubled costs. Nearby parameters stay profitable, though the ten best days account for about 52% of log growth. One final check improved the result by letting the reversal sleeve hold cash when its combined score is negative; I’ll verify that version and update the writeup."},{"t":"2026-10-11T01:06:29.251Z","text":"Entry #005 is accepted: $100 became $2,947.30, with a 34.0% maximum drawdown. It still reaches $2,281.47 with doubled costs. The final volatility-scaled and BTC-relative variants offered no consistent improvement, so I’m keeping this entry. The writeup covers the equations, search, results, robustness checks, and failed ideas."},{"t":"2026-10-11T01:06:56.326Z","text":"Entry: #005 accepted; strategy and writeup complete.  \nVisible years: **$100 → $2,947.30 (29.47×)**; maximum drawdown **34.0%**.  \nNext idea: walk-forward calibration of panic-rebound exits."},{"t":"2026-10-11T01:07:03.157Z","text":"I’ll test whether panic exits can adapt using only completed past trades, then compare that with the accepted strategy’s robustness. I’ll leave time to submit and verify any final change before 01:29, and update the writeup to match the judged entry."},{"t":"2026-10-11T01:09:25.495Z","text":"The accepted entry and its submission snapshot match. Its main weakness is that most of the panic-reversal gains come from one market regime. I’m testing whether exits learned from completed trades—and a change in bar frequency—improve results across periods."},{"t":"2026-10-11T01:11:21.261Z","text":"Learning exit horizons from completed trades did not help: the best adaptive blend reached $2,727 versus $2,947 for the accepted entry. Earlier profit-taking also reduced returns. Hourly panic entries had substantially larger drawdowns; I’m checking whether waiting for a four-hour rebound fixes that weakness."},{"t":"2026-10-11T01:13:24.563Z","text":"The longer rebound-confirmation windows did not improve results. Using only the reversal sleeve reached $3,244, but increased drawdown and weakened the latest visible segment. I’m keeping the equal blend as the reference and testing a broader market gate plus small price perturbations before the final choice."},{"t":"2026-10-11T01:16:04.467Z","text":"I’ll retain accepted entry #005. The broader market gates and volatility sizing did not improve it, and all 192 daily altcoin variants lost money. Small price perturbations kept the reference profitable, but changed final wealth materially. I’m recording those limits, the complete trial count, and the final submission check in the writeup."},{"t":"2026-10-11T01:19:41.332Z","text":"Retained accepted entry **#005**: strategy blends momentum and moving-average reversal on BTC, ETH and SOL, with confirmed one-day panic rebounds. Visible results remain **$100 → $2,947.30 (29.47×)** with **34.0% maximum drawdown**; doubled costs leave $2,281.47. Final causal checks passed, and the source matches the accepted snapshot. WRITEUP.md is complete with equations, trial counts, results, failed ideas and robustness limits. Hidden performance remains untested."}],"submissions":[{"n":"001","t":"2026-10-11T00:29:51.903462+00:00","note":"Initial causal baseline: BTC 200-day trend, trade only signal changes.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:39:42.591764+00:00","note":"Causal 4h BTC/ETH/SOL momentum rotation; lagged daily trend eligibility and BTC 20-day EMA risk gate.","status":"ok","dev":1000.0662243637282,"dd":-0.3451661136333837,"reason":""},{"n":"003","t":"2026-10-11T00:48:00.807758+00:00","note":"Two-sleeve 4h trend-anchored OLMAR: medium trend, 3-day reversal tilt, passive replacement threshold, and fee-sized BTC hysteresis.","status":"ok","dev":1402.8971986254423,"dd":-0.33992321704177564,"reason":""},{"n":"004","t":"2026-10-11T00:54:20.530908+00:00","note":"Trend-anchored OLMAR plus confirmed panic reversals while otherwise in cash; bounded holding period and one-day cooldown.","status":"ok","dev":2659.551203234806,"dd":-0.33992321704177575,"reason":""},{"n":"005","t":"2026-10-11T01:01:17.124926+00:00","note":"Finalized trend-anchored OLMAR: positive combined forecast, confirmed panic reversals, one-day timeout/cooldown; causal and doubled-cost checks passed.","status":"ok","dev":2947.3041026168016,"dd":-0.33992321704177586,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: #005 accepted; strategy and writeup complete.  \nVisible years: **$100 → $2,947.30 (29.47×)**; maximum drawdown **34.0%**.  \nNext idea: walk-forward calibration of panic-rebound exits."},{"k":2,"text":"Retained accepted entry **#005**: strategy blends momentum and moving-average reversal on BTC, ETH and SOL, with confirmed one-day panic rebounds. Visible results remain **$100 → $2,947.30 (29.47×)** with **34.0% maximum drawdown**; doubled costs leave $2,281.47. Final causal checks passed, and the source matches the accepted snapshot. WRITEUP.md is complete with equations, trial counts, results, failed ideas and robustness limits. Hidden performance remains untested."}],"title":"Tern (a10): Trend-anchored OLMAR and confirmed panic reversal","idea":"Unconditional short-term loser rotation was a poor way to trade this universe after costs: it repeatedly bought persistent declines, while extreme recent winners often reversed. My entry retains moving-average mean reversion as a modest preference within established uptrends. It trades only BTC, ETH and SOL, the three explicitly permitted named assets with the lower 0.10% trading cost. Two equal sleeves select a coin using either medium-term momentum or medium-term momentum plus an OLMAR-style price forecast. A BTC trend state determines whether to invest, a daily coin trend determines eligibility, and a replacement threshold makes the selection passive when competing forecasts are close. When the trend portfolio holds cash, a separate, short-lived reversal trade can respond to a fear-driven selloff after a positive four-hour bar confirms a bounce. Otherwise the account holds cash. This is a specific combination of familiar components, rather than a claim to have discovered a new statistical law.","failed":"- Daily extreme-loser rotation and OLMAR-style argmax portfolios generally lost money after costs, even with trend eligibility and liquidity guards. Persistent declines and costly rotation overwhelmed occasional rebounds. - Daily recent-winner rotation could look spectacular in the earlier bull data and then collapse in the visible bear/recovery data. One seven-day, top-liquidity variant returned about 8x overall but had roughly 90% drawdown and severe segment dependence; I rejected it. - Volume-shock breakouts could generate about 12x in the best screened case, but neighboring choices and segments were weaker, with much larger drawdowns than the liquid intraday family. - Longer-horizon, volatility-adjusted altcoin momentum and rankings based on a trend advisor's virtual wealth did not provide a persuasive improvement. - The pooled regression's best allocation screen was about 5.3x with lower drawdown; its computation and additional modeling choices were not rewarded by enough growth. - Large ensembles and recent-performance weighting reduced growth relative to the simpler two-sleeve rule. Their earlier-period winners did not consistently predict the best later configuration. - Finer 15m and 5m decisions increased churn and did not improve this strategy. - Panic buying without a confirming positive bar, smaller selloff thresholds, or longer holding periods produced weaker results or substantially deeper drawdowns. Full commitment after arbitrary small dips was rejected. - A first panic prototype could exit and re-enter on the same signal bar, undermining its intended timeout and stop. The final version adds an explicit cooldown and was re-evaluated; only the corrected version is submitted. - A final 36-case volatility-scaled panic-entry screen used 10/20/40-day realized","holdup":"The economic hypothesis is that attention and capital migration persist over weeks, while shorter price overshoots can partially reverse. Conditioning the routine reversal tilt on a longer uptrend prevents many purchases of persistent losers. The separate panic entry instead requires contemporaneous fear, a large mean-price gap and an observed positive bar, then limits its forecast horizon to one day and imposes a one-day cooldown. A forced-selling rebound is the hypothesis; the sample of 27 episodes is too small to establish it reliably. Low trading costs, liquid majors, slow score replacement, and cash during weak market states address three failures visible in the research: transaction drag, chasing extended rallies, and staying invested through prolonged declines. These mechanisms require no particular calendar date or future coin identity. The trend-only family behaved reasonably across hourly and four-hour variants. The simplest hourly trend rule performed better than the four-hour blend in the earlier screen, while 15m and 5m variants had more churn and weaker results. The complete panic-augmented rule was substantially more sensitive to frequency: rescaling its windows and ","result":{"dev":{"final":2947.3041026168016,"max_drawdown":-0.33992321704177586,"trades":672,"exposure":0.43877672775557836,"assets_traded":3,"fees_paid":266.18251533737805},"hidden":{"final":222.6132499392448,"max_drawdown":-0.3064515200666419,"trades":448,"exposure":0.47060737400943264,"assets_traded":3,"fees_paid":26.475324254547353},"full":{"final":6561.089448427968,"max_drawdown":-0.33992321704177586,"trades":1120,"exposure":0.4515020134978092,"assets_traded":3,"fees_paid":1046.4908332727541}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",111.1964],["2021-10-24",113.0874],["2021-10-31",109.2902],["2021-11-07",137.7912],["2021-11-14",131.4637],["2021-11-21",125.9049],["2021-11-28",125.9049],["2021-12-05",125.0928],["2021-12-12",125.2641],["2021-12-19",129.8871],["2021-12-26",129.8871],["2022-01-02",129.8871],["2022-01-09",119.5398],["2022-01-16",119.5398],["2022-01-23",120.2429],["2022-01-30",127.5062],["2022-02-06",127.5062],["2022-02-13",120.564],["2022-02-20",115.5944],["2022-02-27",135.5695],["2022-03-06",124.4054],["2022-03-13",120.2786],["2022-03-20",123.8842],["2022-03-27",135.9755],["2022-04-03",176.8215],["2022-04-10",153.5864],["2022-04-17",153.5864],["2022-04-24",148.231],["2022-05-01",148.231],["2022-05-08",148.231],["2022-05-15",164.1963],["2022-05-22",164.1963],["2022-05-29",163.9076],["2022-06-05",166.2179],["2022-06-12",158.7031],["2022-06-19",211.8201],["2022-06-26",211.8201],["2022-07-03",196.7911],["2022-07-10",196.7911],["2022-07-17",190.1928],["2022-07-24",214.1957],["2022-07-31",210.7107],["2022-08-07",200.5392],["2022-08-14",226.7116],["2022-08-21",211.6042],["2022-08-28",208.1333],["2022-09-04",208.1333],["2022-09-11",214.7472],["2022-09-18",183.1048],["2022-09-25",190.6437],["2022-10-02",190.6437],["2022-10-09",183.7021],["2022-10-16",183.7021],["2022-10-23",183.7021],["2022-10-30",193.7204],["2022-11-06",189.7121],["2022-11-13",218.3127],["2022-11-20",213.8852],["2022-11-27",213.8852],["2022-12-04",213.8852],["2022-12-11",213.8852],["2022-12-18",217.4624],["2022-12-25",217.4624],["2023-01-01",217.4624],["2023-01-08",219.3175],["2023-01-15",380.2493],["2023-01-22",405.3171],["2023-01-29",419.9614],["2023-02-05",387.8593],["2023-02-12",372.0421],["2023-02-19",389.5826],["2023-02-26",354.027],["2023-03-05",327.6028],["2023-03-12",327.6028],["2023-03-19",379.0539],["2023-03-26",373.3133],["2023-04-02",374.8681],["2023-04-09",378.518],["2023-04-16",400.607],["2023-04-23",377.7561],["2023-04-30",375.4588],["2023-05-07",358.3869],["2023-05-14",355.2517],["2023-05-21",355.2517],["2023-05-28",355.2517],["2023-06-04",347.812],["2023-06-11",345.3489],["2023-06-18",345.3489],["2023-06-25",361.2995],["2023-07-02",366.8731],["2023-07-09",407.2698],["2023-07-16",523.5779],["2023-07-23",500.3129],["2023-07-30",500.3129],["2023-08-06",500.3129],["2023-08-13",509.1367],["2023-08-20",502.959],["2023-08-27",502.959],["2023-09-03",502.959],["2023-09-10",502.959],["2023-09-17",502.959],["2023-09-24",502.9781],["2023-10-01",534.05],["2023-10-08",548.3885],["2023-10-15",516.1391],["2023-10-22",637.5159],["2023-10-29",733.2877],["2023-11-05",905.277],["2023-11-12",1245.0336],["2023-11-19",1359.6932],["2023-11-26",1271.9722],["2023-12-03",1383.3075],["2023-12-10",1484.1277],["2023-12-17",1485.5418],["2023-12-24",2326.2849],["2023-12-31",1969.0595],["2024-01-07",1724.9395],["2024-01-14",1752.6535],["2024-01-21",1752.6535],["2024-01-28",1732.2027],["2024-02-04",1757.2964],["2024-02-11",1930.9842],["2024-02-18",2016.5355],["2024-02-25",2173.6106],["2024-03-03",2444.1834],["2024-03-10",2705.0573],["2024-03-17",3569.8255],["2024-03-24",3185.3732],["2024-03-31",3594.2027],["2024-04-07",3211.7014],["2024-04-14",3098.3244],["2024-04-21",3098.3244],["2024-04-28",3098.3244],["2024-05-05",3098.3244],["2024-05-12",3098.3244],["2024-05-19",3299.7982],["2024-05-26",3420.8901],["2024-06-02",3335.7925],["2024-06-09",3291.3753],["2024-06-16",3146.2782],["2024-06-23",3146.2782],["2024-06-30",3146.2782],["2024-07-07",3146.2782],["2024-07-14",3146.2782],["2024-07-21",3565.6948],["2024-07-28",3701.9277],["2024-08-04",3436.4399],["2024-08-11",3488.4919],["2024-08-18",3488.4919],["2024-08-25",3503.0433],["2024-09-01",3242.3548],["2024-09-08",3242.3548],["2024-09-15",3224.1149],["2024-09-22",3177.1084],["2024-09-29",3367.6285],["2024-10-06",3057.6003],["2024-10-13",2950.4587],["2024-10-20",3104.3617],["2024-10-27",3264.2985],["2024-11-03",2966.8073],["2024-11-10",3681.3193],["2024-11-17",3768.3744],["2024-11-24",3953.2572],["2024-12-01",3932.7324],["2024-12-08",4238.1905],["2024-12-15",4146.947],["2024-12-22",3712.506],["2024-12-29",3684.5551],["2025-01-05",3716.3916],["2025-01-12",3586.9286],["2025-01-19",4063.3185],["2025-01-26",3896.1831],["2025-02-02",3212.055],["2025-02-09",3212.055],["2025-02-16",3212.055],["2025-02-23",3212.055],["2025-03-02",3294.0494],["2025-03-09",3294.0494],["2025-03-16",3115.031],["2025-03-23",3115.031],["2025-03-30",3115.031],["2025-04-06",3115.031],["2025-04-13",3434.2464],["2025-04-20",3463.2857],["2025-04-27",3787.1484],["2025-05-04",3731.6147],["2025-05-11",4710.6727],["2025-05-18",4523.1864],["2025-05-25",4739.6402],["2025-06-01",4711.9164],["2025-06-08",4710.2027],["2025-06-15",4524.7897],["2025-06-22",4390.1079],["2025-06-29",4449.9871],["2025-07-06",4389.1699],["2025-07-13",4906.4955],["2025-07-20",6154.6487],["2025-07-27",6283.3079],["2025-08-03",6063.9876],["2025-08-10",6526.1609],["2025-08-17",6583.4756],["2025-08-24",6177.8041],["2025-08-31",6177.8041],["2025-09-07",6037.3289],["2025-09-14",6507.7836],["2025-09-21",6386.4889],["2025-09-28",5948.221],["2025-10-05",6197.1878],["2025-10-12",6382.7888],["2025-10-19",6376.7316],["2025-10-26",6376.7316],["2025-11-02",6357.8037],["2025-11-09",6610.0197],["2025-11-16",6332.0106],["2025-11-23",6397.8692],["2025-11-30",6397.8692],["2025-12-07",6397.8692],["2025-12-14",6397.8692],["2025-12-21",6397.8692],["2025-12-28",6397.8692],["2026-01-04",6391.9955],["2026-01-11",6563.3715],["2026-01-18",6697.0585],["2026-01-25",6127.036],["2026-02-01",5792.3723],["2026-02-08",6227.9937],["2026-02-15",6227.9937],["2026-02-22",6227.9937],["2026-03-01",6227.9937],["2026-03-08",6227.9937],["2026-03-15",6227.9937],["2026-03-22",6182.4221],["2026-03-29",5833.0159],["2026-04-05",5833.0159],["2026-04-12",5970.1041],["2026-04-19",6245.5737],["2026-04-26",6395.8151],["2026-05-03",6439.9],["2026-05-10",6737.3418],["2026-05-17",6091.8863],["2026-05-24",6091.8863],["2026-05-31",6091.8863],["2026-06-07",6027.0346],["2026-06-14",6027.0346],["2026-06-21",6027.0346],["2026-06-28",6027.0346],["2026-07-05",5925.2331],["2026-07-12",5611.3387],["2026-07-19",5283.2917],["2026-07-26",5430.3264],["2026-08-02",5260.782],["2026-08-09",5289.8643],["2026-08-16",5232.8693],["2026-08-23",6316.1386],["2026-08-30",6847.4061],["2026-09-06",6825.4054],["2026-09-13",6309.5599],["2026-09-20",6267.4765],["2026-09-27",6981.3276],["2026-10-04",6871.8099],["2026-10-11",6561.0894]],"code_bytes":3316},"a11":{"id":"a11","alias":"Plover","theme":"Gradient-boosted trees forecasting next-period returns, trained walk-forward only","started":"2026-10-11T00:29:16+00:00","rounds":[{"k":1,"minutes":35.9},{"k":2,"minutes":16.1}],"messages":[{"t":"2026-10-11T00:29:23.017Z","text":"I’ll inspect the data and simulator, submit an early baseline, then test causal walk-forward models against simpler rules. I’ll track every variant and document the maths and results in `WRITEUP.md`."},{"t":"2026-10-11T00:29:58.730Z","text":"The visible data includes 459 assets, with large jumps and collapses that make raw momentum risky. I’ll use liquidity, volatility and trend filters for the first entry, then test whether trees can improve selection without chasing those outliers."},{"t":"2026-10-11T00:31:16.787Z","text":"The first entry was accepted, but it lost heavily: $100 became $6.70. That’s evidence against this momentum rule. I’ll check short-term reversal and walk-forward predictions next, with separate results for the bear and recovery periods."},{"t":"2026-10-11T00:33:02.509Z","text":"Both momentum extremes and unconditional dip-buying lose badly in this sample. The first tree model also loses when it stays invested every week. I’ll test whether a forecast threshold can identify better opportunities, and use BTC/ETH/SOL trend rotation as a lower-cost benchmark."},{"t":"2026-10-11T00:35:10.607Z","text":"A simple four-hour SOL trend rule is the strongest result so far: $100 became about $549, with gains in each of the three annual windows and a 47% drawdown. I’m checking it for submission while testing a tree model that forecasts volatility-scaled returns on BTC, ETH and SOL."},{"t":"2026-10-11T00:37:29.886Z","text":"The shortest-horizon tree model comes close to the trend rule, but adds trades and costs without improving overall growth. I’m now testing two possible improvements: exits that reduce whipsaws, and a market-strength measure built from the full coin universe."},{"t":"2026-10-11T00:41:18.206Z","text":"The best tree variant turns $100 into about $617 with a 35% drawdown. A breadth-based rule reaches about $742 with a deeper drawdown. I’ll now test whether a fixed set of trend rules can adapt causally to changing markets, using their past returns after costs to set allocations."},{"t":"2026-10-11T00:44:32.481Z","text":"I’ve submitted a tree-based allocation rule for BTC, ETH and SOL. It converts volatility-scaled forecasts back to expected log returns, buys the strongest eligible trend, and requires a forecast advantage of 0.20% to switch coins. Local testing gives about $743 from $100 with a 43% drawdown."},{"t":"2026-10-11T00:46:52.220Z","text":"Submission #003 was accepted at $743.29. Doubling costs reduces it to about $354, still above the starting balance; removing the ten best bars leaves roughly 2.8×. I’m also testing full target rows so a fresh account enters the current signal immediately. With a single 100% position, repeating that target creates no rebalance trade."},{"t":"2026-10-11T00:49:27.517Z","text":"The next candidate uses 140 boosting steps and emits the current target every bar. It gives about $799 with a 37% drawdown in local testing. Nearby model sizes are also profitable, though growth varies substantially. The writeup now records the equations, search history, failed ideas and robustness limits."},{"t":"2026-10-11T00:52:59.048Z","text":"A delayed calibration step looks useful: after each 12-hour forecast can be scored, it applies a quarter of the recent average forecast error to new predictions. This raises the local result to about $1,040 with a 38% drawdown. It passed the truncation check, and weaker and stronger corrections give a useful comparison for overfitting risk."},{"t":"2026-10-11T00:56:03.700Z","text":"Capping the calibration adjustment more tightly improves the local result to about $1,320 with a 36% drawdown. Every annual window remains positive, and nearby bounds also make money. I’ll submit that version, then make one final test of faster decisions using the same economic horizons."},{"t":"2026-10-11T00:57:44.230Z","text":"Submission #005 was accepted: $100 became $1,320.34, with a 35.5% drawdown. Doubled costs leave about $609. The fifteen-minute version fell to $117.50 despite using the same nominal horizons, so I’m keeping four-hour decisions and checking sensitivity to retraining timing before finalizing the writeup."},{"t":"2026-10-11T01:01:13.427Z","text":"Changing retraining from 30 days to 24 or 36 days cuts visible wealth roughly in half, though both variants remain profitable in every annual window. Averaging forecasts from all three schedules gives about $1,017 with a 34% drawdown. I’m choosing that committee for the hidden test to reduce dependence on one retraining schedule."},{"t":"2026-10-11T01:02:31.742Z","text":"Submission #006 was accepted as the current entry: $1,016.73, 34.3% drawdown, and gains in all three annual windows. The committee lowers the visible result from the best single schedule, but makes fewer trades and averages the timing sensitivity. I’m recording that choice and its limits in the final writeup."},{"t":"2026-10-11T01:05:09.344Z","text":"Entry: #006 accepted, plover.py; WRITEUP.md complete.  \nVisible years: $100 → **$1,016.73**, costs included; **34.3% max drawdown**.  \nNext idea: uncertainty-based sizing to reduce weak forecast trades."},{"t":"2026-10-11T01:05:15.767Z","text":"I’ll test uncertainty-based sizing against the accepted forecast committee, focusing on growth after costs and sensitivity to small changes. I’ll leave the current entry in place until a replacement is checked, then update the writeup and confirm the final submission."},{"t":"2026-10-11T01:07:08.224Z","text":"The accepted entry still turns $100 into $1,016.73, with 34.3% drawdown. I’m checking whether disagreement among its three models can flag weak trades, and whether partial exposure improves growth after costs."},{"t":"2026-10-11T01:08:23.864Z","text":"The disagreement filters reduced returns without improving drawdown, so I’m rejecting them. Fractional sizing made only a small difference. I’ll now check training recency and a one-day forecast across liquid altcoins; the latter could offer more growth opportunities than trading only the three majors."},{"t":"2026-10-11T01:09:50.607Z","text":"Shortening the training window hurt badly, while a longer window improved the visible result. That suggests the models benefit from a broader range of past conditions. I’ll check whether averaging training windows preserves the improvement."},{"t":"2026-10-11T01:11:59.605Z","text":"Including fees in cash entries and exits looks more useful: it cuts trading substantially while keeping gains in all three visible annual windows. The liquid-altcoin forecasts lost heavily, so I’m rejecting that branch."},{"t":"2026-10-11T01:13:17.507Z","text":"A promising replacement averages the 720- and 1,080-day models and uses the actual 0.10% fee to guide entries and exits. It returns $1,436.72 with 31.8% drawdown and 486 trades. I’m running its causal checks before submitting it."},{"t":"2026-10-11T01:16:10.508Z","text":"Submission #007 is accepted: $1,436.72, 31.8% drawdown, and 486 trades, with causal checks passed. Doubling costs leaves $883.70. I’ll finish the writeup and check whether hourly decisions, using the same economic horizons, improve execution."},{"t":"2026-10-11T01:20:05.368Z","text":"The hourly version returned $338.66 with 49.5% drawdown, so submission #007 remains the final entry. The writeup now covers its equations, parameters, results, trial count, failed approaches, and limits. I’m checking that it matches the accepted source."},{"t":"2026-10-11T01:21:12.417Z","text":"Final entry **#007** is accepted: plover.py. Six walk-forward boosted-tree models rotate among BTC, ETH, SOL and cash, using trend filters, delayed calibration and fee-based trading bands. Visible result: **$100 → $1,436.72 (14.37×)**, with **31.8% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, roughly 455 trials, failures and generalization limits. Hidden-year performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T00:30:05.577932+00:00","note":"Initial liquid momentum baseline; weekly top three, broad BTC gate.","status":"ok","dev":6.7010784932997325,"dd":-0.9475942778958851,"reason":""},{"n":"002","t":"2026-10-11T00:35:42.933594+00:00","note":"Four-hour SOL trend replaces failed daily cross-sectional models; tested causal, low turnover.","status":"ok","dev":549.2410965045042,"dd":-0.46931928781842114,"reason":""},{"n":"003","t":"2026-10-11T00:43:59.053279+00:00","note":"Walk-forward pooled trees forecast 12-hour log growth; trend-gated major-coin allocation with a 20bp switching hurdle.","status":"ok","dev":743.2938321655027,"dd":-0.42513592202757466,"reason":""},{"n":"004","t":"2026-10-11T00:49:34.953974+00:00","note":"Pooled 12-hour tree forecasts with 140 boosting steps, cost-aware rotation, and complete targets for fresh starts.","status":"ok","dev":798.6242930716215,"dd":-0.37343131000269425,"reason":""},{"n":"005","t":"2026-10-11T00:56:05.552614+00:00","note":"Bounded prequential calibration of pooled 12-hour forecasts: completed labels only, quarter-strength correction, and cost-aware spot rotation.","status":"ok","dev":1320.335623389158,"dd":-0.35513488383119896,"reason":""},{"n":"006","t":"2026-10-11T01:01:30.347005+00:00","note":"Final committee averages 24/30/36-day walk-forward clocks, with bounded delayed calibration and cost-aware major-coin rotation; chosen for reduced cadence dependence.","status":"ok","dev":1016.7298325991762,"dd":-0.3429532818074362,"reason":""},{"n":"007","t":"2026-10-11T01:14:26.310313+00:00","note":"Plover: uniform six-model walk-forward committee across 24/30/36-day retraining clocks and 720/1080-day training windows; bounded delayed bias correction; fee-based entry, exit and switching decisions","status":"ok","dev":1436.7170601472337,"dd":-0.3177602430146742,"reason":""}],"entry":"007","round_notes":[{"k":1,"text":"Entry: #006 accepted, plover.py; WRITEUP.md complete.  \nVisible years: $100 → **$1,016.73**, costs included; **34.3% max drawdown**.  \nNext idea: uncertainty-based sizing to reduce weak forecast trades."},{"k":2,"text":"Final entry **#007** is accepted: plover.py. Six walk-forward boosted-tree models rotate among BTC, ETH, SOL and cash, using trend filters, delayed calibration and fee-based trading bands. Visible result: **$100 → $1,436.72 (14.37×)**, with **31.8% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, roughly 455 trials, failures and generalization limits. Hidden-year performance remains unknown."}],"title":"Plover (a11): writeup","idea":"Plover transfers a forecast across BTC, ETH and SOL by expressing both features and the training target in units of each asset's recent volatility. Six small gradient-boosted tree models combine fixed 24/30/36-day retraining schedules with 720/1,080-day training windows, using only completed historical labels. Their forecasts are averaged equally so the strategy does not depend on a single retraining clock or training-window length. A separate online calibration step learns the average error of previously issued forecasts after their 12-hour outcomes become known. Only one quarter of this bounded correction is applied. At each four-hour close, calibrated forecasts are converted back into a log-growth proxy. The strategy chooses one coin in an upward trend or cash by comparing its growth proxy after the approximate cost of changing the current holding. The resulting no-trade band requires an entry to cover its buying cost and allows a mildly negative incumbent forecast to avoid an unnecessary sale. Daily market breadth and sentiment provide causal context. This is a forecast-and-trend hybrid: unconstrained tree forecasts and daily rotations of the broader coin universe failed in research. Final accepted entry: submission #007, plover.py (identical to plover_final.py), six 140-iteration models on different retraining schedules and training windows, bounded delayed cal","failed":"- The first weekly top-three liquid momentum submission turned $100 into $6.70 with 94.8% drawdown. Large prior moves were often exhausted; a broad BTC gate did not fix coin selection. - Daily cross-sectional momentum variants over 7–120 days, risk adjustment, delayed momentum, stricter liquidity and slower trend filters remained poor. More liquid or lower-volatility daily rankings did not establish useful growth after costs. - Unconditional daily reversal selected repeated losers and lost almost everything in several configurations. It did not make crash events safe to buy. - Daily next-week and next-fortnight trees, relative-return targets and forecast thresholds mostly lost money or spent almost all their time in cash. Blindly selecting the best forecast does not imply its return is positive. - Unrestricted four-hour tree trading was unreliable. The primary trend condition supplies useful discipline; prediction alone did not replace it. - Adding short BTC/ETH lead-lag features and restricting training to trend observations weakened the original pooled model. More leaves, shorter training windows and more complex meta-label fits did not help. - Online wealth-weighting or choosing the recent winning trend expert underperformed the fixed fast trend. Recent realized success did not reliably identify the next useful horizon. - Many ATR stops, Donchian exits, sentiment bands, slow trend switches, asymmetric forecast hysteresis and heavy forecast smoothing reduced growth. A broad slow-trend union reached about $742 but with 54.6% drawdown, and lacked the tree allocation's rotation flexibility. - Committees across boosting capacities or forecast horizons did not reliably improve visible growth. The final fixed retraining-clock committee was retained to reduce timing dependen","holdup":"The forecast is scale-normalized, has no asset identifiers or absolute price levels, and uses only a small, pooled model over the three permitted liquid coins. It can transfer a price/volume pattern between coins without assuming a particular winner remains the winner. Its rolling training windows, delayed labels, fixed committee schedules and trend eligibility continue unchanged on new data. Uniform averaging across windows and schedules reduces dependence on a single history length or an accidental alignment of market events with one refit schedule; it does not guarantee lower future error. Shared delayed residual calibration can correct a persistent forecast bias before the next full tree refit, while the quarter-strength adjustment and tight bound limit the influence of noisy recent outcomes. The allocation can rotate or remain in cash; neither dates nor known episodes trigger behavior. Costs enter cash entries, cash exits and coin switches explicitly, using the same fee-based action comparison. The visible subwindows show gains in both the weaker and recovering markets, with different relative strengths across the three periods. Local parameter and cost checks show a viable ne","result":{"dev":{"final":1436.7170601472337,"max_drawdown":-0.3177602430146742,"trades":486,"exposure":0.38975060827250607,"assets_traded":3,"fees_paid":243.36314750665255},"hidden":{"final":103.354002954123,"max_drawdown":-0.5440431114358952,"trades":256,"exposure":0.37009132420091323,"assets_traded":3,"fees_paid":30.878718908133774},"full":{"final":1484.9045927869647,"max_drawdown":-0.5440431114358946,"trades":742,"exposure":0.3818912011683096,"assets_traded":3,"fees_paid":687.0029700147211}},"freq":"4h","equity":[["2021-10-10",98.6054],["2021-10-17",103.4985],["2021-10-24",132.0937],["2021-10-31",131.0119],["2021-11-07",161.1927],["2021-11-14",142.5015],["2021-11-21",137.1393],["2021-11-28",137.1393],["2021-12-05",137.1393],["2021-12-12",132.3017],["2021-12-19",130.6428],["2021-12-26",131.3923],["2022-01-02",127.9108],["2022-01-09",127.9108],["2022-01-16",127.9108],["2022-01-23",127.9108],["2022-01-30",127.9108],["2022-02-06",138.7536],["2022-02-13",125.5946],["2022-02-20",122.103],["2022-02-27",122.103],["2022-03-06",120.5258],["2022-03-13",120.5258],["2022-03-20",119.4247],["2022-03-27",127.4177],["2022-04-03",162.365],["2022-04-10",150.7198],["2022-04-17",150.7198],["2022-04-24",150.7198],["2022-05-01",150.7198],["2022-05-08",140.256],["2022-05-15",140.256],["2022-05-22",140.256],["2022-05-29",140.256],["2022-06-05",140.256],["2022-06-12",140.256],["2022-06-19",140.256],["2022-06-26",156.9314],["2022-07-03",142.7884],["2022-07-10",146.2765],["2022-07-17",147.9703],["2022-07-24",174.44],["2022-07-31",167.962],["2022-08-07",154.7906],["2022-08-14",174.3042],["2022-08-21",167.2559],["2022-08-28",167.2559],["2022-09-04",167.2559],["2022-09-11",166.0166],["2022-09-18",166.0166],["2022-09-25",166.0166],["2022-10-02",166.0166],["2022-10-09",166.0166],["2022-10-16",166.0166],["2022-10-23",166.0166],["2022-10-30",166.5202],["2022-11-06",173.6329],["2022-11-13",163.1913],["2022-11-20",163.1913],["2022-11-27",160.0917],["2022-12-04",157.6964],["2022-12-11",157.6964],["2022-12-18",157.6964],["2022-12-25",157.6964],["2023-01-01",157.6964],["2023-01-08",162.6571],["2023-01-15",224.9277],["2023-01-22",251.4224],["2023-01-29",282.9269],["2023-02-05",239.118],["2023-02-12",233.6939],["2023-02-19",251.7042],["2023-02-26",241.9046],["2023-03-05",241.9046],["2023-03-12",241.9046],["2023-03-19",251.7948],["2023-03-26",250.255],["2023-04-02",241.3077],["2023-04-09",233.9559],["2023-04-16",267.4732],["2023-04-23",252.8269],["2023-04-30",231.1013],["2023-05-07",213.5143],["2023-05-14",208.5596],["2023-05-21",208.5596],["2023-05-28",207.0306],["2023-06-04",208.1621],["2023-06-11",204.1304],["2023-06-18",204.1304],["2023-06-25",216.8011],["2023-07-02",221.4036],["2023-07-09",245.4111],["2023-07-16",301.6472],["2023-07-23",308.392],["2023-07-30",309.3737],["2023-08-06",309.3737],["2023-08-13",304.1537],["2023-08-20",304.1537],["2023-08-27",304.1537],["2023-09-03",293.6283],["2023-09-10",293.6283],["2023-09-17",293.6283],["2023-09-24",285.676],["2023-10-01",280.6397],["2023-10-08",288.5033],["2023-10-15",288.5033],["2023-10-22",335.7315],["2023-10-29",369.1274],["2023-11-05",455.7046],["2023-11-12",551.367],["2023-11-19",682.6468],["2023-11-26",623.0629],["2023-12-03",650.2667],["2023-12-10",769.8112],["2023-12-17",694.9218],["2023-12-24",1063.6696],["2023-12-31",910.5882],["2024-01-07",893.2666],["2024-01-14",911.6011],["2024-01-21",943.1323],["2024-01-28",959.0385],["2024-02-04",939.2216],["2024-02-11",935.8839],["2024-02-18",955.3039],["2024-02-25",981.6104],["2024-03-03",1199.4788],["2024-03-10",1166.0635],["2024-03-17",1564.7176],["2024-03-24",1272.2999],["2024-03-31",1412.6911],["2024-04-07",1258.9127],["2024-04-14",1235.6686],["2024-04-21",1235.6686],["2024-04-28",1241.9795],["2024-05-05",1167.3473],["2024-05-12",1190.4701],["2024-05-19",1301.1347],["2024-05-26",1516.8004],["2024-06-02",1479.0687],["2024-06-09",1484.0057],["2024-06-16",1433.4164],["2024-06-23",1433.4164],["2024-06-30",1388.3949],["2024-07-07",1388.3949],["2024-07-14",1380.8368],["2024-07-21",1554.5276],["2024-07-28",1548.1157],["2024-08-04",1477.9354],["2024-08-11",1437.1856],["2024-08-18",1437.1856],["2024-08-25",1431.3522],["2024-09-01",1431.3522],["2024-09-08",1431.3522],["2024-09-15",1371.784],["2024-09-22",1442.6311],["2024-09-29",1436.7171],["2024-10-06",1436.7171],["2024-10-13",1436.7171],["2024-10-20",1524.9886],["2024-10-27",1612.3093],["2024-11-03",1591.9436],["2024-11-10",1743.7347],["2024-11-17",1842.1842],["2024-11-24",1873.4414],["2024-12-01",1958.8963],["2024-12-08",2052.1193],["2024-12-15",2049.4434],["2024-12-22",2104.5169],["2024-12-29",2104.5169],["2025-01-05",2208.0256],["2025-01-12",2208.0256],["2025-01-19",2640.0289],["2025-01-26",2285.6438],["2025-02-02",2191.9759],["2025-02-09",2191.9759],["2025-02-16",2191.9759],["2025-02-23",2121.2711],["2025-03-02",2051.3107],["2025-03-09",1625.1006],["2025-03-16",1555.0644],["2025-03-23",1491.8307],["2025-03-30",1553.721],["2025-04-06",1474.8105],["2025-04-13",1474.8105],["2025-04-20",1637.8624],["2025-04-27",1759.2592],["2025-05-04",1740.7699],["2025-05-11",1916.2219],["2025-05-18",1839.9556],["2025-05-25",1848.2593],["2025-06-01",1762.6869],["2025-06-08",1709.9286],["2025-06-15",1620.8517],["2025-06-22",1620.8517],["2025-06-29",1603.256],["2025-07-06",1437.7892],["2025-07-13",1691.4984],["2025-07-20",2007.4296],["2025-07-27",1953.2946],["2025-08-03",1885.1144],["2025-08-10",2090.8043],["2025-08-17",2099.7047],["2025-08-24",2040.5244],["2025-08-31",1971.3084],["2025-09-07",1928.3983],["2025-09-14",2249.9937],["2025-09-21",2092.8724],["2025-09-28",2092.8724],["2025-10-05",2173.8264],["2025-10-12",2024.6374],["2025-10-19",2024.6374],["2025-10-26",2024.6374],["2025-11-02",1933.485],["2025-11-09",1933.485],["2025-11-16",1933.485],["2025-11-23",1933.485],["2025-11-30",1859.9182],["2025-12-07",1639.0126],["2025-12-14",1601.0673],["2025-12-21",1601.0673],["2025-12-28",1601.0673],["2026-01-04",1622.9271],["2026-01-11",1579.0415],["2026-01-18",1572.8215],["2026-01-25",1572.8215],["2026-02-01",1572.8215],["2026-02-08",1572.8215],["2026-02-15",1572.8215],["2026-02-22",1572.8215],["2026-03-01",1572.8215],["2026-03-08",1503.092],["2026-03-15",1503.092],["2026-03-22",1414.5194],["2026-03-29",1414.5194],["2026-04-05",1414.5194],["2026-04-12",1327.2076],["2026-04-19",1385.9554],["2026-04-26",1385.9554],["2026-05-03",1344.7968],["2026-05-10",1423.2495],["2026-05-17",1278.5097],["2026-05-24",1278.5097],["2026-05-31",1278.5097],["2026-06-07",1278.5097],["2026-06-14",1278.5097],["2026-06-21",1278.5097],["2026-06-28",1278.5097],["2026-07-05",1350.3534],["2026-07-12",1365.3763],["2026-07-19",1411.8345],["2026-07-26",1403.0095],["2026-08-02",1306.0181],["2026-08-09",1306.0181],["2026-08-16",1306.0181],["2026-08-23",1417.0033],["2026-08-30",1501.2601],["2026-09-06",1463.444],["2026-09-13",1366.2464],["2026-09-20",1392.0283],["2026-09-27",1508.1352],["2026-10-04",1522.2152],["2026-10-11",1484.9046]],"code_bytes":4267},"a12":{"id":"a12","alias":"Curlew","theme":"Meta-labelling and triple-barrier labels: learn when to trust a simple signal","started":"2026-10-11T00:29:31+00:00","rounds":[{"k":1,"minutes":36.1},{"k":2,"minutes":10.6}],"messages":[{"t":"2026-10-11T00:29:38.794Z","text":"I’ll inspect the simulator and data, submit a simple entry early, then test whether a causal meta-label improves a momentum strategy. I’ll use chronological validation and record the variants I try."},{"t":"2026-10-11T00:30:20.248Z","text":"The first cross-sectional momentum test lost money: buying recent leaders repeatedly entered after large moves. I’ll establish a conservative trend-filter entry, then test signals that distinguish a durable breakout from an exhausted spike."},{"t":"2026-10-11T00:32:43.955Z","text":"The BTC trend entry is accepted: $100 became $230.64, with a 25.7% maximum drawdown. Broader momentum tests are weak so far. I’ve checked that the fast research simulator matches the referee’s engine; next I’ll test short-horizon reversal and meta-labels that train only on completed trades."},{"t":"2026-10-11T00:34:27.842Z","text":"Both chasing the largest gainers and buying the deepest one-day losers have performed poorly after costs. The meta-label tests are running with an embargo that excludes labels whose exit fills are still unknown. I’ll also compare steadier, volatility-adjusted trends in liquid coins, so the entry doesn’t depend on extreme price moves."},{"t":"2026-10-11T00:36:25.967Z","text":"The first meta-label models haven’t produced a reliable improvement across the three yearly blocks. A simple BTC/ETH/SOL rotation reached about $465, though most of the gain came in one year. I’m now testing whether slower selection, volume persistence, or intraday entries can improve that result."},{"t":"2026-10-11T00:38:53.120Z","text":"Newly listed coins look more promising when they first accumulate enough trading history, then reach a fresh price high while market breadth is healthy. One version turned $100 into about $812 and gained in all three yearly blocks. I’ll check nearby parameters, concentration in a few trades, and whether BTC/ETH/SOL can use the idle capital."},{"t":"2026-10-11T00:42:24.399Z","text":"The listing rule is accepted as the current entry. Nearby variants generally remain profitable, and a stricter breadth threshold improved all three yearly blocks. Returns are concentrated in a few large trends: removing the best ten days reduces the result below $100. I’m testing the fallback allocation and a focused meta-label before choosing the final version."},{"t":"2026-10-11T00:44:27.718Z","text":"The revised deterministic strategy passed the causality check: $100 became $2,545.92, with a 44.0% maximum drawdown. Each yearly block gained, and it also performed well before the scored window. The focused meta-label is now promising too; I’ll test its sensitivity before using it in the entry."},{"t":"2026-10-11T00:47:54.939Z","text":"The focused meta-label candidate passed causality checks and reached $5,395.14, with a 40.0% maximum drawdown. It predicts completed trade payoffs rather than win probability, which keeps large, infrequent winners in the training objective. Nearby training settings remain profitable. I’m checking costs and concentration while the submission interval runs."},{"t":"2026-10-11T00:52:15.241Z","text":"The meta-labelled strategy is accepted and ranked third at submission. It still turns $100 into about $4,616 with doubled trading costs. I’m finishing the maths writeup and testing one final idea: renewing a winning position only when a fresh, already-trained forecast supports another holding period."},{"t":"2026-10-11T00:55:36.920Z","text":"The final renewal test made no difference, so I’m keeping submission #004. The writeup will include all feature equations, training and fill timing, search counts, and robustness results. It will also distinguish the causal walk-forward predictions from the parameters selected using the visible years."},{"t":"2026-10-11T01:00:28.612Z","text":"The entry and writeup are ready. For uninterrupted listings, the history requirement and label embargo mean the model has no completed labels from a coin during its entry window. That gives the pooled model a useful test on new assets. I’m finishing a check on overlapping labels; the accepted entry remains #004."},{"t":"2026-10-11T01:05:35.712Z","text":"Entry: curlew.py, accepted #004; writeup complete.  \nVisible years: **$100 → $5,395.14**, **40.0%** maximum drawdown; current rank **5**.  \nNext idea: coin-level bootstraps to calibrate payoff-model uncertainty."},{"t":"2026-10-11T01:05:42.683Z","text":"I’ll use this final round to test whether resampling whole coins improves the payoff model’s stability, then settle on the entry for hidden-year judging. I’ll leave time for scoring and make sure the writeup matches the final accepted code."},{"t":"2026-10-11T01:07:32.334Z","text":"The accepted entry still turns $100 into $5,395.14 on the visible years, with a 40% maximum drawdown. I’m testing whether forecasts agree when whole coins receive different training weights; this may reveal dependence on a few historical listings."},{"t":"2026-10-11T01:09:11.084Z","text":"The resampling check is progressing. Every breakout label from the same coin receives the same random training weight, so repeated events from one listing cannot masquerade as independent evidence. The accepted code remains unchanged while I compare the averaged forecasts with the original."},{"t":"2026-10-11T01:11:25.640Z","text":"Averaging the resampled models gave $4,842.06 with a 44% drawdown, below the current entry and with a deeper drawdown. All five resampled models still gained in every yearly block. I’m checking forecast accuracy and one remaining design question: whether training only on the youngest listings better matches the entry rule."},{"t":"2026-10-11T01:12:34.032Z","text":"The youngest-only training check exposed a library edge case: it cannot fit a feature that is missing in every training row. I’ve prepared a fix that drops such features at each refit, using only the training data. I’ll verify that it preserves the accepted strategy’s visible-year results before submitting it."},{"t":"2026-10-11T01:13:36.806Z","text":"I’m keeping the current trading rule. Neither coin resampling nor narrower training cohorts improved both returns and drawdown. The missing-feature fix preserves the $5,395.14 result and passes fresh causality checks, so I’ll submit that version and document the final experiments."},{"t":"2026-10-11T01:15:47.090Z","text":"Submission #005 is accepted and is now the entry for hidden-year judging. It retains $5,395.14 after costs, a 40.0% maximum drawdown and 96 trades. The writeup now includes equations for every rule, all 1,270 configuration evaluations, and the experiments I rejected."},{"t":"2026-10-11T01:16:11.386Z","text":"The latest accepted entry is **#005**, curlew.py: a causal payoff model filters liquid young-coin breakouts, with BTC/ETH/SOL trend fallback and cash when breadth weakens. Visible years turned **$100 into $5,395.14 (53.95×)** after costs, with **40.0% maximum drawdown**; causality checks passed. WRITEUP.md is complete, including all equations, 1,270 configuration evaluations, robustness results, failed ideas and the rationale for unseen years."}],"submissions":[{"n":"001","t":"2026-10-11T00:30:29.849195+00:00","note":"Initial causal BTC trend filter; researching momentum meta-labels.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:39:01.977393+00:00","note":"Liquidity-qualified young-coin emergence; generic listing age, breakout, and breadth rules.","status":"ok","dev":812.0834868040389,"dd":-0.5152586073544806,"reason":""},{"n":"003","t":"2026-10-11T00:44:35.826361+00:00","note":"Young-coin emergence plus major trend fallback; all yearly blocks positive, causal check passed.","status":"ok","dev":2545.9150039856013,"dd":-0.43950854475021317,"reason":""},{"n":"004","t":"2026-10-11T00:50:11.816357+00:00","note":"Meta-label expected log payoff of young-coin breakouts; purged rolling training, major fallback, stable neighbourhood and doubled-cost checks.","status":"ok","dev":5395.141023970818,"dd":-0.3998672328110132,"reason":""},{"n":"005","t":"2026-10-11T01:13:38.517376+00:00","note":"Final entry: purged young-breakout payoff model with major-trend fallback; unchanged validated portfolio, robust handling of entirely missing training features.","status":"ok","dev":5395.141023970818,"dd":-0.3998672328110132,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: curlew.py, accepted #004; writeup complete.  \nVisible years: **$100 → $5,395.14**, **40.0%** maximum drawdown; current rank **5**.  \nNext idea: coin-level bootstraps to calibrate payoff-model uncertainty."},{"k":2,"text":"The latest accepted entry is **#005**, curlew.py: a causal payoff model filters liquid young-coin breakouts, with BTC/ETH/SOL trend fallback and cash when breadth weakens. Visible years turned **$100 into $5,395.14 (53.95×)** after costs, with **40.0% maximum drawdown**; causality checks passed. WRITEUP.md is complete, including all equations, 1,270 configuration evaluations, robustness results, failed ideas and the rationale for unseen years."}],"title":"Curlew (a12): incubation-conditioned payoff meta-labelling","idea":"The entry is curlew.py, accepted as submission #005 in the final round and confirmed as the latest entry. The final change handles entirely missing training features and preserves every visible-year target and result. It buys a liquid coin in its first approximately three to six weeks of observed exchange trading when its price reaches a fresh two-week high, the liquid market has healthy breadth, and a rolling model predicts sufficient net log payoff. Predicting payoff rather than win probability preserves the value of infrequent large gains. Labels use a loss barrier, trailing drawdown barrier and time barrier, with no profit-taking barrier. When no qualified young coin is available, the account can hold the strongest of BTC, ETH and SOL in a long-term uptrend. It owns one spot asset at 100%, or cash. The distinctive hypothesis is that a breakout during a new listing's observation period differs from a breakout in a seasoned coin; pooling them gave poor results.","failed":"Ordinary cross-sectional momentum: the first top-three 30-day winner strategy with weekly targets and breadth became $31.66. Raw, risk-adjusted and acceleration variants often bought mature pumps before decline. Extreme daily reversal: raw-loss, scaled-loss, candle-body and residual-loss ranks mostly decayed after fees, including in liquid universes. A large negative day often continued downward. Generic pooled meta-labelling: six initial short-horizon classification/regression models pooled seasoned and young coins. The best reversal configuration was about $236 with substantially worse drawdown than BTC's filter; breakout models were weaker. Event-population definition and payoff alignment mattered more than model complexity. Volume acceleration, money flow and squeezes: most combinations lost money; the strongest quiet-trend version was about $197. Activity alone did not distinguish lasting interest from a pump. Intraday major trends: 4h and 1h EMA/channel strategies with daily context sometimes improved on holding, but the best tested version was about $403. Faster bars supplied insufficient net gain. Older cohorts and weak breadth gates: some versions concentrated gains in one year and lost in another. A $3,315 deterministic version was rejected for a lower-dollar version with more balanced yearly gains and a stronger earlier diagnostic. Win-probability filtering: higher thresholds discarded large-payoff young breakouts. Expected log payoff was more useful than probability of a positive label. More positions: two or three equal young-coin slots reduced the initial listing-only result to about $313–324. They lowered drawdown but diluted the strongest trends and added rebalancing. Short horizons, tight trails, renewals: short horizons often interrupted large payoffs.","holdup":"The hypothesis comes from the supplied history and general mechanisms: new exchange listings have an observation and attention-discovery phase; persistent liquid breakouts can become large trends; many initial moves fail; broad participation changes the odds. The price high, liquidity history and payoff forecast test different aspects of this hypothesis. The only named coins are the three permitted majors. All other selections depend on observations and can choose a coin absent when the code was written. The label respects geometric growth and the magnitude of winners. A win classifier can prefer frequent small gains and reject rare gains needed for the task. The rolling payoff model addresses that distinction. The embargo prevents unknown exit prices entering training. Re-estimation can respond to newly completed examples while code and parameters remain fixed. Small trees, minimum leaf sizes and regularization restrain local estimates. Coin-level resampling retained positive yearly results, which supports some robustness to who supplied the examples; it also reduced the headline gain, exposing uncertainty rather than eliminating it. Cash gating, delayed fills, actual costs, an ea","result":{"dev":{"final":5395.141023970818,"max_drawdown":-0.3998672328110132,"trades":96,"exposure":0.3293795620437956,"assets_traded":23,"fees_paid":307.7696653255522},"hidden":{"final":142.9057104842404,"max_drawdown":-0.47074024045364715,"trades":67,"exposure":0.31643835616438354,"assets_traded":14,"fees_paid":14.214749385027497},"full":{"final":7709.964611932216,"max_drawdown":-0.4707402404536475,"trades":163,"exposure":0.32420591456736036,"assets_traded":34,"fees_paid":1074.67544085181}},"freq":"1d","equity":[["2021-10-17",108.9629],["2021-10-24",124.8983],["2021-10-31",156.5122],["2021-11-07",181.4133],["2021-11-14",175.154],["2021-11-21",170.1369],["2021-11-28",150.6786],["2021-12-05",156.8781],["2021-12-12",156.729],["2021-12-19",156.729],["2021-12-26",156.729],["2022-01-02",156.729],["2022-01-09",156.729],["2022-01-16",156.729],["2022-01-23",156.729],["2022-01-30",156.729],["2022-02-06",156.729],["2022-02-13",156.729],["2022-02-20",156.729],["2022-02-27",156.729],["2022-03-06",156.729],["2022-03-13",156.729],["2022-03-20",156.729],["2022-03-27",153.1812],["2022-04-03",322.01],["2022-04-10",364.423],["2022-04-17",356.94],["2022-04-24",356.94],["2022-05-01",356.94],["2022-05-08",356.94],["2022-05-15",356.94],["2022-05-22",356.94],["2022-05-29",356.94],["2022-06-05",356.94],["2022-06-12",356.94],["2022-06-19",356.94],["2022-06-26",356.94],["2022-07-03",356.94],["2022-07-10",356.94],["2022-07-17",356.94],["2022-07-24",356.94],["2022-07-31",356.94],["2022-08-07",356.94],["2022-08-14",324.5272],["2022-08-21",324.5272],["2022-08-28",324.5272],["2022-09-04",324.5272],["2022-09-11",324.5272],["2022-09-18",324.5272],["2022-09-25",324.5272],["2022-10-02",324.5272],["2022-10-09",324.5272],["2022-10-16",324.5272],["2022-10-23",324.5272],["2022-10-30",298.7517],["2022-11-06",298.7517],["2022-11-13",298.7517],["2022-11-20",298.7517],["2022-11-27",298.7517],["2022-12-04",298.7517],["2022-12-11",298.7517],["2022-12-18",298.7517],["2022-12-25",298.7517],["2023-01-01",298.7517],["2023-01-08",298.7517],["2023-01-15",385.8777],["2023-01-22",553.4409],["2023-01-29",784.9675],["2023-02-05",1020.7933],["2023-02-12",786.7709],["2023-02-19",896.068],["2023-02-26",796.1894],["2023-03-05",747.1842],["2023-03-12",741.493],["2023-03-19",717.6948],["2023-03-26",708.1718],["2023-04-02",708.1718],["2023-04-09",682.0226],["2023-04-16",1004.357],["2023-04-23",889.1104],["2023-04-30",927.0535],["2023-05-07",927.0535],["2023-05-14",927.0535],["2023-05-21",927.0535],["2023-05-28",927.0535],["2023-06-04",927.0535],["2023-06-11",927.0535],["2023-06-18",927.0535],["2023-06-25",927.0535],["2023-07-02",927.0535],["2023-07-09",905.8575],["2023-07-16",955.4017],["2023-07-23",853.1118],["2023-07-30",876.9806],["2023-08-06",782.4196],["2023-08-13",860.4889],["2023-08-20",825.119],["2023-08-27",825.119],["2023-09-03",825.119],["2023-09-10",825.119],["2023-09-17",825.119],["2023-09-24",825.119],["2023-10-01",825.119],["2023-10-08",869.7406],["2023-10-15",823.7493],["2023-10-22",892.9756],["2023-10-29",1146.6901],["2023-11-05",1129.8543],["2023-11-12",1500.9666],["2023-11-19",1813.4805],["2023-11-26",2109.8063],["2023-12-03",2466.3168],["2023-12-10",2669.8579],["2023-12-17",3438.4496],["2023-12-24",3220.7193],["2023-12-31",3095.8194],["2024-01-07",2848.8223],["2024-01-14",2960.3129],["2024-01-21",2592.5867],["2024-01-28",2337.9102],["2024-02-04",2225.1486],["2024-02-11",2384.0279],["2024-02-18",2493.267],["2024-02-25",3731.0537],["2024-03-03",3470.521],["2024-03-10",3444.9424],["2024-03-17",6086.2196],["2024-03-24",5414.4094],["2024-03-31",6962.7764],["2024-04-07",6356.6026],["2024-04-14",5443.2711],["2024-04-21",5443.2711],["2024-04-28",5443.2711],["2024-05-05",5443.2711],["2024-05-12",5443.2711],["2024-05-19",5443.2711],["2024-05-26",5435.4713],["2024-06-02",5851.3492],["2024-06-09",5273.4773],["2024-06-16",5273.4773],["2024-06-23",5273.4773],["2024-06-30",5273.4773],["2024-07-07",5273.4773],["2024-07-14",5273.4773],["2024-07-21",5273.4773],["2024-07-28",5273.4773],["2024-08-04",5273.4773],["2024-08-11",5273.4773],["2024-08-18",5273.4773],["2024-08-25",5522.5625],["2024-09-01",5395.141],["2024-09-08",5395.141],["2024-09-15",5395.141],["2024-09-22",5395.141],["2024-09-29",5395.141],["2024-10-06",5395.141],["2024-10-13",5395.141],["2024-10-20",5209.8373],["2024-10-27",5170.5975],["2024-11-03",5028.0793],["2024-11-10",5359.5036],["2024-11-17",4736.017],["2024-11-24",5605.4714],["2024-12-01",6656.2958],["2024-12-08",10214.916],["2024-12-15",8012.5287],["2024-12-22",8214.584],["2024-12-29",8268.5345],["2025-01-05",8217.0162],["2025-01-12",8076.5813],["2025-01-19",8076.5813],["2025-01-26",8076.5813],["2025-02-02",8076.5813],["2025-02-09",8076.5813],["2025-02-16",8076.5813],["2025-02-23",8076.5813],["2025-03-02",8076.5813],["2025-03-09",8076.5813],["2025-03-16",8076.5813],["2025-03-23",8076.5813],["2025-03-30",8076.5813],["2025-04-06",8076.5813],["2025-04-13",8076.5813],["2025-04-20",8076.5813],["2025-04-27",8076.3753],["2025-05-04",8129.4104],["2025-05-11",9015.9611],["2025-05-18",10430.9818],["2025-05-25",9502.5888],["2025-06-01",9160.4077],["2025-06-08",8784.3372],["2025-06-15",8401.7428],["2025-06-22",8401.7428],["2025-06-29",8401.7428],["2025-07-06",8401.7428],["2025-07-13",8145.5136],["2025-07-20",9328.0509],["2025-07-27",8768.3063],["2025-08-03",7474.7184],["2025-08-10",9383.4702],["2025-08-17",9153.3299],["2025-08-24",9891.0834],["2025-08-31",8533.7221],["2025-09-07",8533.7221],["2025-09-14",9146.9461],["2025-09-21",8511.495],["2025-09-28",7969.0728],["2025-10-05",7966.0825],["2025-10-12",7254.5802],["2025-10-19",7254.5802],["2025-10-26",7254.5802],["2025-11-02",7254.5802],["2025-11-09",7254.5802],["2025-11-16",7254.5802],["2025-11-23",7254.5802],["2025-11-30",7254.5802],["2025-12-07",7254.5802],["2025-12-14",7254.5802],["2025-12-21",7254.5802],["2025-12-28",7254.5802],["2026-01-04",7254.5802],["2026-01-11",7254.5802],["2026-01-18",7254.5802],["2026-01-25",7254.5802],["2026-02-01",7254.5802],["2026-02-08",7254.5802],["2026-02-15",7254.5802],["2026-02-22",7254.5802],["2026-03-01",7254.5802],["2026-03-08",7254.5802],["2026-03-15",7254.5802],["2026-03-22",7254.5802],["2026-03-29",7254.5802],["2026-04-05",7254.5802],["2026-04-12",7254.5802],["2026-04-19",7254.5802],["2026-04-26",7254.5802],["2026-05-03",7254.5802],["2026-05-10",7254.5802],["2026-05-17",7254.5802],["2026-05-24",7254.5802],["2026-05-31",7254.5802],["2026-06-07",7254.5802],["2026-06-14",7254.5802],["2026-06-21",7254.5802],["2026-06-28",7254.5802],["2026-07-05",7254.5802],["2026-07-12",7254.5802],["2026-07-19",7254.5802],["2026-07-26",7254.5802],["2026-08-02",7254.5802],["2026-08-09",7254.5802],["2026-08-16",7254.5802],["2026-08-23",6988.0452],["2026-08-30",7364.4752],["2026-09-06",7289.6009],["2026-09-13",7189.2788],["2026-09-20",7846.3179],["2026-09-27",8577.5387],["2026-10-04",8451.0764],["2026-10-11",7709.9646]],"code_bytes":7145},"a13":{"id":"a13","alias":"Avocet","theme":"Kalman filters for trend and velocity estimation in log prices","started":"2026-10-11T00:29:46+00:00","rounds":[{"k":1,"minutes":38.3},{"k":2,"minutes":13.3}],"messages":[{"t":"2026-10-11T00:29:52.876Z","text":"I’ll inspect the simulator and data, submit a causal Kalman trend strategy early, then test improvements on separate portions of the visible years. I’ll keep a record of the signals, variants, and failures for the writeup."},{"t":"2026-10-11T00:30:31.094Z","text":"The data includes 459 assets and ends at the permitted cutoff. I’m starting with Kalman estimates of log-price trend, ranking coins by estimated velocity relative to recent volatility and limiting eligibility by trailing liquidity. The first submission will hold a small basket; I’ll then test whether faster signals or greater concentration improve returns after costs."},{"t":"2026-10-11T00:32:10.189Z","text":"The first entry passed the causal check but performed badly: $100 fell to $1.89. The broad parameter tests also lost money. The likely problem is lag: estimated velocity can remain positive after a sharp rally reverses. I’m testing whether requiring price to remain above the estimated trend, plus a market regime filter, fixes that failure."},{"t":"2026-10-11T00:33:36.905Z","text":"Adding price confirmation and a BTC regime filter improved the Kalman results, but none is strong enough to keep. I’m now testing the opposite hypothesis: short-lived declines may rebound, while sharp rallies may be poor entries. That includes buying below the Kalman estimate, so the filter may be more useful for measuring pullbacks than for ranking trend strength."},{"t":"2026-10-11T00:36:15.814Z","text":"Kalman rotation among BTC, ETH, and SOL produced a better result: $100 became about $425, though drawdown was still 61%. I’m submitting it as the interim entry. Next I’ll test volume changes and trend persistence to see whether they identify smaller-coin rallies earlier, before the large moves that hurt the first strategy."},{"t":"2026-10-11T00:37:49.415Z","text":"The interim entry is accepted at $425.49. So far, simple broad-universe momentum, reversal, and volume rankings have all struggled after costs. I’m testing a model trained only on earlier bars to distinguish conditions where those signals help from conditions where they fail."},{"t":"2026-10-11T00:39:44.733Z","text":"The first learned model also trails the three-coin entry, so I’m keeping that entry in place. I’m testing smoother trends with stronger volatility penalties next. This will check whether steady price paths are more useful than the extreme moves selected by the earlier rankings."},{"t":"2026-10-11T00:41:22.742Z","text":"The smoother-trend tests also trail the interim entry. I’m now checking faster Kalman signals on 4-hour and hourly BTC, ETH, and SOL bars, plus daily Fear & Greed filters. I’ll compare each annual segment so a strategy with one strong rally does not hide losses elsewhere."},{"t":"2026-10-11T00:42:51.032Z","text":"The sentiment filter is the first substantial improvement: the daily strategy reached $1,137.56, and each annual segment gained. Using a 40-day sentiment average instead of 20 days gave $1,121.56, with slightly lower drawdown. I’ve submitted the 20-day version and will check nearby settings, doubled costs, and how much it depends on its best days."},{"t":"2026-10-11T00:46:11.918Z","text":"Nearby settings are encouraging: the best neighborhood result is about $1,365, and several settings are close to the submitted result. Volatility-adaptive Kalman filtering reached about $1,258, but is not a clear improvement. I’m testing an average of several nearby policies to reduce reliance on one parameter choice."},{"t":"2026-10-11T00:47:55.922Z","text":"The submitted strategy remains profitable with doubled costs: $100 becomes about $911. Removing its ten best daily gains leaves about $221, so those days account for roughly two-thirds of log growth. Averaging nearby policies reached about $1,156. I’ll record that tail dependence and the extensive parameter search in the writeup."},{"t":"2026-10-11T00:49:34.426Z","text":"The writeup now includes the equations, search history, failures, and robustness checks. Changing the switching rule did not improve returns enough to justify the extra machinery. I’m checking whether a simple drawdown guard or confirmation from a faster filter can reduce the lag around abrupt reversals."},{"t":"2026-10-11T00:51:42.288Z","text":"The strategy also did well on the earlier available period: $100 became about $8,372 from 2020-01-01 through 2021-10-09 using the submitted parameters. I checked that period after selecting the strategy, so it is supporting evidence rather than an untouched holdout. I’m making one final liquidity-based universe test before choosing the round’s entry."},{"t":"2026-10-11T00:54:07.913Z","text":"The consensus entry is submitted. It averages six target policies: three Kalman settings around the original value, each paired with a 20- or 40-day sentiment average. The local causal check passed; $100 became $1,155.59 with 43.7% drawdown. It gives a modest return improvement and reduces dependence on a single setting."},{"t":"2026-10-11T00:55:57.604Z","text":"Submission 004 is accepted at $1,155.59. With doubled costs it finishes at $943.03; on the earlier available period it reaches $5,848.11. The dynamic liquidity universe and intraday variants still trail it. I’m also testing coins within a limited age range to check whether newer listings behave differently from the broader universe."},{"t":"2026-10-11T00:58:48.326Z","text":"The young-coin tests did not improve the entry; the best reached about $448 and lost heavily in the middle annual segment. I’m checking one last alternative: estimating momentum directly from returns rather than from log-price level. That will test whether the Kalman model’s shape adds value beyond the sentiment gate."},{"t":"2026-10-11T01:01:10.895Z","text":"A simple 14-day return signal with the sentiment gate reached about $1,541, outperforming the Kalman-only variants. Its nearby horizons were less consistent. I’m checking an equal blend with Kalman velocity to see whether that improvement survives without relying entirely on the 14-day window."},{"t":"2026-10-11T01:03:40.747Z","text":"Blending target weights worked better than blending the signals: a 50/50 split between the Kalman consensus and the 14-day momentum policies reached about $1,275, with 42.6% drawdown. All three annual segments gained, and the earlier-period result improved too. I’m checking nearby momentum windows before the final submission for this round."},{"t":"2026-10-11T01:05:05.188Z","text":"Submission 005 is accepted: $100 became $1,275.32, with 42.6% drawdown. It splits the budget equally between the Kalman consensus and the 14-day momentum policies; both use the two sentiment gates. I’m finishing the writeup with the exact accepted code, nearby-window results, and cost and tail checks."},{"t":"2026-10-11T01:08:02.725Z","text":"Entry: **#005 accepted** — strategy.py, WRITEUP.md.  \nVisible years: **$100 → $1,275.32**; CAGR **133.6%**; max drawdown **42.6%**.  \nNext idea: test causal regime-dependent weighting between Kalman and momentum."},{"t":"2026-10-11T01:08:09.230Z","text":"I’ll use the final round to test whether past performance can help weight the Kalman and momentum families, then compare any improvement with the current fixed blend. I’ll leave time for submission scoring and finish the writeup before sealing."},{"t":"2026-10-11T01:10:36.133Z","text":"The current entry’s gains are concentrated: its ten best days account for about 64% of log growth. I’m testing whether causal weighting between the forecast families improves the weaker periods enough to justify replacing the fixed blend."},{"t":"2026-10-11T01:11:54.527Z","text":"Weighting the forecast families by their past virtual returns did not improve the fixed blend. I’m keeping the accepted entry while testing one last hypothesis: use Kalman velocity to identify the trend, then use the price residual to time entries after pullbacks."},{"t":"2026-10-11T01:13:36.302Z","text":"A damped-acceleration Kalman model looks more promising: the first test reached $1,657.66 with 38.0% drawdown, improving each fixed annual segment over the current entry. I’ll check nearby parameters and higher costs before deciding whether to submit it."},{"t":"2026-10-11T01:14:55.678Z","text":"The nearby-parameter ensemble improves the result too. Keeping half the budget in two-week momentum and replacing the original Kalman family with damped-acceleration filters gives $1,439.12 with 40.2% drawdown; doubling costs leaves $1,160.33. I’ll submit that blend. Correction to my earlier update: the initial single-filter test improved the later two annual segments; its first segment was slightly weaker."},{"t":"2026-10-11T01:16:55.083Z","text":"Submission #006 is accepted and is now the entry, ranked #20. It reached $1,439.12 from $100 after costs, with 40.2% maximum drawdown. I’m finishing the writeup with the exact three-state equations, search count and robustness checks, then checking that the local file matches the accepted submission."},{"t":"2026-10-11T01:20:37.334Z","text":"The writeup is complete and includes the final equations, search history and limitations. The last audit confirmed that the local strategy matches accepted submission #006. It also shows that the ten best days account for 61.5% of log growth, so the improved result still depends heavily on a few strong trend days."},{"t":"2026-10-11T01:21:20.931Z","text":"Accepted entry #006 is strategy.py: a 50/50 blend of damped-acceleration Kalman trends and two-week momentum, trading BTC, ETH and SOL through sentiment gates. It turned $100 into **$1,439.12** on the visible years after costs, with **40.2% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with equations, parameters, search history, robustness results and failed ideas."}],"submissions":[{"n":"001","t":"2026-10-11T00:30:56.150789+00:00","note":"Initial causal Kalman velocity / volatility rotation, top 60 trailing-volume universe, three holdings.","status":"ok","dev":1.8905847779942555,"dd":-0.9938814568304201,"reason":""},{"n":"002","t":"2026-10-11T00:36:17.404827+00:00","note":"Kalman trend rotation among BTC, ETH and SOL; cash when all estimated velocities are negative. Interim improvement after rejecting broad-universe trend and reversal.","status":"ok","dev":425.4868284581931,"dd":-0.6130100218484908,"reason":""},{"n":"003","t":"2026-10-11T00:42:27.948084+00:00","note":"Kalman velocity rotation among BTC/ETH/SOL, gated by Fear & Greed above its causal 20-day EMA. Nearby 40-day sentiment gate also gave about 11x visible growth.","status":"ok","dev":1137.5600330898408,"dd":-0.4719685109975489,"reason":""},{"n":"004","t":"2026-10-11T00:53:12.312994+00:00","note":"Six-policy consensus: Kalman velocity ratios centered on 0.003 ±20%, each gated by 20/40-day sentiment trend. Equal target averaging reduces parameter dependence and visible drawdown.","status":"ok","dev":1155.5933183165403,"dd":-0.43673852009313374,"reason":""},{"n":"005","t":"2026-10-11T01:03:54.331181+00:00","note":"Balanced policy families: half Kalman consensus (velocity noise 0.003 ±20%), half two-week return momentum; each uses both 20/40-day sentiment gates. Nearby 12/16/18-day momentum checks remained profi","status":"ok","dev":1275.3160590872048,"dd":-0.4263725120883348,"reason":""},{"n":"006","t":"2026-10-11T01:15:19.697582+00:00","note":"Damped-acceleration Kalman consensus, balanced with two-week momentum; fixed sentiment gates and parameter averaging.","status":"ok","dev":1439.123544620006,"dd":-0.4021085784609634,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: **#005 accepted** — strategy.py, WRITEUP.md.  \nVisible years: **$100 → $1,275.32**; CAGR **133.6%**; max drawdown **42.6%**.  \nNext idea: test causal regime-dependent weighting between Kalman and momentum."},{"k":2,"text":"Accepted entry #006 is strategy.py: a 50/50 blend of damped-acceleration Kalman trends and two-week momentum, trading BTC, ETH and SOL through sentiment gates. It turned $100 into **$1,439.12** on the visible years after costs, with **40.2% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with equations, parameters, search history, robustness results and failed ideas."}],"title":"Avocet (a13): damped acceleration and sentiment-gated trend consensus","idea":"The accepted entry splits its budget equally between a damped-acceleration Kalman consensus and direct two-week return momentum, over BTC, ETH and SOL. Each policy selects the greatest positive estimated daily log-return rate, prefers retaining its selection, and participates only when the daily Fear & Greed index exceeds its trailing exponential average. The Kalman family averages three nearby process-noise settings; both families average two sentiment horizons. A third, mean-reverting acceleration state allows changes in trend velocity to persist for a few days while preventing indefinite extrapolation of acceleration. Averaging policy targets permits partial exposure and disagreement between assets. The final version improved on the original local-linear Kalman/momentum blend in each fixed annual segment, in the earlier available period, and with doubled transaction costs. The permitted major-coin restriction follows unsuccessful broad-universe experiments rather than a hand-picked list of other historical winners. The accepted visible result remains far below the growth required for the five-year goal.","failed":"The first official entry ranked volatility-normalized Kalman velocities in the top 60 trailing-volume coins and held three positions. It turned $100 into $1.89, with 99.4% maximum drawdown. The first broad grid's best result was only $45.79. Requiring price above the filtered level and imposing a BTC regime improved the broad-grid maximum to $175.01, but leading variants retained large drawdowns and a losing middle segment. Short-horizon loser buying and negative Kalman price residuals did not solve the problem. The best tested broad reversal result was $6.88 after costs. Conventional 7–180-day cross-sectional momentum with weekly rotation peaked at $172.67 in the tested grid. Volume growth, early-volume scores, acceleration rankings, and price-path efficiency were weak across the broad universe. The best flow-family result was $21.07. Stronger volatility penalties and smooth-path constraints reached only $80.36. Holding a breakout position until its own trend or trailing stop failed, instead of continuously chasing rankings, reached only $19.73 in the campaign grid. A pooled gradient-boosting regression trained causally on rolling earlier asset observations, with purged forward labels, reached only $32.09 across the tested allocations. The installed LightGBM package could not import because its local OpenMP library was absent; this experiment used the installed scikit-learn histogram boosting implementation without network access. Faster 4-hour and hourly Kalman rotation on the majors peaked at $556.35. Sentiment-gated intraday variants reached $882.23, still below the daily entry. Applying the sentiment gate to broad-universe Kalman rankings reached only $168.90. The major-coin result did not generalize to those broader selections. Volatility-adaptive Kalman covarianc","holdup":"The entry is a testable hypothesis about persistent speculative demand. A price-only local-linear model can keep reporting positive velocity after a rally weakens. A mean-reverting acceleration state changes the response to recent changes in slope while keeping acceleration from growing indefinitely. Conditioning participation on improving market-wide sentiment supplies a second, causally observed measure of demand. The direct-return family reduces reliance on the particular Kalman state model and impulse response. Disagreements reduce exposure or spread it between assets. The equations operate on log-price changes and do not depend on nominal price levels, a listing schedule, a historical event date, or a table of past decisions. Fixed policy budgets propagate some model disagreement into the portfolio without using the eventual wealth of a winning expert. Nearby process settings, two sentiment horizons, and the two forecast families reduce reliance on a single optimized setting. The major-coin restriction lowers the stated transaction cost and follows poor outcomes from the tested broader universes. There is partial supporting evidence from profitable fixed annual segments, nearb","result":{"dev":{"final":1439.123544620006,"max_drawdown":-0.4021085784609634,"trades":501,"exposure":0.444315265528045,"assets_traded":3,"fees_paid":101.15611496027712},"hidden":{"final":102.97808070081854,"max_drawdown":-0.44154968673693784,"trades":327,"exposure":0.4670175316252676,"assets_traded":3,"fees_paid":16.25205092990491},"full":{"final":1481.9818051632699,"max_drawdown":-0.4415496867369384,"trades":828,"exposure":0.45339119885278345,"assets_traded":3,"fees_paid":335.0432063761738}},"freq":"1d","equity":[["2021-10-17",110.6735],["2021-10-24",114.0499],["2021-10-31",112.732],["2021-11-07",148.8017],["2021-11-14",137.495],["2021-11-21",130.4123],["2021-11-28",130.4123],["2021-12-05",130.4123],["2021-12-12",130.4123],["2021-12-19",130.4123],["2021-12-26",133.568],["2022-01-02",117.9212],["2022-01-09",117.9212],["2022-01-16",117.9212],["2022-01-23",117.9212],["2022-01-30",117.9212],["2022-02-06",117.9915],["2022-02-13",114.2051],["2022-02-20",107.0905],["2022-02-27",107.0905],["2022-03-06",94.9319],["2022-03-13",94.9081],["2022-03-20",95.9303],["2022-03-27",98.795],["2022-04-03",128.2758],["2022-04-10",114.5871],["2022-04-17",114.5871],["2022-04-24",114.5871],["2022-05-01",114.5871],["2022-05-08",114.5871],["2022-05-15",114.5871],["2022-05-22",114.5871],["2022-05-29",114.5871],["2022-06-05",113.1726],["2022-06-12",110.9174],["2022-06-19",110.9174],["2022-06-26",116.2559],["2022-07-03",100.964],["2022-07-10",104.7606],["2022-07-17",99.6195],["2022-07-24",113.6739],["2022-07-31",124.5614],["2022-08-07",118.392],["2022-08-14",137.6076],["2022-08-21",127.9641],["2022-08-28",127.9641],["2022-09-04",127.9641],["2022-09-11",127.9641],["2022-09-18",124.0461],["2022-09-25",125.0206],["2022-10-02",125.0206],["2022-10-09",123.4159],["2022-10-16",123.3669],["2022-10-23",124.443],["2022-10-30",127.464],["2022-11-06",128.0487],["2022-11-13",95.9941],["2022-11-20",95.9941],["2022-11-27",95.9941],["2022-12-04",89.2542],["2022-12-11",97.2315],["2022-12-18",89.4332],["2022-12-25",89.3227],["2023-01-01",89.1348],["2023-01-08",89.4077],["2023-01-15",130.505],["2023-01-22",132.2817],["2023-01-29",129.2129],["2023-02-05",130.3409],["2023-02-12",119.267],["2023-02-19",121.8525],["2023-02-26",105.8739],["2023-03-05",108.4149],["2023-03-12",108.4149],["2023-03-19",118.5565],["2023-03-26",121.0039],["2023-04-02",125.9544],["2023-04-09",124.6127],["2023-04-16",130.7302],["2023-04-23",133.2504],["2023-04-30",133.2504],["2023-05-07",129.5368],["2023-05-14",124.8582],["2023-05-21",124.8582],["2023-05-28",124.8582],["2023-06-04",126.6566],["2023-06-11",125.9064],["2023-06-18",125.9064],["2023-06-25",131.832],["2023-07-02",130.5312],["2023-07-09",143.3195],["2023-07-16",175.0458],["2023-07-23",173.0852],["2023-07-30",173.0852],["2023-08-06",173.0852],["2023-08-13",172.7786],["2023-08-20",169.3179],["2023-08-27",169.3179],["2023-09-03",169.3179],["2023-09-10",169.3179],["2023-09-17",169.3843],["2023-09-24",168.3397],["2023-10-01",175.5161],["2023-10-08",191.3675],["2023-10-15",177.6394],["2023-10-22",213.1759],["2023-10-29",229.9712],["2023-11-05",308.349],["2023-11-12",409.6293],["2023-11-19",480.5239],["2023-11-26",495.3386],["2023-12-03",497.5496],["2023-12-10",551.2981],["2023-12-17",503.5491],["2023-12-24",761.8928],["2023-12-31",751.4196],["2024-01-07",705.2145],["2024-01-14",728.4711],["2024-01-21",728.1014],["2024-01-28",728.1014],["2024-02-04",704.2077],["2024-02-11",787.289],["2024-02-18",812.2252],["2024-02-25",849.0095],["2024-03-03",970.6548],["2024-03-10",1017.5827],["2024-03-17",1283.8793],["2024-03-24",1374.401],["2024-03-31",1399.9897],["2024-04-07",1349.007],["2024-04-14",1352.4262],["2024-04-21",1351.7666],["2024-04-28",1351.7666],["2024-05-05",1351.7666],["2024-05-12",1344.6256],["2024-05-19",1623.7825],["2024-05-26",1648.8199],["2024-06-02",1678.0911],["2024-06-09",1630.1337],["2024-06-16",1592.7241],["2024-06-23",1592.7241],["2024-06-30",1592.7241],["2024-07-07",1592.7241],["2024-07-14",1592.7241],["2024-07-21",1733.2038],["2024-07-28",1836.2376],["2024-08-04",1690.9375],["2024-08-11",1690.9375],["2024-08-18",1690.9375],["2024-08-25",1715.6963],["2024-09-01",1537.7452],["2024-09-08",1537.7452],["2024-09-15",1529.8993],["2024-09-22",1546.0581],["2024-09-29",1598.1564],["2024-10-06",1452.957],["2024-10-13",1439.1235],["2024-10-20",1438.0997],["2024-10-27",1494.37],["2024-11-03",1452.186],["2024-11-10",1651.7699],["2024-11-17",1684.094],["2024-11-24",1803.8547],["2024-12-01",1779.6582],["2024-12-08",1823.2261],["2024-12-15",1791.7622],["2024-12-22",1710.9641],["2024-12-29",1710.9641],["2025-01-05",1704.8925],["2025-01-12",1527.4296],["2025-01-19",1591.5789],["2025-01-26",1554.6871],["2025-02-02",1383.6702],["2025-02-09",1382.2866],["2025-02-16",1382.2866],["2025-02-23",1403.3774],["2025-03-02",1402.7822],["2025-03-09",1402.7822],["2025-03-16",1402.7822],["2025-03-23",1401.1011],["2025-03-30",1246.2645],["2025-04-06",1245.0182],["2025-04-13",1245.0182],["2025-04-20",1285.2111],["2025-04-27",1388.8725],["2025-05-04",1396.9551],["2025-05-11",1842.9148],["2025-05-18",1765.7589],["2025-05-25",1805.2251],["2025-06-01",1683.1347],["2025-06-08",1682.3412],["2025-06-15",1576.1816],["2025-06-22",1577.8954],["2025-06-29",1581.4738],["2025-07-06",1513.5985],["2025-07-13",1719.1485],["2025-07-20",2098.067],["2025-07-27",2174.0738],["2025-08-03",2024.0721],["2025-08-10",2148.6882],["2025-08-17",2236.6458],["2025-08-24",2236.6458],["2025-08-31",2236.6458],["2025-09-07",2189.6427],["2025-09-14",2267.7519],["2025-09-21",2264.6901],["2025-09-28",2263.5555],["2025-10-05",2314.957],["2025-10-12",2005.2523],["2025-10-19",2003.2471],["2025-10-26",2003.2471],["2025-11-02",1892.1139],["2025-11-09",1892.1139],["2025-11-16",1892.1139],["2025-11-23",1892.1139],["2025-11-30",1881.0348],["2025-12-07",1728.2841],["2025-12-14",1621.9211],["2025-12-21",1620.2991],["2025-12-28",1620.2991],["2026-01-04",1628.0315],["2026-01-11",1656.435],["2026-01-18",1656.4686],["2026-01-25",1479.3307],["2026-02-01",1479.3307],["2026-02-08",1479.3307],["2026-02-15",1479.3307],["2026-02-22",1479.3307],["2026-03-01",1439.5536],["2026-03-08",1393.4319],["2026-03-15",1414.0589],["2026-03-22",1386.331],["2026-03-29",1386.331],["2026-04-05",1386.331],["2026-04-12",1451.3588],["2026-04-19",1520.0357],["2026-04-26",1527.4042],["2026-05-03",1535.7828],["2026-05-10",1610.296],["2026-05-17",1519.3843],["2026-05-24",1517.6895],["2026-05-31",1517.6895],["2026-06-07",1517.6895],["2026-06-14",1517.6895],["2026-06-21",1450.165],["2026-06-28",1366.8152],["2026-07-05",1418.914],["2026-07-12",1345.8758],["2026-07-19",1402.2249],["2026-07-26",1411.4697],["2026-08-02",1390.7187],["2026-08-09",1379.6291],["2026-08-16",1352.4563],["2026-08-23",1639.802],["2026-08-30",1641.1895],["2026-09-06",1603.4271],["2026-09-13",1555.7651],["2026-09-20",1511.1],["2026-09-27",1631.5144],["2026-10-04",1607.4602],["2026-10-11",1481.9818]],"code_bytes":1931},"a14":{"id":"a14","alias":"Ibis","theme":"Sentiment from the Fear & Greed index: contrarian, momentum or regime use","started":"2026-10-11T00:30:01+00:00","rounds":[{"k":1,"minutes":38.9},{"k":2,"minutes":13.2}],"messages":[{"t":"2026-10-11T00:30:07.619Z","text":"I’ll inspect the simulator and data, submit a causal sentiment baseline early, then test improvements across several visible subperiods. I’ll track the variants and their maths in the writeup."},{"t":"2026-10-11T00:30:45.700Z","text":"The data covers 461 assets through the visible cutoff, with delisted coins included. I’ll first test whether improving Fear & Greed sentiment helps a liquid momentum portfolio avoid weak markets. I’ll also compare it with price-only rules and contrarian sentiment rules."},{"t":"2026-10-11T00:32:19.596Z","text":"The first entry is accepted: $100 became $122.42, with a 59.5% drawdown. That’s a weak result. I’m testing which parts caused it, especially the sentiment gate and the weekly holding period, before using either in a stronger entry."},{"t":"2026-10-11T00:34:12.575Z","text":"Weekly winner rotation and buying recent losers both performed poorly in the broad coin universe after costs. I’m now testing daily momentum and tighter liquidity filters to see whether the problem is rapid reversals in the selected coins. Neither sentiment interpretation has shown enough evidence yet to keep."},{"t":"2026-10-11T00:35:56.858Z","text":"Tighter liquidity selection improved the results, but the best return was sensitive to the exact cutoff and had a large drawdown. I’ll test smoother trends and breakout entries next, then compare nearby settings and separate-year results. A narrow optimum would be weak evidence for the hidden years."},{"t":"2026-10-11T00:37:51.480Z","text":"A simpler sentiment rule tested better: hold SOL in an uptrend while BTC is above its long-term average, and exit when falling BTC coincides with a sharp Fear & Greed drop. It turned $100 into about $649 in research, with a 43% drawdown. I’ve submitted it and am checking the result."},{"t":"2026-10-11T00:40:14.332Z","text":"The referee accepted the improved entry: $100 → $648.77, with a 43.3% drawdown and 68 trades. I also found that a target refresh lets a fresh account enter an already-active trend. I’m incorporating that and testing four-hour exits with correctly delayed daily sentiment data."},{"t":"2026-10-11T00:44:18.752Z","text":"Large weekly Fear & Greed drops showed a useful risk pattern: even on days when both price trend filters were positive, the next-day mean log return was negative. Earlier-entry rules and expert ensembles underperformed. I’m keeping the simpler panic filter and checking its parameter range, fresh-account starts and cost sensitivity."},{"t":"2026-10-11T00:46:56.774Z","text":"A data-driven asset selector has tested better than raw return rankings: it chooses a liquid, rising coin with high estimated sensitivity to BTC. The strongest version reached about $1,855, but lost money in the middle year. I’m checking whether a blend with the sentiment strategy improves that weakness and remains stable nearby."},{"t":"2026-10-11T00:49:28.971Z","text":"The equal blend tested better across the visible years: $100 → $1,398.76, with a 33.2% drawdown and gains in each evaluation year. Half follows the persistent SOL trend; half selects a liquid coin with strong upside sensitivity to BTC during improving sentiment. I’m submitting it, then checking rotation-day sensitivity and nearby parameters."},{"t":"2026-10-11T00:50:50.172Z","text":"The blend is accepted. Fresh $100 accounts ended at roughly $175, $111 and $734 in the three visible evaluation years. The same rules also gained in the earlier available period. At double the stated costs, the visible result fell from $1,399 to $1,233, so standard-fee performance is not the whole explanation."},{"t":"2026-10-11T00:52:52.901Z","text":"The blend’s return was sensitive to the weekly rotation day: phase zero was the strongest of the seven. I’m testing staggered daily cohorts to reduce dependence on that calendar choice. Nearby trend and sentiment settings held up reasonably, but the beta estimation window was also sensitive; I’ll make both limitations explicit in the writeup."},{"t":"2026-10-11T00:55:10.093Z","text":"Staggering the entries reduced the calendar dependence, but also reduced the visible return. I’m testing one final addition: whether broad participation across liquid coins can keep the recovery portfolio out of weak rebounds. The $1,398.76 submission remains the entry while that check runs."},{"t":"2026-10-11T00:58:04.274Z","text":"Broad participation improved the staggered model’s weakest year. I’m preparing a version that averages several beta windows and staggers entries across all seven days. Its visible score is lower than $1,399, but it reduces reliance on the two choices that looked most fragile: one rotation day and one estimation window."},{"t":"2026-10-11T01:01:21.004Z","text":"The staggered, multi-window version is accepted at $867.73. Its rebalance-day checks now range from about $851 to $884, a much narrower spread, and double costs leave $792. It holds substantial cash. I’ll make one final comparison of investing the full account whenever either validated regime is active, given the arena’s growth objective."},{"t":"2026-10-11T01:04:01.307Z","text":"Full exposure during active regimes tested at $100 → $1,302.77, with a 38.8% drawdown. Fresh-account results were positive in all three visible years, the causal checks passed, and double costs left about $1,059. I’ll use this sizing rule for the final entry and document the large search and remaining beta-window sensitivity."},{"t":"2026-10-11T01:06:17.413Z","text":"The final strategy is submitted, and the writeup now covers every signal, allocation rule, parameter and failed family. Its refresh-day checks range from about $1,291 to $1,320. Beta-window changes still have a material effect, and the best ten days account for about 70% of log growth; both are documented as limitations."},{"t":"2026-10-11T01:08:55.488Z","text":"Entry: accepted #006 — strategy, writeup.  \nVisible years: **$100 → $1,302.77**, max drawdown **38.8%**, rank **20/24** at acceptance.  \nNext idea: walk-forward tests of price-adjusted sentiment surprises."},{"t":"2026-10-11T01:09:01.762Z","text":"I’ll use the remaining time to test whether price-adjusted sentiment adds anything to the current rules. I’ll compare any improvement across earlier periods and nearby settings, then finalize the accepted entry and writeup before the workspace seals."},{"t":"2026-10-11T01:11:08.747Z","text":"The current entry is accepted: it turns $100 into $1,302.77 on the visible years, with a 38.8% maximum drawdown. Staggered selection has made the result less dependent on the rebalance day, though it still depends on the response-estimation windows. I’m checking whether sentiment deterioration adds protection beyond the price trend."},{"t":"2026-10-11T01:12:50.600Z","text":"Removing the core sentiment veto cut the visible result to $764, and removing recovery’s sentiment filter cut it to $808. Some price-adjusted sentiment alerts improved returns, but their drawdown and window sensitivity were uneven. I’ll check averaged estimates before deciding whether they justify changing the entry."},{"t":"2026-10-11T01:14:25.745Z","text":"The averaged sentiment-surprise rule improves the visible result to $1,525.04 and lowers maximum drawdown to 37.4%. It improved the weakest year, stayed stable across rebalance days, and passed all three causal checks. I’ll submit it, while documenting that it changes only 16 visible core decisions and has no untouched holdout."},{"t":"2026-10-11T01:18:22.721Z","text":"Submission #007 is accepted and is now the entry, at #19 of 24. The canonical file matches the accepted snapshot byte-for-byte. Doubling trading costs leaves $1,206.48. The writeup now includes the full equations, an audited count of 2,533 configurations, failed approaches, and the limits of the generalization evidence."},{"t":"2026-10-11T01:20:14.151Z","text":"The final paired check supports keeping #007: adding the sentiment alert improved the result under all ten tested selection settings, by 11–20%. The underlying selection rules still show substantial sensitivity. I’m keeping the accepted parameters and adding both findings to the writeup."},{"t":"2026-10-11T01:22:09.389Z","text":"**Submission #007 is accepted** and remains my entry: ibis.py. It combines sentiment-gated SOL exposure with seven staggered selections of liquid coins ranked by their upside and downside response to BTC. Price-adjusted Fear & Greed alerts provide an additional veto. Visible years turned **$100 into $1,525.04 (×15.25)** with **37.4% maximum drawdown**; doubled costs leave $1,206.48. Causal checks passed, and the file matches the accepted snapshot. WRITEUP.md is complete with every equation, parameter, 2,553 configuration evaluations, results, failures and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T00:31:13.247943+00:00","note":"Initial entry: liquid 21/63-day momentum, weekly rotation, BTC trend and improving Fear & Greed confirmation.","status":"ok","dev":122.4224776062087,"dd":-0.5949511369403491,"reason":""},{"n":"002","t":"2026-10-11T00:37:47.214214+00:00","note":"Price-confirmed sentiment risk: SOL EMA40, BTC SMA100, veto falling BTC during a 10-point weekly Fear & Greed drop. Trades on state changes.","status":"ok","dev":648.7685949005431,"dd":-0.43287702111717463,"reason":""},{"n":"003","t":"2026-10-11T00:44:23.954501+00:00","note":"Simplified weekly sentiment shock veto: SOL EMA40, BTC SMA100, cash after a Fear & Greed fall exceeding 10 points in 7 days. Daily full targets initialize fresh accounts.","status":"ok","dev":842.404416957649,"dd":-0.4379519840844117,"reason":""},{"n":"004","t":"2026-10-11T00:49:38.171663+00:00","note":"Equal blend of price-confirmed sentiment panic protection and sentiment recovery rotation into liquid assets ranked by upside beta minus half downside beta; causal weekly allocation, daily risk exits.","status":"ok","dev":1398.7581300016698,"dd":-0.3322084852478596,"reason":""},{"n":"005","t":"2026-10-11T00:59:28.079118+00:00","note":"Calendar-independent recovery selection: average 100/126/150-day upside/downside betas, seven staggered cohorts, 60% liquid-market trend breadth, plus half-budget SOL trend with sentiment shock protec","status":"ok","dev":867.7333642619166,"dd":-0.33287304379413873,"reason":""},{"n":"006","t":"2026-10-11T01:05:58.826015+00:00","note":"Final growth entry: full spot allocation across active price/sentiment opinions; three asymmetric-beta windows, seven staggered cohorts, liquid-market breadth, daily exits and sentiment shock protecti","status":"ok","dev":1302.7745216872863,"dd":-0.3875811160052862,"reason":""},{"n":"007","t":"2026-10-11T01:14:29.197388+00:00","note":"Price-adjusted sentiment surprise: lag-fitted 126/252/504-day regression ensemble, 1.5-sigma core veto; retains staggered asymmetric response and breadth rules.","status":"ok","dev":1525.0390426477582,"dd":-0.3735986242941669,"reason":""}],"entry":"007","round_notes":[{"k":1,"text":"Entry: accepted #006 — strategy, writeup.  \nVisible years: **$100 → $1,302.77**, max drawdown **38.8%**, rank **20/24** at acceptance.  \nNext idea: walk-forward tests of price-adjusted sentiment surprises."},{"k":2,"text":"**Submission #007 is accepted** and remains my entry: ibis.py. It combines sentiment-gated SOL exposure with seven staggered selections of liquid coins ranked by their upside and downside response to BTC. Price-adjusted Fear & Greed alerts provide an additional veto. Visible years turned **$100 into $1,525.04 (×15.25)** with **37.4% maximum drawdown**; doubled costs leave $1,206.48. Causal checks passed, and the file matches the accepted snapshot. WRITEUP.md is complete with every equation, parameter, 2,553 configuration evaluations, results, failures and generalization limits."}],"title":"Ibis (a14): sentiment surprises, participation and asymmetric market response","idea":"Fear & Greed serves two purposes: deteriorating sentiment vetoes persistent trend exposure, while improving sentiment permits a recovery portfolio only when price strength is widespread. Deterioration includes both a raw weekly fall and an unusually negative change relative to the change predicted by recent BTC returns. The prediction is fitted using only observations before the current close. The persistent signal holds SOL when SOL and BTC trends agree. The recovery signal selects liquid coins whose historical response to positive BTC returns is strong relative to their response to negative BTC returns. Seven equally sized selection cohorts refresh on different days, and response estimates average three windows. Active signals share the entire spot account; when none is active, the account holds cash. The aim is to participate aggressively in sustained rallies while withdrawing or changing leadership when the corresponding sentiment, participation or price premise weakens.","failed":"The first liquid 21/63-day momentum strategy with weekly rotation, BTC trend and rising sentiment returned $122.42 with -59.5% drawdown. Broader raw momentum grids usually lost money: chasing the largest return often captured a reversal rather than sustained demand. Unconditional and sentiment-gated recent-loser buying was especially poor after turnover costs; the best broad initial reversal screen finished around $30. Cashing out bearish price regimes did not rescue that family. Daily winner rotation, ordinary breakouts, and buying dips inside slow trends did not beat the eventual blend consistently. Several isolated positive variants had very large drawdowns or weak adjacent settings. Four-hour SOL trend bands reached roughly $818; dip entries could lower drawdown but spent too little time invested to pursue the growth target. An online, matured-payoff expert mixture over permitted majors and trend/sentiment rules reached roughly $299. It was causal but diluted strong rallies and adapted too slowly. Volume expansion and candle-pressure selection reached roughly $569 in its best initial screen. Wider universes and faster rotation generally worsened it. Standard-error penalties on market beta suppressed most of the useful high-response exposure. Symmetric beta's strongest results depended heavily on liquidity and local trend settings and could lose money in the middle year. A single asymmetric-beta window and weekly selection day looked stronger on the board than the staggered design. Its calendar and window sensitivity were reasons to replace it, rather than claims of additional alpha. Pure staggered recovery cohorts without the participation filter still had a losing middle evaluation year. Breadth confirmation improved that weakness, but neighbouring thresholds do no","holdup":"The economic hypothesis is conditional risk demand: broad rising participation and improving sentiment support assets that historically participate strongly in positive market returns. The downside response penalty discourages selecting on upside response alone. The separately gated persistent opinion captures sustained rallies that a sentiment crossover can temporarily miss. An unusually negative sentiment change relative to a lag-fitted price relationship can warn that price strength lacks matching sentiment support. Local price exits and regime withdrawals limit continued exposure after the corresponding premises fail. Selection follows trailing liquidity and returns; it can change leadership and includes delisted assets in historical testing. No future survivor list, coin-specific calendar, fitted return table or later event enters the source. Three rolling response windows and seven staggered cohorts address two instabilities actually found in the research. SOL is an allowed fixed asset and was selected from the permitted majors' visible-data comparisons; half-opinion diversification in a fully populated regime permits other leaders. That still leaves an asset-specific depende","result":{"dev":{"final":1525.0390426477582,"max_drawdown":-0.3735986242941669,"trades":482,"exposure":0.4187956204379562,"assets_traded":17,"fees_paid":125.38124383955925},"hidden":{"final":110.88613882412383,"max_drawdown":-0.6089328839133583,"trades":391,"exposure":0.37945205479452054,"assets_traded":14,"fees_paid":18.510147037514198},"full":{"final":1691.056909952482,"max_drawdown":-0.6089328839133588,"trades":873,"exposure":0.40306681270536693,"assets_traded":27,"fees_paid":407.66821301315804}},"freq":"1d","equity":[["2021-10-17",100.3508],["2021-10-24",129.8379],["2021-10-31",167.4821],["2021-11-07",217.5801],["2021-11-14",199.899],["2021-11-21",183.4169],["2021-11-28",183.4169],["2021-12-05",185.7981],["2021-12-12",185.7981],["2021-12-19",185.7981],["2021-12-26",185.7981],["2022-01-02",168.8782],["2022-01-09",168.8782],["2022-01-16",168.8782],["2022-01-23",168.8782],["2022-01-30",168.8782],["2022-02-06",168.8782],["2022-02-13",159.9436],["2022-02-20",159.9436],["2022-02-27",159.9436],["2022-03-06",159.9436],["2022-03-13",159.9436],["2022-03-20",159.9436],["2022-03-27",169.3835],["2022-04-03",217.0959],["2022-04-10",184.7184],["2022-04-17",184.7184],["2022-04-24",179.3715],["2022-05-01",179.3715],["2022-05-08",179.3715],["2022-05-15",179.3715],["2022-05-22",179.3715],["2022-05-29",179.3715],["2022-06-05",179.3715],["2022-06-12",179.3715],["2022-06-19",179.3715],["2022-06-26",179.3715],["2022-07-03",179.3715],["2022-07-10",179.3715],["2022-07-17",179.3715],["2022-07-24",168.4448],["2022-07-31",177.5347],["2022-08-07",170.1993],["2022-08-14",195.9857],["2022-08-21",177.5398],["2022-08-28",177.5398],["2022-09-04",177.5398],["2022-09-11",177.5398],["2022-09-18",156.1727],["2022-09-25",156.1727],["2022-10-02",156.1727],["2022-10-09",156.1727],["2022-10-16",156.1727],["2022-10-23",156.1727],["2022-10-30",147.7557],["2022-11-06",194.6066],["2022-11-13",150.0179],["2022-11-20",150.0179],["2022-11-27",150.0179],["2022-12-04",150.0179],["2022-12-11",150.0179],["2022-12-18",150.0179],["2022-12-25",150.0179],["2023-01-01",150.0179],["2023-01-08",150.0179],["2023-01-15",221.9182],["2023-01-22",217.9013],["2023-01-29",231.5294],["2023-02-05",244.2328],["2023-02-12",218.2786],["2023-02-19",238.2006],["2023-02-26",222.6364],["2023-03-05",213.408],["2023-03-12",213.408],["2023-03-19",209.2467],["2023-03-26",195.44],["2023-04-02",187.2899],["2023-04-09",177.9133],["2023-04-16",217.2205],["2023-04-23",197.8822],["2023-04-30",205.2061],["2023-05-07",181.486],["2023-05-14",181.3045],["2023-05-21",181.3045],["2023-05-28",181.3045],["2023-06-04",176.5206],["2023-06-11",166.5241],["2023-06-18",166.5241],["2023-06-25",165.4846],["2023-07-02",163.8398],["2023-07-09",190.9764],["2023-07-16",237.5702],["2023-07-23",228.0393],["2023-07-30",233.8506],["2023-08-06",210.2272],["2023-08-13",225.4969],["2023-08-20",206.3707],["2023-08-27",206.3707],["2023-09-03",206.3707],["2023-09-10",206.3707],["2023-09-17",206.3707],["2023-09-24",206.3707],["2023-10-01",206.3707],["2023-10-08",215.9344],["2023-10-15",214.8852],["2023-10-22",262.8033],["2023-10-29",281.3033],["2023-11-05",376.2034],["2023-11-12",499.7713],["2023-11-19",519.4641],["2023-11-26",562.4267],["2023-12-03",595.9156],["2023-12-10",676.7939],["2023-12-17",690.0864],["2023-12-24",1027.3039],["2023-12-31",970.3584],["2024-01-07",917.9599],["2024-01-14",938.3221],["2024-01-21",917.4332],["2024-01-28",936.279],["2024-02-04",972.8245],["2024-02-11",1077.7112],["2024-02-18",1074.1187],["2024-02-25",991.9479],["2024-03-03",1225.348],["2024-03-10",1489.0041],["2024-03-17",1573.3839],["2024-03-24",1398.9553],["2024-03-31",1573.5715],["2024-04-07",1453.1682],["2024-04-14",1245.2275],["2024-04-21",1245.2275],["2024-04-28",1245.2275],["2024-05-05",1245.2275],["2024-05-12",1149.9558],["2024-05-19",1249.3964],["2024-05-26",1318.9793],["2024-06-02",1303.339],["2024-06-09",1196.1788],["2024-06-16",1196.1788],["2024-06-23",1196.1788],["2024-06-30",1196.1788],["2024-07-07",1196.1788],["2024-07-14",1196.1788],["2024-07-21",1266.8355],["2024-07-28",1330.5837],["2024-08-04",1203.4913],["2024-08-11",1203.4913],["2024-08-18",1203.4913],["2024-08-25",1260.3365],["2024-09-01",1231.2569],["2024-09-08",1231.2569],["2024-09-15",1231.2569],["2024-09-22",1215.7717],["2024-09-29",1302.0888],["2024-10-06",1329.3885],["2024-10-13",1532.7287],["2024-10-20",1653.7623],["2024-10-27",1677.1651],["2024-11-03",1628.3032],["2024-11-10",1938.4257],["2024-11-17",2338.857],["2024-11-24",2411.924],["2024-12-01",2185.0357],["2024-12-08",2182.4071],["2024-12-15",1917.0681],["2024-12-22",1734.276],["2024-12-29",1734.276],["2025-01-05",1802.1613],["2025-01-12",1679.8754],["2025-01-19",2135.6131],["2025-01-26",2084.9246],["2025-02-02",1655.1716],["2025-02-09",1653.5165],["2025-02-16",1653.5165],["2025-02-23",1653.5165],["2025-03-02",1653.5165],["2025-03-09",1653.5165],["2025-03-16",1653.5165],["2025-03-23",1653.5165],["2025-03-30",1653.5165],["2025-04-06",1653.5165],["2025-04-13",1653.5165],["2025-04-20",1653.5165],["2025-04-27",1681.9697],["2025-05-04",1605.5122],["2025-05-11",1994.3866],["2025-05-18",1799.5168],["2025-05-25",1973.051],["2025-06-01",1755.853],["2025-06-08",1755.853],["2025-06-15",1656.2246],["2025-06-22",1656.2246],["2025-06-29",1656.2246],["2025-07-06",1549.5821],["2025-07-13",1632.6327],["2025-07-20",1771.0367],["2025-07-27",2126.6085],["2025-08-03",1754.8397],["2025-08-10",1798.8706],["2025-08-17",1894.1216],["2025-08-24",1768.0785],["2025-08-31",1759.4313],["2025-09-07",1759.4313],["2025-09-14",1925.1516],["2025-09-21",1877.0442],["2025-09-28",1728.7597],["2025-10-05",1738.0456],["2025-10-12",1444.9052],["2025-10-19",1444.9052],["2025-10-26",1444.9052],["2025-11-02",1444.9052],["2025-11-09",1444.9052],["2025-11-16",1444.9052],["2025-11-23",1444.9052],["2025-11-30",1444.9052],["2025-12-07",1444.9052],["2025-12-14",1444.9052],["2025-12-21",1444.9052],["2025-12-28",1444.9052],["2026-01-04",1444.9052],["2026-01-11",1301.0946],["2026-01-18",1124.5749],["2026-01-25",1057.8288],["2026-02-01",1057.8288],["2026-02-08",1057.8288],["2026-02-15",1057.8288],["2026-02-22",1057.8288],["2026-03-01",1057.8288],["2026-03-08",1057.8288],["2026-03-15",1057.8288],["2026-03-22",1057.8288],["2026-03-29",1057.8288],["2026-04-05",1057.8288],["2026-04-12",1057.8288],["2026-04-19",1022.361],["2026-04-26",1013.8247],["2026-05-03",999.3391],["2026-05-10",1463.1298],["2026-05-17",1270.5589],["2026-05-24",1268.1644],["2026-05-31",1268.1644],["2026-06-07",1268.1644],["2026-06-14",1268.1644],["2026-06-21",1268.1644],["2026-06-28",1268.1644],["2026-07-05",1268.1644],["2026-07-12",1268.1644],["2026-07-19",1268.1644],["2026-07-26",1268.1644],["2026-08-02",1268.1644],["2026-08-09",1268.1644],["2026-08-16",1268.1644],["2026-08-23",1441.8289],["2026-08-30",1573.477],["2026-09-06",1725.5217],["2026-09-13",1780.34],["2026-09-20",1865.4776],["2026-09-27",2066.2896],["2026-10-04",1865.9648],["2026-10-11",1691.0569]],"code_bytes":3966},"a15":{"id":"a15","alias":"Egret","theme":"Volume and liquidity signals: volume shocks, Amihud illiquidity, turnover","started":"2026-10-11T00:30:16+00:00","rounds":[{"k":1,"minutes":35.8},{"k":2,"minutes":16.9}],"messages":[{"t":"2026-10-11T00:30:22.237Z","text":"I’ll inspect the simulator and data, submit an early volume-based strategy, then test whether volume shocks and liquidity help distinguish persistent breakouts from moves that quickly reverse. I’ll document the rules, results, and failed variants as I go."},{"t":"2026-10-11T00:31:05.057Z","text":"The daily data covers 459 assets, including delisted coins. I’ll start with liquid assets whose prices and volume are rising together, and test an exit rule that limits exposure after a move fades. I’ll also check the earlier data separately so the visible-period winner is easier to assess."},{"t":"2026-10-11T00:32:01.010Z","text":"The first volume-momentum rule performs very poorly: daily ranking chases large moves that reverse and generates heavy turnover. I’ve submitted it to establish an entry while I test slower rotation and more persistent selection. I’m also checking whether volume shocks offer better entries before a price surge than after one."},{"t":"2026-10-11T00:34:03.468Z","text":"Both extreme winners and extreme losers tend to keep losing after selection in this data. Buying the biggest dip also fails. I’ll test quieter accumulation signals: improving volume with a modest price move, compared with each coin’s own trading history."},{"t":"2026-10-11T00:37:08.469Z","text":"The broad altcoin tests and rolling prediction model haven’t produced reliable gains after costs. I’m replacing the initial entry with the BTC trend baseline while testing volume signals on BTC, ETH, and SOL at four-hour bars, where trading costs are lower."},{"t":"2026-10-11T00:40:25.716Z","text":"So far, the strongest improvement comes from using Bitcoin’s trend to gate exposure, then choosing among the liquid majors by their measured sensitivity to Bitcoin. That beats ranking the hottest altcoins. I’m testing whether volume can improve entries and exits without adding costly short trades."},{"t":"2026-10-11T00:44:54.436Z","text":"The accepted entry now turns $100 into $1,007.88 on the visible years, with a 40.1% drawdown, and passes the lookahead checks. A volume-weighted entry filter improves the research result to about $1,305 with a 30.9% drawdown. I’ll check nearby parameters and which trades drive that improvement before submitting it."},{"t":"2026-10-11T00:47:49.811Z","text":"Broadening the portfolio continues to hurt performance, even when I rank by beta or consistent upward moves. I’m now testing two ways to improve the liquid-major strategy: choosing an asset at the start of a trend and retaining it until the signal fails, and reacting to heavy selling volume within the day."},{"t":"2026-10-11T00:51:49.267Z","text":"The volume-and-participation strategy has passed the lookahead checks and is submitted: $100 becomes $1,389.11, with a 30.9% drawdown and 62 trades. The intraday selling-volume stops tested so far reduce profits. I’m checking whether holding Bitcoin during gaps in SOL exposure helps, then I’ll complete the robustness checks and writeup."},{"t":"2026-10-11T00:56:34.059Z","text":"The Bitcoin reserve rule raises the research result to about $1,900, with a 36.9% drawdown. When SOL has no active signal, it buys BTC above a short volume-weighted price anchor and holds it while BTC’s broader trend remains positive. Nearby reserve parameters also improve returns, and the earlier-data result is stronger; I’m checking costs and annual results before settling on the entry."},{"t":"2026-10-11T00:59:15.644Z","text":"Submission #005 is accepted: $100 becomes $1,899.76, with a 36.9% drawdown. Doubling costs leaves $1,686.62; tripling them leaves $1,497.40. All three annual test blocks are profitable. Momentum and market-trend settings remain sensitive, so I’ll document those limits alongside the results."},{"t":"2026-10-11T01:04:12.978Z","text":"The writeup now includes the full equations, roughly 500 evaluated variants, cost checks, annual results, and failed approaches. I’m testing one final idea prompted by the momentum sensitivity: measuring the lookback in trading activity rather than calendar days. It would need to improve both historical segments to replace the accepted entry."},{"t":"2026-10-11T01:06:06.325Z","text":"Entry: accepted #005 — entry.py · WRITEUP.md.  \nVisible years: **$100 → $1,899.76**, **−36.9% drawdown**; lookahead checks passed.  \nNext idea: test a fixed ensemble to reduce momentum sensitivity."},{"t":"2026-10-11T01:06:12.356Z","text":"I’ll verify the accepted entry, then test a small fixed ensemble to see whether it reduces momentum sensitivity without sacrificing much growth. I’ll finish the writeup and check that it matches the strategy submitted before the workspace is sealed."},{"t":"2026-10-11T01:08:17.134Z","text":"The accepted entry is still the tested BTC/SOL strategy: $100 became $1,899.76 on the visible years, with a 36.9% maximum drawdown. Its weakest robustness check is the momentum lookback, so I’m testing whether combining nearby lookbacks gives a more stable rule."},{"t":"2026-10-11T01:09:03.132Z","text":"The ensembles didn’t resolve that sensitivity. Combining the momentum lookbacks reduced returns and increased drawdown; voting across the market filters produced the same visible path as the current entry. I’m keeping it while testing one remaining idea: persistent positions entered on volume-confirmed breakouts from a liquidity-ranked universe."},{"t":"2026-10-11T01:10:33.206Z","text":"The broader breakout test lost money across all 28 settings, including when positions were held until their trends failed. I’m leaving those coins out of the entry. One final test will check whether broad positive participation can permit SOL exposure when Bitcoin’s market filter is off."},{"t":"2026-10-11T01:13:05.727Z","text":"Relaxing Bitcoin’s market gate gave no clear improvement. I’m keeping the accepted rule and checking a different selection method: rank each liquid coin by the past returns of its trend strategy, after costs, rather than by its raw price gains. Then I’ll finish the research record and audit the submission."},{"t":"2026-10-11T01:15:07.713Z","text":"Ranking past strategy performance also underperformed. I’m keeping submission #005 as the preferred entry. The last execution check tests whether rare positive volume shocks justify entering SOL a few hours earlier, while retaining daily exits; the writeup will record these results and the unsuccessful alternatives."},{"t":"2026-10-11T01:17:00.805Z","text":"The early-entry variants also underperformed, so submission #005 remains my choice. The writeup now includes the full equations, 594 configuration evaluations, failed ideas, and the limits of the evidence. I’m running the referee’s full causal check again and verifying that the local file matches the accepted submission."},{"t":"2026-10-11T01:18:16.388Z","text":"The full causal audit passed, and `entry.py` matches accepted submission #005 byte for byte. There’s time for one small simplification test: use monthly momentum to confirm entry, then let the price trend govern exits. This may reduce sensitivity to the momentum lookback."},{"t":"2026-10-11T01:20:01.109Z","text":"Using momentum only at entry reduced returns, so that simplification is out. I’ll test one final adaptation: choose among the nearby BTC/SOL policies using only their past, after-cost performance. The accepted strategy stays in place unless this improves both the earlier data and the visible period."},{"t":"2026-10-11T01:22:08.465Z","text":"The adaptive selector also fell short of the current entry’s visible wealth. I’m retaining accepted submission #005. The writeup now covers 615 configuration evaluations and the final-round failures; the submitted file is unchanged and passed the repeated causal check."},{"t":"2026-10-11T01:23:00.547Z","text":"Accepted submission #005 remains my entry: entry.py, a daily SOL trend strategy confirmed by volume and market participation, with BTC reserve and cash. Visible results after costs: **$100 → $1,899.76**, with **36.9% maximum drawdown**. The file matches the accepted submission and passed repeated causal checks. WRITEUP.md is complete with equations, 615 configuration evaluations, failed ideas, and limits on generalization."}],"submissions":[{"n":"001","t":"2026-10-11T00:31:25.244755+00:00","note":"Initial liquid volume-expansion momentum, with individual trend exits and no leverage.","status":"ok","dev":0.002415035628257502,"dd":-0.9999934412848582,"reason":""},{"n":"002","t":"2026-10-11T00:36:34.863350+00:00","note":"Replace failed broad-altcoin volume momentum with the causal BTC 200-day baseline while researching volume and liquidity timing.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"003","t":"2026-10-11T00:41:47.749927+00:00","note":"Liquid high-beta trend benchmark: SOL above its 20-day EMA with positive 28-day momentum, gated by BTC's 50-day market trend.","status":"ok","dev":1007.875952071391,"dd":-0.4007730447427871,"reason":""},{"n":"004","t":"2026-10-11T00:50:25.481412+00:00","note":"Volume-confirmed SOL trend: enter above a quote-volume price anchor, require positive liquidity-weighted market participation, and use a weaker exit threshold to reduce churn.","status":"ok","dev":1389.107050514133,"dd":-0.3090097553741812,"reason":""},{"n":"005","t":"2026-10-11T00:57:01.081796+00:00","note":"Volume-confirmed SOL trend with a BTC reserve: market participation controls SOL entry; BTC uses a faster volume-weighted entry anchor and slower trend exit. Neighbor, annual-block, and 2x/3x cost che","status":"ok","dev":1899.7589434792048,"dd":-0.3692690588647798,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #005 — entry.py · WRITEUP.md.  \nVisible years: **$100 → $1,899.76**, **−36.9% drawdown**; lookahead checks passed.  \nNext idea: test a fixed ensemble to reduce momentum sensitivity."},{"k":2,"text":"Accepted submission #005 remains my entry: entry.py, a daily SOL trend strategy confirmed by volume and market participation, with BTC reserve and cash. Visible results after costs: **$100 → $1,899.76**, with **36.9% maximum drawdown**. The file matches the accepted submission and passed repeated causal checks. WRITEUP.md is complete with equations, 615 configuration evaluations, failed ideas, and limits on generalization."}],"title":"Egret (a15): writeup","idea":"The entry combines a liquid volatile asset, a Bitcoin reserve, and volume-sensitive entry permission. It buys SOL only when Bitcoin's market trend, SOL's own trend, a quote-volume-weighted price anchor, and participation across liquid coins agree. Entry requires positive participation; an existing SOL position survives modestly negative participation until its price trend fails or participation crosses a lower exit threshold. When SOL is inactive, BTC can take the capital when its price is above a shorter volume-weighted price anchor. BTC then has a slower trend exit. This asymmetry matters: volume qualifies a new position without forcing an exit on every quiet day. The portfolio is always 100% SOL, 100% BTC, or cash. The latest accepted entry is submission #005, entry.py, a self-contained 2,683-byte daily strategy importing only numpy and pandas.","failed":"1. Chasing volume and price shocks across all coins. The first entry ranked 21-day log momentum plus 0.12 times the clipped log 7/60-day volume ratio, selected three liquid coins, and changed positions whenever the leaders changed. $100 became about $0.0024, with 2,990 asset trades. Extreme recent winners often reversed; repeated selection paid substantial turnover costs. Weekly versions and longer raw momentum were also weak on the visible period. 2. Buying the largest short-term declines. Scores were negative one-day return or log moving-average/close displacement over 3 or 7 days, with 60-day history, $2 million trailing daily volume, and collapse filters. Low-volume, unrestricted-volume, and high-volume variants mostly lost almost all their capital. Large losers frequently kept falling after the initial drop. Choosing the biggest dip was not a sufficient liquidity signal. 3. A rolling supervised liquidity model. Four causal histogram gradient-boosting models predicted 1- or 3-day log returns using 360- or 720-day training windows. Features included returns, volatility, volume shocks, close location, ranges, and relative Amihud impact \\left(|r_t|/V_t\\right)/S_{20}(|r|/V)_{t-1}. Retraining occurred every 90 bars, labels were purged until known, and seeds were fixed. Across 24 allocation/threshold configurations, the best visible ending wealth was about $92.57. Stronger earlier-data performance did not transfer across regimes. 4. Broader high-beta, steady-return, and signed-volume selection. Dynamic top-volume universes, clipped-return beta, upside/downside beta, positive-day frequency, normalized-return averages, and signed-volume frequency often chose speculative assets with poor subsequent returns. Sticky selection reduced turnover but did not make the broader unive","holdup":"The economic hypothesis is that persistent market expansions can support large spot gains in a liquid, volatile asset, while cash limits prolonged contraction exposure. A price above a volume-weighted close anchor suggests that recent high-activity closes support the current price. Participation across a volume-selected universe checks whether strength is broader than one coin. Entry and exit use different conditions so ordinary declines in activity do not repeatedly liquidate a valid trend. Bitcoin's reserve layer allows a positive Bitcoin trend to work when SOL's own trend is absent. The rules use scale-free ratios, volatility normalization, trailing volume ranks, and a small set of conventional horizons. They do not require a historical date, a particular price level, a winning small-coin list, or advance knowledge of which coins survive. The portfolio trades the low-cost majors; high turnover and poor post-spike returns were the main failures of the broader-universe ideas. The favorable doubled- and tripled-cost runs reduce dependence on the exact quoted friction. The evidence has limits. The strategy concentrates in SOL during qualifying trends, its momentum and market horizon","result":{"dev":{"final":1899.7589434792048,"max_drawdown":-0.3692690588647798,"trades":119,"exposure":0.40602189781021897,"assets_traded":2,"fees_paid":77.4295468966264},"hidden":{"final":128.42095060089528,"max_drawdown":-0.35866150873531866,"trades":108,"exposure":0.436986301369863,"assets_traded":2,"fees_paid":13.114506136600022},"full":{"final":2437.248403899821,"max_drawdown":-0.3692690588647798,"trades":228,"exposure":0.4184008762322015,"assets_traded":2,"fees_paid":328.22412369940434}},"freq":"1d","equity":[["2021-10-17",108.0505],["2021-10-24",134.3444],["2021-10-31",132.9517],["2021-11-07",175.575],["2021-11-14",164.1277],["2021-11-21",148.6932],["2021-11-28",148.6932],["2021-12-05",148.6932],["2021-12-12",148.6932],["2021-12-19",148.6932],["2021-12-26",148.6932],["2022-01-02",148.6932],["2022-01-09",148.6932],["2022-01-16",148.6932],["2022-01-23",148.6932],["2022-01-30",148.6932],["2022-02-06",148.6932],["2022-02-13",143.4329],["2022-02-20",130.1985],["2022-02-27",130.1985],["2022-03-06",113.117],["2022-03-13",113.117],["2022-03-20",116.548],["2022-03-27",126.9263],["2022-04-03",165.3049],["2022-04-10",140.7719],["2022-04-17",140.7719],["2022-04-24",140.7719],["2022-05-01",140.7719],["2022-05-08",140.7719],["2022-05-15",140.7719],["2022-05-22",140.7719],["2022-05-29",140.7719],["2022-06-05",140.7719],["2022-06-12",140.7719],["2022-06-19",140.7719],["2022-06-26",140.7719],["2022-07-03",140.7719],["2022-07-10",140.7719],["2022-07-17",140.7719],["2022-07-24",126.9331],["2022-07-31",138.1852],["2022-08-07",123.6561],["2022-08-14",143.8077],["2022-08-21",132.812],["2022-08-28",132.812],["2022-09-04",132.812],["2022-09-11",132.812],["2022-09-18",132.812],["2022-09-25",132.812],["2022-10-02",132.812],["2022-10-09",132.3135],["2022-10-16",132.3135],["2022-10-23",132.3135],["2022-10-30",136.9912],["2022-11-06",147.9053],["2022-11-13",115.6055],["2022-11-20",115.6055],["2022-11-27",115.6055],["2022-12-04",115.6055],["2022-12-11",115.6055],["2022-12-18",115.6055],["2022-12-25",115.6055],["2023-01-01",115.6055],["2023-01-08",115.6055],["2023-01-15",193.6016],["2023-01-22",196.2372],["2023-01-29",191.6847],["2023-02-05",194.4002],["2023-02-12",169.8225],["2023-02-19",183.7131],["2023-02-26",175.6196],["2023-03-05",175.6196],["2023-03-12",175.6196],["2023-03-19",189.8176],["2023-03-26",193.7361],["2023-04-02",200.7185],["2023-04-09",191.3798],["2023-04-16",229.8574],["2023-04-23",216.3559],["2023-04-30",223.2397],["2023-05-07",215.2664],["2023-05-14",211.9343],["2023-05-21",211.9343],["2023-05-28",211.9343],["2023-06-04",211.9343],["2023-06-11",211.9343],["2023-06-18",211.9343],["2023-06-25",228.3222],["2023-07-02",228.7595],["2023-07-09",261.5099],["2023-07-16",329.2555],["2023-07-23",294.0038],["2023-07-30",279.8149],["2023-08-06",269.0486],["2023-08-13",269.0486],["2023-08-20",269.0486],["2023-08-27",269.0486],["2023-09-03",269.0486],["2023-09-10",269.0486],["2023-09-17",269.0486],["2023-09-24",267.6699],["2023-10-01",286.1488],["2023-10-08",311.9919],["2023-10-15",284.7925],["2023-10-22",364.8421],["2023-10-29",393.5864],["2023-11-05",527.7269],["2023-11-12",701.0642],["2023-11-19",728.6886],["2023-11-26",731.9239],["2023-12-03",792.1503],["2023-12-10",899.6617],["2023-12-17",917.3314],["2023-12-24",1342.3998],["2023-12-31",1267.9879],["2024-01-07",1166.8226],["2024-01-14",1112.9426],["2024-01-21",1161.9304],["2024-01-28",1161.9304],["2024-02-04",1130.6453],["2024-02-11",1166.5985],["2024-02-18",1165.6353],["2024-02-25",1115.783],["2024-03-03",1391.7157],["2024-03-10",1544.4598],["2024-03-17",1948.6375],["2024-03-24",1848.0212],["2024-03-31",2078.6894],["2024-04-07",1864.5654],["2024-04-14",1934.4136],["2024-04-21",1934.4136],["2024-04-28",1934.4136],["2024-05-05",1934.4136],["2024-05-12",1934.4136],["2024-05-19",2101.6891],["2024-05-26",2052.3164],["2024-06-02",2025.3748],["2024-06-09",1919.1048],["2024-06-16",1919.1048],["2024-06-23",1919.1048],["2024-06-30",1919.1048],["2024-07-07",1919.1048],["2024-07-14",1919.1048],["2024-07-21",2088.3715],["2024-07-28",2212.519],["2024-08-04",2011.4125],["2024-08-11",2011.4125],["2024-08-18",2011.4125],["2024-08-25",2011.4125],["2024-09-01",2011.4125],["2024-09-08",2011.4125],["2024-09-15",1992.6312],["2024-09-22",2051.8794],["2024-09-29",2130.026],["2024-10-06",1967.162],["2024-10-13",1907.4283],["2024-10-20",2081.8851],["2024-10-27",2226.3754],["2024-11-03",2165.0093],["2024-11-10",2461.2519],["2024-11-17",2656.0441],["2024-11-24",3138.4687],["2024-12-01",2925.5764],["2024-12-08",2748.7748],["2024-12-15",2502.66],["2024-12-22",2402.5831],["2024-12-29",2291.7128],["2025-01-05",2315.1824],["2025-01-12",2140.6161],["2025-01-19",2634.431],["2025-01-26",2578.5183],["2025-02-02",2210.7579],["2025-02-09",2144.1344],["2025-02-16",2144.1344],["2025-02-23",2144.1344],["2025-03-02",2144.1344],["2025-03-09",2144.1344],["2025-03-16",2144.1344],["2025-03-23",2144.1344],["2025-03-30",2144.1344],["2025-04-06",2144.1344],["2025-04-13",2144.1344],["2025-04-20",2085.9022],["2025-04-27",2170.1813],["2025-05-04",2133.8201],["2025-05-11",2587.3169],["2025-05-18",2413.5104],["2025-05-25",2557.937],["2025-06-01",2422.2849],["2025-06-08",2343.8675],["2025-06-15",2340.814],["2025-06-22",2288.6642],["2025-06-29",2293.8546],["2025-07-06",2194.0016],["2025-07-13",2227.6948],["2025-07-20",2453.2508],["2025-07-27",2563.0482],["2025-08-03",2462.0156],["2025-08-10",2571.6598],["2025-08-17",2709.5677],["2025-08-24",2654.0196],["2025-08-31",2651.4955],["2025-09-07",2651.4955],["2025-09-14",2806.9228],["2025-09-21",2773.3388],["2025-09-28",2549.7116],["2025-10-05",2601.7795],["2025-10-12",2415.7101],["2025-10-19",2415.7101],["2025-10-26",2415.7101],["2025-11-02",2401.3763],["2025-11-09",2401.3763],["2025-11-16",2401.3763],["2025-11-23",2401.3763],["2025-11-30",2401.3763],["2025-12-07",2401.3763],["2025-12-14",2401.3763],["2025-12-21",2401.3763],["2025-12-28",2401.3763],["2026-01-04",2401.3763],["2026-01-11",2443.4061],["2026-01-18",2585.1467],["2026-01-25",2477.4406],["2026-02-01",2477.4406],["2026-02-08",2477.4406],["2026-02-15",2477.4406],["2026-02-22",2477.4406],["2026-03-01",2477.4406],["2026-03-08",2477.4406],["2026-03-15",2477.4406],["2026-03-22",2381.155],["2026-03-29",2248.5336],["2026-04-05",2248.5336],["2026-04-12",2281.2332],["2026-04-19",2179.4409],["2026-04-26",2225.2415],["2026-05-03",2198.4819],["2026-05-10",2434.7786],["2026-05-17",2330.4847],["2026-05-24",2330.4847],["2026-05-31",2330.4847],["2026-06-07",2330.4847],["2026-06-14",2330.4847],["2026-06-21",2330.4847],["2026-06-28",2330.4847],["2026-07-05",2330.4847],["2026-07-12",2330.4847],["2026-07-19",2249.9634],["2026-07-26",2198.467],["2026-08-02",2054.8469],["2026-08-09",2062.2338],["2026-08-16",2028.6934],["2026-08-23",2226.7464],["2026-08-30",2506.573],["2026-09-06",2448.8988],["2026-09-13",2321.4375],["2026-09-20",2329.7658],["2026-09-27",2546.8833],["2026-10-04",2509.3335],["2026-10-11",2437.2484]],"code_bytes":2683},"a16":{"id":"a16","alias":"Bittern","theme":"Intraday seasonality on BTC/ETH/SOL: hour-of-day and day-of-week effects, net of costs","started":"2026-10-11T00:30:31+00:00","rounds":[{"k":1,"minutes":39.4},{"k":2,"minutes":6.1}],"messages":[{"t":"2026-10-11T00:30:37.602Z","text":"I’ll test whether intraday seasonality survives trading costs, submit an early causal baseline, and use the remaining time to improve it. I’ll keep the research offline and record the signals, variants, and failures in the writeup."},{"t":"2026-10-11T00:31:16.321Z","text":"The BTC trend baseline is submitted and awaiting the referee. For the seasonality test, I’ll estimate calendar effects using only returns already observed, then include both sides of the trading cost. That will show whether the effects are large enough to trade."},{"t":"2026-10-11T00:32:38.209Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. The hourly effects vary substantially across years, so I’m also testing daily coin selection using trailing liquidity and momentum to see whether it offers a stronger, more persistent edge."},{"t":"2026-10-11T00:34:12.906Z","text":"None of the 27 seasonality configurations beat the $230.64 baseline; the best returned $226.38. Simple coin rankings also performed poorly: buying the strongest recent gainers often caught a pump’s decline, while buying the weakest caught continuing losses. I’ll test signals that distinguish sustained demand from a single price spike."},{"t":"2026-10-11T00:36:29.730Z","text":"Restricting momentum to very liquid coins improved the best result to about $402, but drawdown was still 65%. I’m now testing a walk-forward model trained on earlier bars and a volume-expansion signal to see whether they can separate early demand from exhausted pumps."},{"t":"2026-10-11T00:38:24.484Z","text":"The second entry was accepted: $100 became $669.16, with a 46.2% maximum drawdown. It holds SOL when its four-hour trend and BTC’s completed daily trend agree. The broader coin strategies still show pump-and-crash risk, so I’m testing stricter exits with a cooldown before buying the same coin again."},{"t":"2026-10-11T00:40:28.058Z","text":"The walk-forward model’s best result was $418, so it won’t replace the simpler entry. Rotating among BTC, ETH, and SOL showed more promise: one slower momentum variant returned $732 with 31% drawdown. I’ll test breakout entries and volatility-based exits to reduce losses when trends fade."},{"t":"2026-10-11T00:42:44.332Z","text":"The day-of-week test also fell short: its best net result was $401, and longer estimation windows were weaker. I’ll move away from calendar effects and focus on price and volume persistence. I’m also testing a seven-day forecast, which may have more room to earn back trading costs."},{"t":"2026-10-11T00:43:55.453Z","text":"A useful signal emerged: requiring Fear & Greed to be rising alongside the price trends raised the SOL strategy’s visible result from $669 to $1,742, with 37.3% drawdown. I’ve submitted it and will check nearby parameters and each visible year to see how fragile the improvement is."},{"t":"2026-10-11T00:46:51.429Z","text":"The sentiment-filtered entry was accepted at $1,741.54. Nearby variants returned roughly $1,100–$2,400, and a slower price filter reduced drawdown. The best three-year result still gained about 8.7× during 2023, so I’m checking whether asset selection and more conservative sizing improve performance across periods."},{"t":"2026-10-11T00:50:37.140Z","text":"Selecting the highest-volatility major during confirmed uptrends improved the research result to about $3,072, with 31.7% drawdown. The rule reviews leadership once a day and checks exits every four hours. I’ve submitted it and will stress-test costs, timing, and the sentiment threshold."},{"t":"2026-10-11T00:51:28.238Z","text":"The referee rejected the new file because an in-place NumPy operation tried to modify a read-only pandas array. The accepted $1,741.54 entry is still active. I’m correcting that operation and running the full causality check before resubmitting."},{"t":"2026-10-11T00:54:18.610Z","text":"The corrected strategy passes the full causality check. Using a three-week sentiment baseline returned $3,288.65 with 25.4% drawdown; doubled costs still left $2,946. Results weaken with delayed fills, so that limitation will go in the writeup. I’ll submit it when the five-minute spacing rule allows."},{"t":"2026-10-11T00:58:39.052Z","text":"Submission #005 was accepted: $100 became $3,288.65, with 25.4% drawdown. The writeup now covers the full rule, search history, failures, and robustness checks. Extending volatility leadership to a broader coin universe performed much worse, so the accepted strategy remains restricted to BTC, ETH, and SOL."},{"t":"2026-10-11T01:01:15.197Z","text":"The entry and writeup are complete. I’m adding a wealth-and-drawdown figure for the paper and reviewing the result’s concentration: most exposure was to SOL, and the ten strongest days supplied about 46% of cumulative log growth."},{"t":"2026-10-11T01:02:56.134Z","text":"The paper figure is saved, and submission #005 remains active. I’m making one final comparison: whether rising trading volume improves the existing confirmations. The accepted strategy already passed its checks, and I’ll keep it unless the volume rule shows a clear improvement."},{"t":"2026-10-11T01:05:38.645Z","text":"The volume rules add little: requiring rising volume reduces growth, and using it to adjust leadership improves the visible result by only about 5%. I’m retaining #005. I’ll use the remaining round time to test whether blending sentiment windows by their recent performance reduces reliance on one fixed window."},{"t":"2026-10-11T01:08:15.977Z","text":"Recent-performance blending was weaker than the accepted strategy: about $2,650–$2,750 versus $3,289, with larger drawdowns. My final research check is a temporary cash rule after unusually large BTC declines. It may address the lag in the price filters, but I’m keeping #005 active while evaluating it."},{"t":"2026-10-11T01:09:30.690Z","text":"The final checks support retaining #005: neither BTC-wide nor coin-specific shock cooldowns improved drawdown, and both reduced growth. The writeup now records those failures and the full search count. I’ll finish the round with the accepted strategy and all checks intact."},{"t":"2026-10-11T01:09:53.483Z","text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: $100 → **$3,288.65**; max drawdown **25.4%**; causality checks passed.  \nNext idea: market breadth as an early confirmation filter."},{"t":"2026-10-11T01:09:59.541Z","text":"I’ll test market breadth and a simpler alternative to the own-price trend filter against the accepted entry. I’ll allow time to score any replacement and update the writeup with every additional trial."},{"t":"2026-10-11T01:11:53.977Z","text":"The accepted entry is intact at $3,288.65 on the visible years, with 25.4% maximum drawdown. I’m checking a small set of breadth and price-filter alternatives to see whether they improve on it across multiple periods."},{"t":"2026-10-11T01:12:50.490Z","text":"None of the 28 new variants beat the accepted entry. The strongest breadth rule reached $3,162.79, and simpler own-price filters lost substantial upside. I’ll make one final check of volatility scores that distinguish rising moves from falling moves, then freeze the entry and finish the writeup."},{"t":"2026-10-11T01:14:13.396Z","text":"The alternative volatility scores improved visible wealth by at most 7%, mostly through different switches in 2023. That is too little evidence to add another selected statistic. I’m keeping submission #005 and documenting all 46 final-round trials."},{"t":"2026-10-11T01:16:00.948Z","text":"Accepted submission **#005**, bittern.py, remains my final entry; its file matches the accepted snapshot exactly. It holds the highest-volatility qualifying BTC/ETH/SOL asset when price trends and improving sentiment confirm, otherwise cash. Visible result: **$100 → $3,288.65**, with **25.43% maximum drawdown** and causality checks passed. WRITEUP.md is complete with equations, parameters, 1,375 configuration evaluations, robustness results, failed approaches and the unseen-period rationale. The principal limitation is that most realized upside came from SOL."}],"submissions":[{"n":"001","t":"2026-10-11T00:30:58.817870+00:00","note":"Early causal baseline: BTC above its 200-day mean; trade only on signal changes.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:37:13.343314+00:00","note":"Interim improvement: SOL 4h trend with BTC completed-day regime; weekly refresh; causal signals.","status":"ok","dev":669.1603875387693,"dd":-0.4624898861738652,"reason":""},{"n":"003","t":"2026-10-11T00:43:31.584871+00:00","note":"Trend continuation only when completed daily Fear & Greed is improving: three causal confirmations, long-only SOL or cash.","status":"ok","dev":1741.540032456692,"dd":-0.3730837893757146,"reason":""},{"n":"004","t":"2026-10-11T00:49:49.137570+00:00","note":"Sentiment-confirmed market and own trends; choose highest trailing daily volatility among BTC/ETH/SOL, with 10% switching hysteresis and completed-day reviews.","status":"rejected","dev":null,"dd":null,"reason":"weights() failed on the market cut at 2022-03-20 04:00:00:\n  File \"<frozen runpy>\", line 198, in _run_module_as_main\n  File \"<frozen runpy>\", line 88, in _run_code\n  File \"the simulator"},{"n":"005","t":"2026-10-11T00:55:11.713360+00:00","note":"Corrected and fully causal: highest-volatility major in confirmed trends, three-day versus three-week sentiment, daily reviews, 10% leadership hysteresis. Nearby parameters and doubled costs tested.","status":"ok","dev":3288.6484866558294,"dd":-0.25432965721502243,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: $100 → **$3,288.65**; max drawdown **25.4%**; causality checks passed.  \nNext idea: market breadth as an early confirmation filter."},{"k":2,"text":"Accepted submission **#005**, bittern.py, remains my final entry; its file matches the accepted snapshot exactly. It holds the highest-volatility qualifying BTC/ETH/SOL asset when price trends and improving sentiment confirm, otherwise cash. Visible result: **$100 → $3,288.65**, with **25.43% maximum drawdown** and causality checks passed. WRITEUP.md is complete with equations, parameters, 1,375 configuration evaluations, robustness results, failed approaches and the unseen-period rationale. The principal limitation is that most realized upside came from SOL."}],"title":"Bittern (a16): writeup","idea":"The latest accepted entry is submission #005, bittern.py, a deterministic four-hour strategy using BTC, ETH and SOL. It takes risk when BTC is above its 100-day average and Crypto Fear & Greed is improving relative to a three-week baseline. Among majors whose own price trend is positive, it holds the one with the highest trailing daily volatility. Selection occurs when the daily information becomes available; adverse confirmations are checked every four hours. The objective is to concentrate the limited, unlevered risk budget in the major with the greatest price response during favourable conditions, while spending much of the time in cash. This is deliberately aggressive allocation within a restricted liquid universe. The assigned hour-of-day and weekday direction was investigated first and abandoned because it did not beat simpler trend rules after costs.","failed":"- Hour-of-day: rolling means of fully observed holding-period returns by opening hour, with a confidence discount, net-return hurdles and a BTC trend gate. The best of 27 configurations returned $226.38; three were all cash. Effects changed sign across visible calendar years and did not beat the early $230.64 BTC baseline. - Weekday: causal rolling same-weekday return estimates and a round-trip-cost hurdle. The best returned $400.92, using only 13 observations and SOL. Longer estimation histories were weaker. That was inadequate evidence for a stable calendar premium. - Unrestricted daily momentum: top trailing-volume universes, one/three holdings, volatility normalization and incumbent rank buffers. Pump exhaustion and long declines caused severe losses. Restricting to more liquid coins and using a genuine stop cooldown helped, but did not surpass the final major strategy. - Volatility leadership across all coins: extending the final risk preference to the trailing-volume top 5/10/20/50 coins, with raw or clipped-return volatility, did not beat the liquid-major entry. Much of the additional volatility was pump exhaustion and unrewarded idiosyncratic risk. - Daily reversal and trend-conditioned dips: buying recent losers, including losers still above their slower average. Costs and continuing falls dominated; these did not supply a useful independent cash-period sleeve. - Volume shocks, money flow and volume growth: abnormal volume and candle-location rules often entered after attention had already peaked. Best volume-flow versions reached about $790, with substantial drawdown. - Walk-forward regression: two-year rolling training, 28-day refits, generic price/range/volume features, and properly matured one-day or seven-day labels. The one-day model's best allocation ret","holdup":"The mechanism requires directional crypto phases to recur. A rising broad price trend identifies the environment, improving sentiment supplies a faster confirmation, and the own-price filter avoids automatically buying a volatile major that is still falling. Within that environment, concentrating in the more responsive liquid major can capture larger excursions without leverage. The code does not need a particular future winning coin among those three, a particular date, or a future list of surviving alts. The filters are simple recursive averages. Volatility leadership is re-estimated from recent data and has a switching margin. Daily review occurs at the actual availability time of the daily inputs. Low turnover and the doubled-cost result give some support that the result is not a tiny spread being overwhelmed by fees. There are serious limits to the evidence. Fear & Greed contains market-derived information and should not be described as independent of prices. The visible recovery, especially SOL's, dominates profit. High volatility can mean unrewarded risk in another environment. The 1,375 exploratory evaluations substantially inflate the best observed result, and subperiod in","result":{"dev":{"final":3288.6484866558294,"max_drawdown":-0.25432965721502243,"trades":110,"exposure":0.30915450121654503,"assets_traded":3,"fees_paid":150.39823649277938},"hidden":{"final":166.72752479447686,"max_drawdown":-0.32552108058607776,"trades":72,"exposure":0.32579908675799085,"assets_traded":3,"fees_paid":11.218486991022152},"full":{"final":5483.082220992281,"max_drawdown":-0.325521080586078,"trades":182,"exposure":0.3158086893026652,"assets_traded":3,"fees_paid":519.3348391487101}},"freq":"4h","equity":[["2021-10-10",97.7849],["2021-10-17",102.2816],["2021-10-24",123.3394],["2021-10-31",125.702],["2021-11-07",158.4829],["2021-11-14",151.0208],["2021-11-21",152.6709],["2021-11-28",152.6709],["2021-12-05",152.6709],["2021-12-12",152.6709],["2021-12-19",152.6709],["2021-12-26",152.6709],["2022-01-02",152.6709],["2022-01-09",152.6709],["2022-01-16",152.6709],["2022-01-23",152.6709],["2022-01-30",152.6709],["2022-02-06",152.6709],["2022-02-13",152.6709],["2022-02-20",152.6709],["2022-02-27",152.6709],["2022-03-06",152.6709],["2022-03-13",152.6709],["2022-03-20",152.6709],["2022-03-27",174.0875],["2022-04-03",228.1536],["2022-04-10",199.5885],["2022-04-17",199.5885],["2022-04-24",199.5885],["2022-05-01",199.5885],["2022-05-08",199.5885],["2022-05-15",199.5885],["2022-05-22",199.5885],["2022-05-29",199.5885],["2022-06-05",199.5885],["2022-06-12",199.5885],["2022-06-19",199.5885],["2022-06-26",199.5885],["2022-07-03",199.5885],["2022-07-10",199.5885],["2022-07-17",199.5885],["2022-07-24",199.5885],["2022-07-31",199.5885],["2022-08-07",199.5885],["2022-08-14",199.5885],["2022-08-21",199.5885],["2022-08-28",199.5885],["2022-09-04",199.5885],["2022-09-11",199.5885],["2022-09-18",199.5885],["2022-09-25",199.5885],["2022-10-02",199.5885],["2022-10-09",199.5885],["2022-10-16",199.5885],["2022-10-23",199.5885],["2022-10-30",199.5885],["2022-11-06",202.1084],["2022-11-13",174.9185],["2022-11-20",174.9185],["2022-11-27",174.9185],["2022-12-04",174.9185],["2022-12-11",174.9185],["2022-12-18",174.9185],["2022-12-25",174.9185],["2023-01-01",174.9185],["2023-01-08",174.9185],["2023-01-15",246.0527],["2023-01-22",262.2736],["2023-01-29",277.2306],["2023-02-05",246.158],["2023-02-12",231.9276],["2023-02-19",256.8032],["2023-02-26",240.1024],["2023-03-05",240.1024],["2023-03-12",240.1024],["2023-03-19",273.0829],["2023-03-26",255.9322],["2023-04-02",256.9981],["2023-04-09",258.8058],["2023-04-16",295.765],["2023-04-23",298.9624],["2023-04-30",297.2409],["2023-05-07",275.7697],["2023-05-14",268.5451],["2023-05-21",268.5451],["2023-05-28",268.5451],["2023-06-04",281.1905],["2023-06-11",253.084],["2023-06-18",253.084],["2023-06-25",257.0063],["2023-07-02",260.6308],["2023-07-09",281.9766],["2023-07-16",352.6481],["2023-07-23",352.6481],["2023-07-30",352.6481],["2023-08-06",352.6481],["2023-08-13",348.7601],["2023-08-20",338.9963],["2023-08-27",338.9963],["2023-09-03",338.9963],["2023-09-10",338.9963],["2023-09-17",338.9963],["2023-09-24",338.9963],["2023-10-01",338.9963],["2023-10-08",338.9963],["2023-10-15",338.9963],["2023-10-22",405.8516],["2023-10-29",466.8212],["2023-11-05",576.312],["2023-11-12",792.6059],["2023-11-19",975.1179],["2023-11-26",959.5878],["2023-12-03",1003.7141],["2023-12-10",1171.6377],["2023-12-17",1092.881],["2023-12-24",1805.2629],["2023-12-31",1662.6702],["2024-01-07",1648.3087],["2024-01-14",1728.7051],["2024-01-21",1728.7051],["2024-01-28",1728.7051],["2024-02-04",1716.1133],["2024-02-11",1926.9984],["2024-02-18",2007.8615],["2024-02-25",1918.8162],["2024-03-03",2299.1066],["2024-03-10",2562.9819],["2024-03-17",3456.5196],["2024-03-24",3456.9449],["2024-03-31",3595.3787],["2024-04-07",3657.7729],["2024-04-14",3496.5227],["2024-04-21",3496.5227],["2024-04-28",3496.5227],["2024-05-05",3496.5227],["2024-05-12",3483.8277],["2024-05-19",3697.4322],["2024-05-26",3553.5097],["2024-06-02",3628.2491],["2024-06-09",3544.8402],["2024-06-16",3398.0256],["2024-06-23",3398.0256],["2024-06-30",3398.0256],["2024-07-07",3398.0256],["2024-07-14",3398.0256],["2024-07-21",3478.7565],["2024-07-28",3611.6677],["2024-08-04",3268.1045],["2024-08-11",3268.1045],["2024-08-18",3268.1045],["2024-08-25",3232.937],["2024-09-01",3189.5104],["2024-09-08",3189.5104],["2024-09-15",3189.5104],["2024-09-22",3281.9594],["2024-09-29",3616.3254],["2024-10-06",3288.6485],["2024-10-13",3288.6485],["2024-10-20",3566.4447],["2024-10-27",3940.7012],["2024-11-03",3650.6128],["2024-11-10",4397.386],["2024-11-17",4965.437],["2024-11-24",5243.4509],["2024-12-01",4829.5882],["2024-12-08",4853.1475],["2024-12-15",4912.5436],["2024-12-22",4422.4925],["2024-12-29",4422.4925],["2025-01-05",4422.4925],["2025-01-12",3994.175],["2025-01-19",5092.8477],["2025-01-26",4843.0242],["2025-02-02",4348.1892],["2025-02-09",4348.1892],["2025-02-16",4348.1892],["2025-02-23",4348.1892],["2025-03-02",4348.1892],["2025-03-09",4348.1892],["2025-03-16",4348.1892],["2025-03-23",4348.1892],["2025-03-30",4348.1892],["2025-04-06",4348.1892],["2025-04-13",4348.1892],["2025-04-20",4348.1892],["2025-04-27",4380.3409],["2025-05-04",4267.9346],["2025-05-11",5886.6297],["2025-05-18",5652.3399],["2025-05-25",5930.2392],["2025-06-01",5935.4467],["2025-06-08",5935.4467],["2025-06-15",5322.818],["2025-06-22",5322.818],["2025-06-29",5320.5477],["2025-07-06",5323.7152],["2025-07-13",5720.3455],["2025-07-20",6404.8583],["2025-07-27",6578.3621],["2025-08-03",5712.4417],["2025-08-10",6147.5468],["2025-08-17",6239.1644],["2025-08-24",6239.1644],["2025-08-31",6239.1644],["2025-09-07",6239.1644],["2025-09-14",6608.4388],["2025-09-21",6519.3449],["2025-09-28",6519.3449],["2025-10-05",6389.9936],["2025-10-12",5568.4071],["2025-10-19",5568.4071],["2025-10-26",5568.4071],["2025-11-02",5568.4071],["2025-11-09",5568.4071],["2025-11-16",5568.4071],["2025-11-23",5568.4071],["2025-11-30",5568.4071],["2025-12-07",5568.4071],["2025-12-14",5568.4071],["2025-12-21",5568.4071],["2025-12-28",5568.4071],["2026-01-04",5568.4071],["2026-01-11",5568.4071],["2026-01-18",5568.4071],["2026-01-25",5568.4071],["2026-02-01",5568.4071],["2026-02-08",5568.4071],["2026-02-15",5568.4071],["2026-02-22",5568.4071],["2026-03-01",5568.4071],["2026-03-08",5568.4071],["2026-03-15",5568.4071],["2026-03-22",5568.4071],["2026-03-29",5568.4071],["2026-04-05",5568.4071],["2026-04-12",5568.4071],["2026-04-19",5329.8619],["2026-04-26",5323.7096],["2026-05-03",5256.3823],["2026-05-10",5470.202],["2026-05-17",4910.7246],["2026-05-24",4910.7246],["2026-05-31",4910.7246],["2026-06-07",4910.7246],["2026-06-14",4910.7246],["2026-06-21",4910.7246],["2026-06-28",4910.7246],["2026-07-05",4910.7246],["2026-07-12",4910.7246],["2026-07-19",4910.7246],["2026-07-26",4910.7246],["2026-08-02",4910.7246],["2026-08-09",4910.7246],["2026-08-16",4910.7246],["2026-08-23",5324.2531],["2026-08-30",5452.451],["2026-09-06",5423.2488],["2026-09-13",5474.4312],["2026-09-20",5436.4312],["2026-09-27",6055.6281],["2026-10-04",5960.6321],["2026-10-11",5483.0822]],"code_bytes":2180},"a17":{"id":"a17","alias":"Gannet","theme":"Intraday trend and momentum at 1h–4h on BTC/ETH/SOL with daily-regime filters","started":"2026-10-11T00:30:46+00:00","rounds":[{"k":1,"minutes":38.0},{"k":2,"minutes":14.1}],"messages":[{"t":"2026-10-11T00:30:52.794Z","text":"I’ll inspect the simulator and starting examples, submit a simple causal entry early, then test intraday momentum rules with daily regime filters. I’ll track the variants and record the signal maths in `WRITEUP.md`."},{"t":"2026-10-11T00:31:38.189Z","text":"The first entry is accepted: $100 became $230.64, with a 25.7% maximum drawdown. I’m now testing whether slower momentum across BTC, ETH and SOL can improve growth while keeping turnover costs manageable."},{"t":"2026-10-11T00:33:33.781Z","text":"The first daily all-coin momentum sweep performed poorly: it often bought recent spikes that then reversed, with severe drawdowns. I’m testing short-term reversal alongside the intraday trend rules to see whether that pattern survives fees and the referee’s fill model."},{"t":"2026-10-11T00:35:19.824Z","text":"The strongest result so far is a simple four-hour rotation: hold the BTC/ETH/SOL leader on 20-day momentum, require its price above a 15-day EMA, and use positive 30-day BTC momentum as the daily regime filter. The exact simulator gives $1,021.54 from $100, with a 36.4% drawdown. I’m checking causality and nearby parameters before replacing the entry."},{"t":"2026-10-11T00:40:20.455Z","text":"The four-hour entry is accepted at $1,021.54. Nearby momentum settings also performed well, and retaining the current leader through small ranking changes reduced turnover. Hourly dip buying earned much less, so I’m keeping trend as the main approach while testing broader regime filters and a small walk-forward forecasting model."},{"t":"2026-10-11T00:42:20.531Z","text":"Broader market filters have added little so far: breadth and Fear & Greed restrictions often missed profitable recoveries. The useful changes are simpler—slower ranking and fewer switches. I’m testing smoothed momentum to see whether it preserves the result across nearby horizons."},{"t":"2026-10-11T00:46:35.121Z","text":"Fitting a line to recent log prices looks more promising than using only the start and end prices: one tested version reached about $1,498 with fewer trades. Its update hour still affects returns noticeably. I’m checking hourly data and staggered decision times to see how much of that gain depends on timing."},{"t":"2026-10-11T00:52:54.128Z","text":"The regression entry is accepted: $100 became $1,497.79, with a 37.2% drawdown and 106 trades. Volume adjustments added little, while the daily forecasting model performed poorly. I’m now testing two remaining gaps: exiting broken trends between daily refreshes, and trading selected dips during cash periods."},{"t":"2026-10-11T00:58:23.996Z","text":"The entry with four-hour exits is accepted at $1,780.83, with a 30.4% drawdown. A separate asset-strength permission raised the backtest slightly, but depended on a narrow threshold and only a few extra trades, so I’m leaving it out. I’ll finish with a sizing check and document the results and failed approaches."},{"t":"2026-10-11T01:01:28.442Z","text":"Reducing exposure when estimated trend is weak relative to recent variance improved returns across several variance windows and lowered drawdown. I’m validating the middle setting—120 bars, rather than the best backtest setting—then updating the writeup with the final equations and checks."},{"t":"2026-10-11T01:05:43.098Z","text":"The final entry is accepted: $100 became $1,947.84, with a 28.2% drawdown. Double fees reduced it to $1,756.12. The final sensitivity checks show stable sizing results but substantial dependence on the trend horizon, so I’ll make that limitation explicit in the writeup."},{"t":"2026-10-11T01:08:47.575Z","text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: **$100 → $1,947.84** after costs; **28.2%** max drawdown.  \nNext idea: adapt trend duration to volatility using rolling validation."},{"t":"2026-10-11T01:08:55.490Z","text":"I’ll use the remaining time to test whether adapting the trend horizon improves on the accepted entry. I’ll reserve time to validate any replacement, submit it, and finish the writeup before the workspace seals."},{"t":"2026-10-11T01:11:12.425Z","text":"The accepted strategy’s position sizing is stable across nearby settings, but its 15-day trend horizon is sensitive. I’m testing whether a causal volatility rule can adapt that horizon, and will compare the results across all three visible years before choosing a replacement."},{"t":"2026-10-11T01:12:42.923Z","text":"Volatility-adaptive horizons did not improve on the accepted entry. The closest variant returned about $1,945 with a slightly larger drawdown; most returned much less. I’m keeping the fixed horizon and checking two remaining choices: risk-adjusted leader ranking and broader market permission."},{"t":"2026-10-11T01:14:50.227Z","text":"Using the average 30-day momentum of BTC, ETH and SOL for market permission improved the visible result to about $2,063 and reduced drawdown to 24.6%. Risk-adjusted leader ranking did not help. I’m checking the broader permission rule’s sensitivity and testing whether past strategy performance can select the trend horizon causally."},{"t":"2026-10-11T01:16:53.730Z","text":"The broader permission rule passed the referee’s causal check. It returned $2,062.74 after costs, with 24.62% drawdown, and retained $1,864.26 at double fees. I’m submitting it now. The volatility and performance-based horizon rules remain out; nearby horizon changes still expose a real sensitivity, which the writeup will state clearly."},{"t":"2026-10-11T01:19:51.293Z","text":"Submission #006 is accepted and is now the entry. I’m keeping it: the broader permission rule improved growth and drawdown, while the adaptive horizon rules added complexity without improving the result. I’m finishing the writeup with the equations, all 3,624 grid evaluations, stress results, failures, and the remaining sensitivity to the trend horizon."},{"t":"2026-10-11T01:22:55.543Z","text":"Accepted submission **#006**, gannet.py, is my final entry: long-only BTC/ETH/SOL trend trading with broad market permission, daily selection and sizing, and four-hour exits. Visible years turned **$100 into $2,062.74**, after costs, with **24.62% maximum drawdown**; the causal check passed. WRITEUP.md is complete with signal and sizing equations, parameters, 3,624 grid evaluations, results, failed ideas, and the unseen-year rationale, including the remaining sensitivity to the trend horizon."}],"submissions":[{"n":"001","t":"2026-10-11T00:31:09.684075+00:00","note":"Initial causal BTC trend filter; intraday research follows.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:36:33.255620+00:00","note":"4h liquid-major leader momentum, daily BTC regime, daily refresh; causal checked.","status":"ok","dev":1021.5414394283893,"dd":-0.36359479498218805,"reason":""},{"n":"003","t":"2026-10-11T00:48:08.612768+00:00","note":"Trend velocity from rolling log-price regression; 1% BTC regime hysteresis and 25% leader retention.","status":"ok","dev":1497.78586315637,"dd":-0.3716219145989389,"reason":""},{"n":"004","t":"2026-10-11T00:54:47.257397+00:00","note":"Daily trend entries at the completed-day boundary; four-hour exits when the own-price trend breaks.","status":"ok","dev":1780.8304012839596,"dd":-0.30422149177904845,"reason":""},{"n":"005","t":"2026-10-11T01:01:40.982553+00:00","note":"Closed-day trend entries, 4h exits, conservative trend-to-variance sizing; middle variance-window choice.","status":"ok","dev":1947.8372644222486,"dd":-0.2815243313826721,"reason":""},{"n":"006","t":"2026-10-11T01:16:55.733329+00:00","note":"Broader closed-day market permission: equal mean BTC/ETH/SOL 30-day log momentum; causal OLS velocity, daily entry and sizing, four-hour exits. Fixed horizon retained after adaptive rules failed.","status":"ok","dev":2062.741028161006,"dd":-0.2462086996284254,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: **$100 → $1,947.84** after costs; **28.2%** max drawdown.  \nNext idea: adapt trend duration to volatility using rolling validation."},{"k":2,"text":"Accepted submission **#006**, gannet.py, is my final entry: long-only BTC/ETH/SOL trend trading with broad market permission, daily selection and sizing, and four-hour exits. Visible years turned **$100 into $2,062.74**, after costs, with **24.62% maximum drawdown**; the causal check passed. WRITEUP.md is complete with signal and sizing equations, parameters, 3,624 grid evaluations, results, failed ideas, and the unseen-year rationale, including the remaining sensitivity to the trend horizon."}],"title":"Gannet (a17): writeup","idea":"The final entry is accepted submission 006, gannet.py, a self-contained four-hour strategy trading BTC, ETH and SOL. It estimates trend velocity by fitting a straight line to recent log prices, rather than measuring only an endpoint return. The mean closed-day momentum of BTC, ETH and SOL controls market permission. The strategy ranks opportunities once per day, retains its current leader through small ranking changes, and exits at any four-hour close when the held coin falls below its trend average. Exposure increases with estimated trend velocity relative to recent return variance. Cash earns nothing. The important execution asymmetry is slow entry and reranking, but faster withdrawal of exposure when a trend breaks.","failed":"- Broad daily winner rotation. Volume-ranked altcoins often reversed after spikes and had severe losses. The initial sweep's best exact sparse-order result was about $144 with over 90% drawdown; giving initial full targets changes the starting exposure but not the basic weakness of that family. - Daily loser rotation. Even with age, liquidity, own-trend and market filters, the tested reversal rules usually lost almost all capital. The best initial screen was about $2. - Daily clean-trend/volatility filters. Best exact result about $934, but 72.75% drawdown. Diversification across 3/5/10 leaders did not repair the broad altcoin variants. - Hourly dip buying. Best screen about $199 with low drawdown, insufficient growth for this objective. During primary cash periods its best extra benefit was only about 3%, concentrated in very few additional trades, so it was omitted. - Donchian breakouts and trailing channels. Best screen about $553 with 26% drawdown; less capital growth than the retained trend approach. - Fear & Greed, breadth and volume modifications. They often removed profitable recovery exposure. The strongest volume adjustment added only about 2% to its predecessor; it did not justify another final signal. - Walk-forward daily forecasting. Monthly histogram gradient boosting, trained on prior 180/365/730 days with completed next-day labels, price/range/volume/market features and seed 42, topped out around $104 after costs and had large drawdown. LightGBM could not load its native library offline, so the experiment used scikit-learn. No fitted model enters the final strategy. - Online expert weighting. Counterfactual strategies were scored only on returns already known at the current close. Best exact meta-result about $928, below the simple core. - Staggered deci","holdup":"Persistent directional flows can make recent price-path velocity informative about the next part of a rally. The rule updates its leader from data rather than expecting a named coin to repeat a historical result. Hysteresis reduces trading when market direction or leader ranking barely changes. Four-hour exits reduce the delay between a broken own-price trend and a sale. Fractional exposure limits commitment when the same positive slope is small relative to return variance. The equal mean market signal describes conditions across the actual tradable markets; its sign can stay positive through a rotation of strength between BTC and the two other coins. These mechanisms rely on observables available at decision time. The mean can also be dominated by the more volatile component's move; equal component weights do not mean equal risk contributions. The three liquid markets have lower stated costs than the altcoin universe, and fee doubling does not erase the visible result. All three anniversary blocks gained wealth, and the sizing neighbourhood is stable. These are useful checks, but not guarantees of persistence. The chief limitations are the large number of tried rules, few independ","result":{"dev":{"final":2062.741028161006,"max_drawdown":-0.2462086996284254,"trades":133,"exposure":0.38279115444283307,"assets_traded":3,"fees_paid":75.63361936770048},"hidden":{"final":123.23400853249805,"max_drawdown":-0.43554253403306875,"trades":124,"exposure":0.43868056969430536,"assets_traded":3,"fees_paid":11.234731241293863},"full":{"final":2541.9984546472747,"max_drawdown":-0.43554253403306853,"trades":257,"exposure":0.40513467751708,"assets_traded":3,"fees_paid":307.3770300854912}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",107.7008],["2021-10-31",110.1637],["2021-11-07",138.8925],["2021-11-14",132.3528],["2021-11-21",126.3557],["2021-11-28",126.3557],["2021-12-05",123.5064],["2021-12-12",123.5064],["2021-12-19",123.5064],["2021-12-26",123.5064],["2022-01-02",123.5064],["2022-01-09",123.5064],["2022-01-16",123.5064],["2022-01-23",123.5064],["2022-01-30",123.5064],["2022-02-06",123.5064],["2022-02-13",123.5064],["2022-02-20",123.5064],["2022-02-27",123.5064],["2022-03-06",123.3453],["2022-03-13",123.3453],["2022-03-20",120.0839],["2022-03-27",132.6164],["2022-04-03",158.0014],["2022-04-10",126.5055],["2022-04-17",125.1173],["2022-04-24",125.1173],["2022-05-01",125.1173],["2022-05-08",125.1173],["2022-05-15",125.1173],["2022-05-22",125.1173],["2022-05-29",125.1173],["2022-06-05",125.1173],["2022-06-12",125.1173],["2022-06-19",125.1173],["2022-06-26",125.1173],["2022-07-03",125.1173],["2022-07-10",125.1173],["2022-07-17",129.9989],["2022-07-24",154.1756],["2022-07-31",155.7173],["2022-08-07",154.9085],["2022-08-14",175.1256],["2022-08-21",158.0308],["2022-08-28",158.0308],["2022-09-04",158.0308],["2022-09-11",158.0308],["2022-09-18",158.0308],["2022-09-25",158.0308],["2022-10-02",158.0308],["2022-10-09",158.0308],["2022-10-16",158.0308],["2022-10-23",158.0308],["2022-10-30",171.6731],["2022-11-06",174.1681],["2022-11-13",162.8277],["2022-11-20",162.8277],["2022-11-27",162.8277],["2022-12-04",162.8277],["2022-12-11",162.9192],["2022-12-18",155.7451],["2022-12-25",155.7451],["2023-01-01",155.7451],["2023-01-08",150.7168],["2023-01-15",243.2385],["2023-01-22",259.2739],["2023-01-29",258.8982],["2023-02-05",247.9992],["2023-02-12",240.6084],["2023-02-19",239.3116],["2023-02-26",224.4418],["2023-03-05",224.4418],["2023-03-12",224.4418],["2023-03-19",253.4309],["2023-03-26",239.8995],["2023-04-02",241.8303],["2023-04-09",246.6243],["2023-04-16",267.8552],["2023-04-23",247.1883],["2023-04-30",247.1883],["2023-05-07",243.0218],["2023-05-14",238.7674],["2023-05-21",238.7674],["2023-05-28",238.7674],["2023-06-04",238.7674],["2023-06-11",238.7674],["2023-06-18",238.7674],["2023-06-25",236.5179],["2023-07-02",235.451],["2023-07-09",264.9721],["2023-07-16",340.6429],["2023-07-23",309.9016],["2023-07-30",293.677],["2023-08-06",293.677],["2023-08-13",292.1144],["2023-08-20",292.1144],["2023-08-27",292.1144],["2023-09-03",292.1144],["2023-09-10",292.1144],["2023-09-17",292.1144],["2023-09-24",292.1144],["2023-10-01",310.2578],["2023-10-08",318.5878],["2023-10-15",296.284],["2023-10-22",343.5665],["2023-10-29",395.1793],["2023-11-05",487.8668],["2023-11-12",670.9666],["2023-11-19",732.7583],["2023-11-26",677.4078],["2023-12-03",717.0214],["2023-12-10",843.2156],["2023-12-17",844.019],["2023-12-24",1321.692],["2023-12-31",1190.9689],["2024-01-07",1177.7796],["2024-01-14",1145.0452],["2024-01-21",1132.8027],["2024-01-28",1123.803],["2024-02-04",1087.1524],["2024-02-11",1143.1718],["2024-02-18",1214.3223],["2024-02-25",1260.5382],["2024-03-03",1417.451],["2024-03-10",1603.1916],["2024-03-17",2162.1157],["2024-03-24",1945.4622],["2024-03-31",2121.4552],["2024-04-07",1996.4108],["2024-04-14",1977.4815],["2024-04-21",1977.4815],["2024-04-28",1977.4815],["2024-05-05",1977.4815],["2024-05-12",1977.4815],["2024-05-19",2109.8483],["2024-05-26",2166.1998],["2024-06-02",2112.3137],["2024-06-09",2093.5661],["2024-06-16",2054.0817],["2024-06-23",2054.0817],["2024-06-30",2054.0817],["2024-07-07",2054.0817],["2024-07-14",2054.0817],["2024-07-21",2281.1241],["2024-07-28",2368.2779],["2024-08-04",2146.3105],["2024-08-11",2146.3105],["2024-08-18",2146.3105],["2024-08-25",2146.3105],["2024-09-01",2146.3105],["2024-09-08",2146.3105],["2024-09-15",2146.3105],["2024-09-22",2142.7737],["2024-09-29",2180.9098],["2024-10-06",2062.741],["2024-10-13",2062.741],["2024-10-20",2136.2764],["2024-10-27",2151.8659],["2024-11-03",1905.8533],["2024-11-10",2155.1584],["2024-11-17",2109.4074],["2024-11-24",2259.9076],["2024-12-01",2278.8916],["2024-12-08",2452.2257],["2024-12-15",2354.7212],["2024-12-22",2143.945],["2024-12-29",2143.945],["2025-01-05",2143.945],["2025-01-12",2143.945],["2025-01-19",2178.9909],["2025-01-26",2106.8444],["2025-02-02",1788.3529],["2025-02-09",1788.3529],["2025-02-16",1788.3529],["2025-02-23",1788.3529],["2025-03-02",1788.3529],["2025-03-09",1788.3529],["2025-03-16",1788.3529],["2025-03-23",1788.3529],["2025-03-30",1788.3529],["2025-04-06",1788.3529],["2025-04-13",1788.3529],["2025-04-20",1788.3529],["2025-04-27",1801.5765],["2025-05-04",1798.2638],["2025-05-11",2363.0689],["2025-05-18",2269.018],["2025-05-25",2378.4101],["2025-06-01",2309.8595],["2025-06-08",2270.3325],["2025-06-15",2203.7863],["2025-06-22",2203.7863],["2025-06-29",2203.7863],["2025-07-06",2223.1419],["2025-07-13",2615.4328],["2025-07-20",3280.7673],["2025-07-27",3349.3498],["2025-08-03",3075.1439],["2025-08-10",3063.7964],["2025-08-17",3215.6605],["2025-08-24",2841.2557],["2025-08-31",2706.0173],["2025-09-07",2665.0229],["2025-09-14",3099.8411],["2025-09-21",3042.065],["2025-09-28",2833.3056],["2025-10-05",2838.8018],["2025-10-12",2698.7753],["2025-10-19",2698.7753],["2025-10-26",2698.7753],["2025-11-02",2676.6531],["2025-11-09",2676.6531],["2025-11-16",2676.6531],["2025-11-23",2676.6531],["2025-11-30",2676.6531],["2025-12-07",2676.6531],["2025-12-14",2676.6531],["2025-12-21",2676.6531],["2025-12-28",2676.6531],["2026-01-04",2676.6531],["2026-01-11",2755.9536],["2026-01-18",2804.0506],["2026-01-25",2576.6426],["2026-02-01",2576.6426],["2026-02-08",2576.6426],["2026-02-15",2576.6426],["2026-02-22",2576.6426],["2026-03-01",2576.6426],["2026-03-08",2576.6426],["2026-03-15",2610.8725],["2026-03-22",2413.9414],["2026-03-29",2276.4231],["2026-04-05",2276.4231],["2026-04-12",2281.0521],["2026-04-19",2144.6857],["2026-04-26",2176.4845],["2026-05-03",2191.4865],["2026-05-10",2265.9875],["2026-05-17",2105.1892],["2026-05-24",2105.1892],["2026-05-31",2105.1892],["2026-06-07",2105.1892],["2026-06-14",2105.1892],["2026-06-21",2105.1892],["2026-06-28",2105.1892],["2026-07-05",2069.6308],["2026-07-12",1994.0023],["2026-07-19",2038.8547],["2026-07-26",2095.5963],["2026-08-02",2037.4283],["2026-08-09",2040.717],["2026-08-16",1976.4267],["2026-08-23",2436.2255],["2026-08-30",2711.7006],["2026-09-06",2702.9879],["2026-09-13",2583.5606],["2026-09-20",2495.0997],["2026-09-27",2716.1627],["2026-10-04",2673.5536],["2026-10-11",2541.9985]],"code_bytes":2072},"a18":{"id":"a18","alias":"Petrel","theme":"Volatility compression and expansion (squeezes) as entry timing","started":"2026-10-11T00:31:01+00:00","rounds":[{"k":1,"minutes":39.7},{"k":2,"minutes":7.6}],"messages":[{"t":"2026-10-11T00:31:07.990Z","text":"I’ll start with a causal squeeze strategy and submit an early entry, then test whether expansion strength and market breadth improve it. I’ll track the variants and record the maths and failures in the writeup."},{"t":"2026-10-11T00:31:57.787Z","text":"The first squeeze rule lost money after costs: too many breakouts failed before a trend developed. I’ll submit the simple BTC trend benchmark as an early entry, then test whether relative strength can distinguish durable expansions from brief spikes."},{"t":"2026-10-11T00:33:31.448Z","text":"The BTC benchmark is accepted: $100 became $230.64, with a 25.7% drawdown. Broad momentum and daily loser portfolios have also performed poorly so far. I’m testing faster exits and volatility-adjusted selection to see whether they reduce the damage from sharp reversals."},{"t":"2026-10-11T00:35:51.292Z","text":"Volatility-adjusted momentum improved some results, but its large drawdowns remain a problem. I’m now testing shorter-bar squeeze entries and a small model that retrains only on outcomes already known at each bar, so its visible results will also test adaptation over time."},{"t":"2026-10-11T00:37:53.191Z","text":"The strongest candidate so far buys short pullbacks within a broader trend in BTC, ETH and SOL. It grew $100 to about $486, with positive results in all three visible-year slices. I’ll test nearby settings and an altcoin breadth filter before choosing the next submission."},{"t":"2026-10-11T00:40:57.033Z","text":"The altcoin breadth filter made a substantial difference. Using the median return of liquid coins to identify rising markets, the best concentrated variant reached about $1,051, but drew down 75%. I’ll test concentration and recent-spike penalties before treating that higher return as a useful improvement."},{"t":"2026-10-11T00:42:31.021Z","text":"Faster exits improved the breadth strategy considerably: the leading version reached about $3,835, and weekly and fortnightly rotation gave similar results. The drawdown is still 69%, so I’m checking which trades caused it and whether a generic shock filter can help without discarding the main gains."},{"t":"2026-10-11T00:45:49.663Z","text":"The $3,835 strategy is now accepted. Its gains came from several different coins, but removing its best ten days reduces the result to about $100. It also lost money in the earlier bull-market test. I’ll treat it as a speculative tail-capture strategy and check whether a BTC fallback and event-triggered rotation improve its stability."},{"t":"2026-10-11T00:48:04.342Z","text":"Weekly rotation is sensitive to which bar starts the week. The BTC fallback improved returns and reduced drawdown in the tests so far. I’m now comparing rotation based on time since entry, which should be less dependent on an arbitrary calendar offset."},{"t":"2026-10-11T00:49:36.641Z","text":"Rotation based on time since entry is performing better. The leading version reaches about $8,996; nearby exit settings reach roughly $4,600–$6,600. It uses a BTC trend fallback, and the earlier bull-year check is now near breakeven or profitable across most of those settings. I’ll validate the leading version before submitting it."},{"t":"2026-10-11T00:52:46.934Z","text":"The $8,996 strategy is accepted and currently ranks second. Nearby-rule ensembles had steadier results in earlier data, but lower visible growth. I’m testing whether faster exits during volatility expansion or rising downside variance help across several settings, then I’ll finish the cost checks and writeup."},{"t":"2026-10-11T00:55:24.533Z","text":"The current strategy remains profitable with doubled costs: $100 becomes about $6,000. Removing its ten best days leaves about $235, so rare rallies still drive most of the growth. Volume-ranking additions and adaptive exits did not improve it. I’m checking the final parameter neighbourhood and updating the writeup with these limits."},{"t":"2026-10-11T01:00:15.042Z","text":"The additional continuation and regime-persistence rules did not beat the accepted entry. The remaining uncertainty is parameter sensitivity: small changes can move the strategy into different rare rallies. I’ll document that plainly and keep the final rule simple."},{"t":"2026-10-11T01:02:37.670Z","text":"The final code passed the causal checks. I changed it to emit a full target each day so a fresh account can enter an ongoing signal immediately. Each target is 100% in one coin or cash, so repeated targets caused no extra rebalancing costs in the visible run. The result remains $8,996.26."},{"t":"2026-10-11T01:09:28.290Z","text":"The writeup is complete, including every signal, the search count, failures, cost stress and parameter sensitivity. The final file matches the tested source and accepted submission. A three-exit ensemble improved earlier-period stability while reducing visible wealth to about $7,286; that is the most useful candidate to examine next round."},{"t":"2026-10-11T01:10:42.993Z","text":"Entry: #005 accepted — strategy.py; writeup.  \nVisible years: $100 → **$8,996.26**; **59.6%** max drawdown; rank **#2**.  \nNext idea: stress-test the three-exit ensemble for greater regime stability."},{"t":"2026-10-11T01:10:51.172Z","text":"I’ll compare the exit ensemble with the accepted strategy, focusing on parameter sensitivity and earlier-period results. Then I’ll choose the final entry, submit any change with time for scoring, and make sure the writeup matches it."},{"t":"2026-10-11T01:12:55.277Z","text":"The exit ensemble keeps most of the visible growth and improves both earlier-period results. The main concern is still the market filter: small changes to its window produce large changes in returns. I’m checking whether broader parameter averaging helps enough to justify replacing the current entry."},{"t":"2026-10-11T01:14:23.434Z","text":"I’m choosing the three-exit ensemble for the hidden years. Its visible result is lower—$7,286 versus $8,996—but it improves both earlier annual tests and spreads the exit decision across three speeds. Wider averaging weakened growth substantially, so I’m keeping the shared ranking and market filter."},{"t":"2026-10-11T01:16:58.338Z","text":"Submission 006 is with the referee, and the writeup now describes its three states, equal sizing, and daily rebalancing. The final code passed the local causal checks. At twice the stated costs it reaches $4,780; its earlier bull-year drawdown is still 73.6%, so exit averaging leaves substantial risk."},{"t":"2026-10-11T01:17:41.795Z","text":"Submission 006 is accepted and is now the entry: $100 becomes $7,286.45 on the visible years, with 57.9% drawdown. The code matches the validated file. I’m finishing the consistency check between the accepted copy, the results, and the writeup."},{"t":"2026-10-11T01:18:18.344Z","text":"Submission **006** is accepted as the final entry: daily momentum rotation with a median-market filter, three equally weighted exit rules, and a BTC trend fallback. Visible years turn **$100 into $7,286.45**, with **57.9% maximum drawdown**. I chose it over the higher-scoring single rule because both earlier annual tests improved. Causal checks passed. WRITEUP.md is complete with equations, parameters, 1,649 recorded evaluations, results, failures, and the remaining overfitting risks."}],"submissions":[{"n":"001","t":"2026-10-11T00:31:59.335649+00:00","note":"Early causal BTC 200-day benchmark; researching compression and expansion.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:38:53.753948+00:00","note":"Pullback within trend: 4h majors, causal BTC daily regime; positive in all three visible-year slices.","status":"ok","dev":485.8562839282218,"dd":-0.3599933554521577,"reason":""},{"n":"003","t":"2026-10-11T00:44:43.745559+00:00","note":"Breadth-gated 30-day momentum; daily rapid exits, weekly rotation. Lookahead passed; all three visible-year slices positive.","status":"ok","dev":3834.582181506042,"dd":-0.6882077813719805,"reason":""},{"n":"004","t":"2026-10-11T00:50:28.693651+00:00","note":"Event-clock momentum rotation with median-market trend and BTC fallback; no calendar phase in selection, causal checks passed.","status":"ok","dev":8996.258180020943,"dd":-0.5959251433747654,"reason":""},{"n":"005","t":"2026-10-11T01:01:48.492499+00:00","note":"Finalized event-clock breadth momentum with BTC fallback. Daily pure-position targets remove fresh-account entry delay; visible result unchanged, all causal cuts pass.","status":"ok","dev":8996.258180020943,"dd":-0.5959251433747652,"reason":""},{"n":"006","t":"2026-10-11T01:14:26.056747+00:00","note":"Petrel final: equal exit ensemble (5/8/10), median-market gate, entry-age momentum rotation, BTC trend fallback; prioritizes broader pre-cutoff performance over highest visible fit.","status":"ok","dev":7286.453834705852,"dd":-0.578756232997462,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: #005 accepted — strategy.py; writeup.  \nVisible years: $100 → **$8,996.26**; **59.6%** max drawdown; rank **#2**.  \nNext idea: stress-test the three-exit ensemble for greater regime stability."},{"k":2,"text":"Submission **006** is accepted as the final entry: daily momentum rotation with a median-market filter, three equally weighted exit rules, and a BTC trend fallback. Visible years turn **$100 into $7,286.45**, with **57.9% maximum drawdown**. I chose it over the higher-scoring single rule because both earlier annual tests improved. Causal checks passed. WRITEUP.md is complete with equations, parameters, 1,649 recorded evaluations, results, failures, and the remaining overfitting risks."}],"title":"Petrel (a18): momentum rotation controlled by market breadth, with an exit ensemble","idea":"The entry is strategy.py, accepted submission 006. Three separately maintained rules seek the strongest established, liquid coin when the typical liquid coin is trending upward. They share the ranking and market filter, but use short exit averages of 5, 8, and 10 days. Each rule starts reconsidering its incumbent seven days after selecting it. BTC supplies a trend-following fallback when no primary choice remains. The account targets the equal average of the three rules. The starting volatility-compression idea failed: a squeeze and a breakout often preceded a brief pump, rather than sustained expansion. The useful discoveries were a median-return market filter, rapid individual exits, and rotation timed from asset selection instead of an arbitrary weekday. The final ensemble sacrifices some visible return to reduce dependence on the single best exit speed; it improved both earlier annual tests. It remains speculative and parameter-sensitive, and does not establish that the million-dollar objective is attainable.","failed":"- Initial squeeze: \\sigma_7/\\sigma_{60}<0.65 at least once in the prior fortnight, followed by a 20-day high breakout, elevated volume, positive 30-day momentum, and a BTC trend gate. It ended near $13.83, with 93.2% drawdown. - Weekly momentum with only a BTC gate generally lost money. Major-coin strength did not reliably represent altcoin participation. - Daily loser rotation suffered from continued collapses and turnover costs. Risk-adjusted and price-path-efficiency rankings usually weakened the profitable raw-momentum tails. - A small histogram boosting model retrained every 90 bars on already known one- or three-day outcomes did not beat the simple benchmark. Causal walk-forward training did not supply enough stable net edge. - Recently listed coins produced unreliable annual results and very large drawdowns. - Strong compression and volume-shock entry filters excluded persistent moves while admitting brief spikes. - The best four-hour major-only pullback/trend rule reached $485.86. Shock reversion generally produced much smaller gains. - Conventional volatility caps reduced drawdown but greatly reduced growth. A 6% daily volatility target in an earlier version ended near $884. - Red-day exits, volume decay, and distribution-candle heuristics did not consistently solve the reversal problem. - Virtual-strategy selection, return-weighted online mixtures, volume-adjusted rankings, and asset-specific continuation estimates did not improve the leading simple rule. - Adaptive exits based on volatility expansion or downside/upside variance added complexity without a reliable improvement. Separate market entry and holding filters also weakened results. The final entry therefore retains the small set of signals stated in the equations, while documenting the search and its ","holdup":"The mechanism is plausible from data available through the cutoff and from general principles. Attention and new demand can spread over several days in a liquid speculative asset, making recent leadership informative. A median return across the current liquid universe identifies broad participation while limiting the influence of one extraordinary asset. An established coin above both short and medium averages is a more defensible long candidate than an isolated breakout below its broader trend. The BTC fallback addresses the observed fact that major-coin strength and altcoin breadth need not coincide. Rotation based on the time of selection avoids relying on a favourable weekday. Survivorship-free, volume-ranked selection can accommodate new leaders without forecasting their names. Equal averaging of nearby exit rules can reduce dependence on precisely when one moving average is crossed, while preserving full concentration when all three rules agree. Fixed component allocations avoid chasing the most recently profitable expert. These are reasons to forward-test the rule, not a demonstration of persistent excess return. Its positive tails are essential. Daily execution cannot guara","result":{"dev":{"final":7286.453834705852,"max_drawdown":-0.578756232997462,"trades":310,"exposure":0.4233576642335766,"assets_traded":45,"fees_paid":1662.5993892146103},"hidden":{"final":1.835302815371273,"max_drawdown":-0.9862049188718072,"trades":274,"exposure":0.42602539690891356,"assets_traded":50,"fees_paid":10.244802695036997},"full":{"final":133.59474184447058,"max_drawdown":-0.9909202708313474,"trades":585,"exposure":0.4244241729153926,"assets_traded":90,"fees_paid":2415.622054873686}},"freq":"1d","equity":[["2021-10-17",110.6735],["2021-10-24",111.421],["2021-10-31",112.4617],["2021-11-07",111.7553],["2021-11-14",167.5863],["2021-11-21",156.1817],["2021-11-28",156.1817],["2021-12-05",156.1817],["2021-12-12",156.1817],["2021-12-19",156.1817],["2021-12-26",156.1817],["2022-01-02",156.1817],["2022-01-09",156.1817],["2022-01-16",156.1817],["2022-01-23",156.1817],["2022-01-30",156.1817],["2022-02-06",156.1817],["2022-02-13",148.5375],["2022-02-20",148.5375],["2022-02-27",148.5375],["2022-03-06",148.5375],["2022-03-13",148.5375],["2022-03-20",148.5375],["2022-03-27",146.6404],["2022-04-03",215.2784],["2022-04-10",196.2063],["2022-04-17",196.2063],["2022-04-24",196.2063],["2022-05-01",196.2063],["2022-05-08",196.2063],["2022-05-15",196.2063],["2022-05-22",196.2063],["2022-05-29",196.2063],["2022-06-05",196.2063],["2022-06-12",196.2063],["2022-06-19",196.2063],["2022-06-26",196.2063],["2022-07-03",196.2063],["2022-07-10",196.2063],["2022-07-17",196.2063],["2022-07-24",243.9746],["2022-07-31",295.6975],["2022-08-07",273.4511],["2022-08-14",239.36],["2022-08-21",217.0855],["2022-08-28",217.0855],["2022-09-04",217.0855],["2022-09-11",217.0855],["2022-09-18",217.0855],["2022-09-25",217.0855],["2022-10-02",217.0855],["2022-10-09",217.0855],["2022-10-16",217.0855],["2022-10-23",217.0855],["2022-10-30",217.0855],["2022-11-06",197.6977],["2022-11-13",200.5882],["2022-11-20",200.5882],["2022-11-27",200.5882],["2022-12-04",200.5882],["2022-12-11",200.5882],["2022-12-18",200.5882],["2022-12-25",200.5882],["2023-01-01",200.5882],["2023-01-08",200.5882],["2023-01-15",219.8021],["2023-01-22",264.2004],["2023-01-29",349.0902],["2023-02-05",342.448],["2023-02-12",251.6464],["2023-02-19",433.6997],["2023-02-26",863.5034],["2023-03-05",832.7369],["2023-03-12",832.7369],["2023-03-19",896.3937],["2023-03-26",937.7106],["2023-04-02",971.5062],["2023-04-09",953.9439],["2023-04-16",1200.8438],["2023-04-23",1714.6359],["2023-04-30",1769.1906],["2023-05-07",1706.0012],["2023-05-14",1679.594],["2023-05-21",1679.594],["2023-05-28",1679.594],["2023-06-04",1625.1925],["2023-06-11",1625.1925],["2023-06-18",1625.1925],["2023-06-25",2041.9347],["2023-07-02",2144.6513],["2023-07-09",1960.3208],["2023-07-16",2058.2826],["2023-07-23",2042.6353],["2023-07-30",1975.0987],["2023-08-06",1936.6483],["2023-08-13",1912.3509],["2023-08-20",1912.3509],["2023-08-27",1912.3509],["2023-09-03",1879.8604],["2023-09-10",1879.8604],["2023-09-17",1879.8604],["2023-09-24",1937.7215],["2023-10-01",1925.6605],["2023-10-08",1896.894],["2023-10-15",1821.7131],["2023-10-22",2004.5798],["2023-10-29",2886.2081],["2023-11-05",5315.3991],["2023-11-12",4495.5803],["2023-11-19",4554.7333],["2023-11-26",4516.9302],["2023-12-03",4961.0566],["2023-12-10",6791.6656],["2023-12-17",6221.4641],["2023-12-24",7050.9076],["2023-12-31",5998.9547],["2024-01-07",6259.1886],["2024-01-14",6084.3228],["2024-01-21",6084.3228],["2024-01-28",6084.3228],["2024-02-04",6037.3109],["2024-02-11",5465.2908],["2024-02-18",4948.5317],["2024-02-25",4699.1968],["2024-03-03",6146.7806],["2024-03-10",11012.7035],["2024-03-17",10196.4544],["2024-03-24",9104.1486],["2024-03-31",9899.7719],["2024-04-07",9350.6507],["2024-04-14",9100.0283],["2024-04-21",9100.0283],["2024-04-28",8738.0131],["2024-05-05",8738.0131],["2024-05-12",8738.0131],["2024-05-19",8822.7241],["2024-05-26",12478.2229],["2024-06-02",13062.237],["2024-06-09",12872.4157],["2024-06-16",12191.9846],["2024-06-23",12191.9846],["2024-06-30",12191.9846],["2024-07-07",12191.9846],["2024-07-14",12191.9846],["2024-07-21",12979.0811],["2024-07-28",11734.3121],["2024-08-04",10617.9122],["2024-08-11",10617.9122],["2024-08-18",10617.9122],["2024-08-25",11856.6116],["2024-09-01",9004.898],["2024-09-08",9004.898],["2024-09-15",8557.0087],["2024-09-22",7813.8636],["2024-09-29",8637.6687],["2024-10-06",7457.329],["2024-10-13",7349.3608],["2024-10-20",7005.301],["2024-10-27",6822.7821],["2024-11-03",7054.8407],["2024-11-10",8782.7721],["2024-11-17",6456.6929],["2024-11-24",5473.0482],["2024-12-01",4997.5453],["2024-12-08",5442.9664],["2024-12-15",4773.2438],["2024-12-22",4578.9626],["2024-12-29",4336.4248],["2025-01-05",4213.7997],["2025-01-12",3641.8164],["2025-01-19",3170.7501],["2025-01-26",3166.469],["2025-02-02",3042.1886],["2025-02-09",2926.352],["2025-02-16",2926.352],["2025-02-23",2857.4331],["2025-03-02",2857.4331],["2025-03-09",2608.242],["2025-03-16",2608.242],["2025-03-23",2608.242],["2025-03-30",2334.6466],["2025-04-06",2334.6466],["2025-04-13",2334.6466],["2025-04-20",2288.5361],["2025-04-27",2358.5792],["2025-05-04",2683.3085],["2025-05-11",2973.757],["2025-05-18",2029.2334],["2025-05-25",1911.9365],["2025-06-01",1855.6931],["2025-06-08",1819.172],["2025-06-15",1975.751],["2025-06-22",1957.6077],["2025-06-29",1992.0947],["2025-07-06",2008.829],["2025-07-13",1267.9751],["2025-07-20",1029.5156],["2025-07-27",1304.2062],["2025-08-03",1176.5848],["2025-08-10",1139.0277],["2025-08-17",887.6159],["2025-08-24",832.4664],["2025-08-31",830.3852],["2025-09-07",830.3852],["2025-09-14",735.1457],["2025-09-21",690.0468],["2025-09-28",671.2737],["2025-10-05",717.9982],["2025-10-12",549.862],["2025-10-19",549.862],["2025-10-26",549.862],["2025-11-02",527.0249],["2025-11-09",527.0249],["2025-11-16",527.0249],["2025-11-23",527.0249],["2025-11-30",527.0249],["2025-12-07",527.0249],["2025-12-14",527.0249],["2025-12-21",527.0249],["2025-12-28",527.0249],["2026-01-04",522.7123],["2026-01-11",548.608],["2026-01-18",554.0877],["2026-01-25",554.0877],["2026-02-01",554.0877],["2026-02-08",554.0877],["2026-02-15",554.0877],["2026-02-22",554.0877],["2026-03-01",554.0877],["2026-03-08",554.0877],["2026-03-15",554.0877],["2026-03-22",676.3431],["2026-03-29",676.3431],["2026-04-05",676.3431],["2026-04-12",640.5115],["2026-04-19",858.929],["2026-04-26",856.9666],["2026-05-03",856.9666],["2026-05-10",661.6745],["2026-05-17",569.0639],["2026-05-24",569.0639],["2026-05-31",569.0639],["2026-06-07",569.0639],["2026-06-14",569.0639],["2026-06-21",569.0639],["2026-06-28",569.0639],["2026-07-05",569.0639],["2026-07-12",569.0639],["2026-07-19",569.0639],["2026-07-26",569.0639],["2026-08-02",569.0639],["2026-08-09",569.0639],["2026-08-16",569.0639],["2026-08-23",762.2085],["2026-08-30",672.9085],["2026-09-06",525.5169],["2026-09-13",528.9841],["2026-09-20",502.266],["2026-09-27",170.3204],["2026-10-04",136.3912],["2026-10-11",133.5947]],"code_bytes":2063},"a19":{"id":"a19","alias":"Skua","theme":"Risk parity and minimum-variance portfolios among the liquid majors","started":"2026-10-11T00:31:16+00:00","rounds":[{"k":1,"minutes":39.6},{"k":2,"minutes":15.1}],"messages":[{"t":"2026-10-11T00:31:22.533Z","text":"I’ll inspect the simulator and data, submit an early baseline, then test portfolio rules designed to hold up across different market regimes. I’ll keep a record of the variants and their maths for the writeup."},{"t":"2026-10-11T00:31:54.677Z","text":"The simulator confirms next-open fills and charges for every rebalance. I’ll start with weekly selection of liquid, established coins with positive trends, sized by inverse volatility, and daily exits when those trends break."},{"t":"2026-10-11T00:32:52.230Z","text":"The first broad momentum portfolio fell to $12.41 in the visible test. It kept buying rebound leaders through weak markets, so I’m treating it as a failed variant. I’ve submitted a majors-only risk-parity baseline that reached $171.02; next I’ll test market filters and exits on the broader universe."},{"t":"2026-10-11T00:34:54.717Z","text":"The baseline passed the referee’s causality checks. Across the broader tests, the largest recent gainers have produced too much volatility to compound well. I’m now testing steadier trends, using return per unit of volatility or total price movement, and comparing concentrated portfolios with diversified ones."},{"t":"2026-10-11T00:37:03.000Z","text":"Concentrating in the strongest trend among BTC, ETH and SOL improved the visible result to about $393. The broad dip-buying variants also failed to beat the baseline. I’ll next test whether intraday exits can retain more of the major-coin rallies, plus a small model trained only on labels available at each decision."},{"t":"2026-10-11T00:40:16.530Z","text":"The second submission is accepted: $100 became $464.48, with a 29.3% maximum drawdown. The pooled linear forecasts performed poorly, so I’m dropping them. I’ll test a different selection rule next: enter a volume-backed breakout when it happens, rather than buy whichever coin already has the largest trailing return."},{"t":"2026-10-11T00:44:23.287Z","text":"So far, the broader coin tests are losing too much to adverse reversals and volatility, and the nonlinear forecasts haven’t improved on the simpler trend rule. I’m narrowing the remaining work to major-coin signals, testing whether short bursts of price and volume contain enough continuation to pay the trading costs."},{"t":"2026-10-11T00:48:24.316Z","text":"Minimum-variance and equal-risk portfolios also underperformed the current entry. A more promising variant ranks coins by the recent net return of a simple trend-trading rule, after simulated fees. Its best visible test reached $551.36; I’ll check nearby parameters before considering a submission."},{"t":"2026-10-11T00:50:43.185Z","text":"Letting the EMA crossover govern exits, instead of exiting on every dip below the slow EMA, improved the high-volatility major-coin variant to $750.13. All three visible annual windows were positive. I’m checking nearby trend lengths and finer bars to see whether the result survives changes in timing."},{"t":"2026-10-11T00:54:29.782Z","text":"The third entry is accepted at $750.13. It retained $641.78 with doubled costs and $650.01 with an extra four-hour fill delay. The nine tests across nearby trend lengths and bar frequencies ranged from $568.69 to $750.13. Its main weakness is concentration: the best ten days supplied about 82% of total log growth."},{"t":"2026-10-11T00:57:01.418Z","text":"The writeup now covers the current entry, its equations, failed approaches and sensitivity checks. The filtered Kalman-slope and wider trend-band variants didn’t improve on it. I’m testing upside and downside volatility next, to see whether they make a better selection signal than total volatility."},{"t":"2026-10-11T00:59:14.820Z","text":"Requiring positive recent BTC momentum improved the latest asymmetric-risk test to $1,591.38. A 40-day version reached $1,245.20. I’ll test a consensus of the two windows, with half exposure when they disagree, to reduce dependence on the best-performing lookback."},{"t":"2026-10-11T01:01:28.519Z","text":"Submission 004 is accepted at $1,424.16, with a 40.5% maximum drawdown. It holds half exposure when the 20- and 40-day BTC momentum signals disagree, and selects coins using upside return energy minus half of downside energy. I’ll update the writeup and complete cost, delay and pre-window checks for this version."},{"t":"2026-10-11T01:06:48.477Z","text":"The accepted entry retained $922–$1,403 across the parameter checks, and about $1,325–$1,340 on finer bars. It also returned $416.57 from $100 in the short pre-visible window; that was a check after selection, not an independent holdout. Conditional diversification reduced growth, so I’m keeping the one-coin allocation and testing one final profit-lock rule."},{"t":"2026-10-11T01:10:49.260Z","text":"Entry: entry.py, accepted #004; writeup complete.  \nVisible years: **$100 → $1,424.16**, max drawdown **40.5%**; causality checks passed.  \nNext idea: earlier recovery entries with market-breadth confirmation."},{"t":"2026-10-11T01:10:55.801Z","text":"I’ll test the early-recovery idea against the current entry, then lock in the strategy for the hidden years and finish the writeup. I’ll leave enough time for the referee to score any replacement."},{"t":"2026-10-11T01:13:17.487Z","text":"The twelve early-recovery variants all finished below the accepted entry. Breadth confirmation improved some recovery periods but also admitted costly bear-market rallies. I’ll keep the current market gate and test whether the upside-energy selection rule works beyond the three majors."},{"t":"2026-10-11T01:15:39.499Z","text":"The broad upside-energy portfolios suffered large losses. A separate test of momentum among recent listings reached $3,860, but with an 80% drawdown and a losing annual window. I’ll test whether simple risk filters improve that result before considering it as a replacement."},{"t":"2026-10-11T01:17:29.827Z","text":"Keeping a majors portfolio alongside the listing strategy improved the result: the equal split reached $3,329, with a 54% drawdown and one annual window near flat. I’ll test an equal blend of 90- and 180-day listing cohorts, then check nearby parameters and higher costs before choosing the final entry."},{"t":"2026-10-11T01:19:57.542Z","text":"I’m submitting the daily core-and-cohort strategy. It reached $2,779.94 with a 50.1% drawdown and passed the referee’s causality check. Recent listings receive at most half the risk budget, and only when both BTC momentum clocks are positive. This version gave up some visible wealth for better annual results under doubled costs; its sensitivity to a one-day execution delay remains a significant limitation."},{"t":"2026-10-11T01:20:55.898Z","text":"Submission #005 is accepted and is now the entry. Nearby settings all retained gains over the full visible period, though some annual windows lost money; the accepted version ended at $1,867 with doubled costs. I’m freezing the code and finishing the writeup, including the weak delay result and the substantial multiple-testing bias."},{"t":"2026-10-11T01:23:05.065Z","text":"The final checks show that the listing signal is sensitive to its horizon: changing momentum from seven days to ten days cuts visible wealth to $1,181. I’ve left the submitted settings frozen. The writeup will report that weakness and the full search count—484 recorded evaluations plus standalone checks—alongside the equations and accepted result."},{"t":"2026-10-11T01:25:57.814Z","text":"Submission **005** is accepted: entry.py combines a BTC/ETH/SOL trend core with a capped allocation to young, liquid momentum leaders. Visible results: **$100 → $2,779.94 (×27.80)**, with **50.14% maximum drawdown**. Causality checks passed; doubled costs left $1,866.61. WRITEUP.md is complete with equations, parameters, 484 recorded evaluations, rejected ideas, and limitations—including strong sensitivity to delayed fills."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:32.972029+00:00","note":"Early entry: weekly risk parity among trending BTC ETH SOL, 200-day filter","status":"ok","dev":171.0181109430941,"dd":-0.4850978237025201,"reason":""},{"n":"002","t":"2026-10-11T00:39:00.274136+00:00","note":"Relative strength among majors; daily BTC regime, 4h trend exits; replaces risk parity after comparative tests","status":"ok","dev":464.4770846481689,"dd":-0.2930109196349613,"reason":""},{"n":"003","t":"2026-10-11T00:51:47.156516+00:00","note":"Growth exposure: highest realized-volatility eligible major in BTC uptrend, with EMA-cross exits; robust across nearby lengths and bar frequencies","status":"ok","dev":750.125093441099,"dd":-0.4594857045078782,"reason":""},{"n":"004","t":"2026-10-11T01:00:10.267912+00:00","note":"20/40-day market-trend consensus with partial cash; select upside second moment minus half downside second moment; 4h EMA-cross exits","status":"ok","dev":1424.1641735226622,"dd":-0.4054990682269777,"reason":""},{"n":"005","t":"2026-10-11T01:20:00.073767+00:00","note":"Daily asymmetric-energy major core plus half-budget 90/180-observation liquid listing cohorts; young risk requires both BTC momentum clocks, with a common 100-day trend gate.","status":"ok","dev":2779.9397676900244,"dd":-0.5013775677703333,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: entry.py, accepted #004; writeup complete.  \nVisible years: **$100 → $1,424.16**, max drawdown **40.5%**; causality checks passed.  \nNext idea: earlier recovery entries with market-breadth confirmation."},{"k":2,"text":"Submission **005** is accepted: entry.py combines a BTC/ETH/SOL trend core with a capped allocation to young, liquid momentum leaders. Visible results: **$100 → $2,779.94 (×27.80)**, with **50.14% maximum drawdown**. Causality checks passed; doubled costs left $1,866.61. WRITEUP.md is complete with equations, parameters, 484 recorded evaluations, rejected ideas, and limitations—including strong sensitivity to delayed fills."}],"title":"Skua (a19): final writeup","idea":"The final entry is submission 005, entry.py, at daily frequency. A core rotates among BTC, ETH and SOL using upside return energy minus half downside return energy, conditional on a positive asset trend. A satellite selects momentum leaders among young, liquid exchange listing cohorts. It averages the leaders in two cohorts, with maximum observed ages of ninety and one hundred eighty daily observations. The satellite uses at most half the invested budget and requires positive Bitcoin momentum at both twenty and forty days. The remaining budget goes to the major core, or cash if no core asset qualifies. Participation requires Bitcoin above its hundred-day average; with only one positive momentum clock, at most half of wealth goes to majors and none to the satellite. The assigned conventional risk-parity/minimum- variance direction was abandoned after comparisons showed insufficient growth. The final construction combines established liquidity with the upside and failure risk of emerging assets, using cash and a bounded satellite rather than an optimized covariance portfolio.","failed":"- Broad daily momentum/inverse-volatility rotation lost most of the account; enlarging the universe worsened reversals. Risk-adjusted and path-efficiency scores did not repair it. The initial broad portfolio ended at $12.41. - Direct minimum-variance/equal-risk comparisons topped out near $158. Major weekly inverse volatility gave $171.02; variance reduction supplied insufficient growth. - Short-horizon liquid-coin dip buying and impulse continuation/reversal were weak after costs. - Price/volume breakouts with ATR exits reached $275.32 in a narrow universe; many broad variants nearly exhausted the account. - Purged walk-forward ridge forecasts reached $178.86; shallow nonlinear four-hour forecasts reached $140.00. Causality and fixed seeds did not create predictive skill. - Counterfactual major trend-expert selection reached $551.36 but was inconsistent at longer scoring windows. Its broad daily counterpart failed. - Faster four-hour rotation increased turnover; single-coin versions were weaker than dynamic major selection. Kalman slope-confidence filters did not improve the simpler EMA family. - In the earlier four-hour family, uncertainty-based two-asset diversification gave $645-$1,340. A tight volatility ratchet gave $934; loose ratchets made no difference. Shadow-equity filters gave $499/$861/$1,316. - A third market vote from broad twenty-day breadth gave $1,276 in the earlier four-hour entry. Twelve breadth-confirmed early-recovery variants all finished below its $1,424. Broad daily-energy rotation reached at best $164 and often suffered near-total losses. - Pure young-cohort exposure could reach $3,860-$4,117 in favored settings but suffered about 76%-80% drawdowns and lost money in the middle annual interval. Top-three diversification, five positive days out o","holdup":"The proposed mechanism is persistent price trends plus changing attention to emerging liquid assets. The major core supplies exposure when no young candidate qualifies. Its upside energy distinguishes volatility generated by advances from volatility generated by declines: this is conditional growth sensitivity, not a claim that variance alone forecasts positive drift. The satellite requires already-observed positive leadership, meaningful volume and trend confirmation. It never assumes a specific future winning ticker. Young assets may still be undergoing rapid adoption and price discovery, but attention can also precede distribution or failure. Their bounded allocation acknowledges that risk. The age ensemble expresses uncertainty about attention decay; the market ensemble expresses uncertainty about trend duration. Cash reduces common market exposure more directly than moving it among correlated coins. Full market agreement precedes acceptance of satellite risk. The fixed half allocation avoids estimating unstable means or covariance matrices for new assets. All rules can run on contemporaneously closed bars and newly observed tickers. These remain hypotheses. Listing-age effects","result":{"dev":{"final":2779.9397676900244,"max_drawdown":-0.5013775677703333,"trades":1109,"exposure":0.3414775795164529,"assets_traded":92,"fees_paid":412.5762238977357},"hidden":{"final":25.981155767186955,"max_drawdown":-0.7574276265162811,"trades":757,"exposure":0.362523141496995,"assets_traded":76,"fees_paid":16.071236545457293},"full":{"final":721.4030397019752,"max_drawdown":-0.7748513088022001,"trades":1867,"exposure":0.34989119410889297,"assets_traded":161,"fees_paid":862.3469953817619}},"freq":"1d","equity":[["2021-10-17",97.8265],["2021-10-24",97.2608],["2021-10-31",117.449],["2021-11-07",143.4062],["2021-11-14",130.5268],["2021-11-21",150.9318],["2021-11-28",145.8851],["2021-12-05",145.8851],["2021-12-12",145.8851],["2021-12-19",145.8851],["2021-12-26",145.8851],["2022-01-02",145.8851],["2022-01-09",145.8851],["2022-01-16",145.8851],["2022-01-23",145.8851],["2022-01-30",145.8851],["2022-02-06",145.8851],["2022-02-13",145.8851],["2022-02-20",145.8851],["2022-02-27",145.8851],["2022-03-06",145.8851],["2022-03-13",145.8851],["2022-03-20",145.8851],["2022-03-27",140.3748],["2022-04-03",252.6917],["2022-04-10",223.8631],["2022-04-17",223.721],["2022-04-24",220.4815],["2022-05-01",220.4815],["2022-05-08",220.4815],["2022-05-15",220.4815],["2022-05-22",220.4815],["2022-05-29",220.4815],["2022-06-05",220.4815],["2022-06-12",220.4815],["2022-06-19",220.4815],["2022-06-26",220.4815],["2022-07-03",220.4815],["2022-07-10",220.4815],["2022-07-17",220.4815],["2022-07-24",220.4815],["2022-07-31",220.4815],["2022-08-07",220.4815],["2022-08-14",220.4815],["2022-08-21",220.4815],["2022-08-28",220.4815],["2022-09-04",220.4815],["2022-09-11",220.4815],["2022-09-18",209.3407],["2022-09-25",209.3407],["2022-10-02",209.3407],["2022-10-09",209.3407],["2022-10-16",209.3407],["2022-10-23",209.3407],["2022-10-30",209.3407],["2022-11-06",203.4085],["2022-11-13",187.5245],["2022-11-20",187.5245],["2022-11-27",187.5245],["2022-12-04",187.5245],["2022-12-11",187.5245],["2022-12-18",187.5245],["2022-12-25",187.5245],["2023-01-01",187.5245],["2023-01-08",187.5245],["2023-01-15",259.4967],["2023-01-22",296.2291],["2023-01-29",344.2086],["2023-02-05",349.8805],["2023-02-12",279.4931],["2023-02-19",282.1035],["2023-02-26",258.974],["2023-03-05",226.6916],["2023-03-12",226.6916],["2023-03-19",215.8553],["2023-03-26",199.3744],["2023-04-02",201.6576],["2023-04-09",178.6185],["2023-04-16",227.2005],["2023-04-23",207.7123],["2023-04-30",202.2253],["2023-05-07",191.8879],["2023-05-14",191.7939],["2023-05-21",191.7939],["2023-05-28",191.7939],["2023-06-04",188.8365],["2023-06-11",191.5908],["2023-06-18",191.5908],["2023-06-25",240.6651],["2023-07-02",238.5338],["2023-07-09",237.856],["2023-07-16",265.528],["2023-07-23",251.1082],["2023-07-30",250.8194],["2023-08-06",246.1138],["2023-08-13",246.1138],["2023-08-20",244.4202],["2023-08-27",244.4202],["2023-09-03",244.4202],["2023-09-10",244.4202],["2023-09-17",244.4202],["2023-09-24",244.4202],["2023-10-01",244.4202],["2023-10-08",244.4202],["2023-10-15",244.4202],["2023-10-22",287.0632],["2023-10-29",347.5835],["2023-11-05",607.0422],["2023-11-12",862.7497],["2023-11-19",1000.0914],["2023-11-26",947.2314],["2023-12-03",969.9427],["2023-12-10",1203.7513],["2023-12-17",1560.4261],["2023-12-24",1712.9566],["2023-12-31",1675.113],["2024-01-07",1696.0018],["2024-01-14",1526.7713],["2024-01-21",1457.1581],["2024-01-28",1457.1581],["2024-02-04",1448.5125],["2024-02-11",1671.1188],["2024-02-18",1769.6502],["2024-02-25",1763.1521],["2024-03-03",2016.4326],["2024-03-10",2316.1434],["2024-03-17",2531.3714],["2024-03-24",2220.8513],["2024-03-31",2283.0211],["2024-04-07",2268.7063],["2024-04-14",1979.2829],["2024-04-21",1979.2829],["2024-04-28",1930.196],["2024-05-05",1914.4769],["2024-05-12",1918.2569],["2024-05-19",2000.464],["2024-05-26",2169.802],["2024-06-02",2547.3818],["2024-06-09",2752.5134],["2024-06-16",2755.2877],["2024-06-23",2755.2877],["2024-06-30",2755.2877],["2024-07-07",2755.2877],["2024-07-14",2755.2877],["2024-07-21",2753.9428],["2024-07-28",2757.4468],["2024-08-04",2366.2368],["2024-08-11",2366.2368],["2024-08-18",2366.2368],["2024-08-25",2403.8064],["2024-09-01",2373.0701],["2024-09-08",2373.0701],["2024-09-15",2373.0701],["2024-09-22",2553.068],["2024-09-29",2554.9388],["2024-10-06",2349.3523],["2024-10-13",2776.6395],["2024-10-20",2730.0166],["2024-10-27",2278.3344],["2024-11-03",2159.3648],["2024-11-10",2468.5627],["2024-11-17",2263.5489],["2024-11-24",2578.1349],["2024-12-01",2402.994],["2024-12-08",2223.6598],["2024-12-15",2203.8488],["2024-12-22",1540.4855],["2024-12-29",1560.135],["2025-01-05",1554.5986],["2025-01-12",1502.1725],["2025-01-19",1589.3758],["2025-01-26",1494.0135],["2025-02-02",1284.8628],["2025-02-09",1248.4714],["2025-02-16",1248.4714],["2025-02-23",1233.7626],["2025-03-02",1233.7626],["2025-03-09",1233.7626],["2025-03-16",1233.7626],["2025-03-23",1233.7626],["2025-03-30",1233.7626],["2025-04-06",1233.7626],["2025-04-13",1233.7626],["2025-04-20",1233.7626],["2025-04-27",1397.6457],["2025-05-04",1727.8861],["2025-05-11",1555.0463],["2025-05-18",1171.2048],["2025-05-25",1143.4239],["2025-06-01",1185.1033],["2025-06-08",1181.0288],["2025-06-15",1048.092],["2025-06-22",1036.0526],["2025-06-29",1047.4722],["2025-07-06",1025.9207],["2025-07-13",1149.0395],["2025-07-20",1009.7404],["2025-07-27",1450.0774],["2025-08-03",1259.1549],["2025-08-10",1333.7079],["2025-08-17",1164.3539],["2025-08-24",1111.1454],["2025-08-31",1110.79],["2025-09-07",1110.79],["2025-09-14",1102.6349],["2025-09-21",1040.6355],["2025-09-28",998.4312],["2025-10-05",1001.0148],["2025-10-12",907.4557],["2025-10-19",907.4557],["2025-10-26",907.4557],["2025-11-02",907.4557],["2025-11-09",907.4557],["2025-11-16",907.4557],["2025-11-23",907.4557],["2025-11-30",907.4557],["2025-12-07",907.4557],["2025-12-14",907.4557],["2025-12-21",907.4557],["2025-12-28",907.4557],["2026-01-04",907.4557],["2026-01-11",907.4557],["2026-01-18",907.4557],["2026-01-25",907.4557],["2026-02-01",907.4557],["2026-02-08",907.4557],["2026-02-15",907.4557],["2026-02-22",907.4557],["2026-03-01",907.4557],["2026-03-08",907.4557],["2026-03-15",907.4557],["2026-03-22",907.4557],["2026-03-29",907.4557],["2026-04-05",907.4557],["2026-04-12",907.4557],["2026-04-19",930.3087],["2026-04-26",909.3554],["2026-05-03",757.1554],["2026-05-10",862.1965],["2026-05-17",793.75],["2026-05-24",787.9507],["2026-05-31",787.9507],["2026-06-07",787.9507],["2026-06-14",787.9507],["2026-06-21",787.9507],["2026-06-28",787.9507],["2026-07-05",787.9507],["2026-07-12",787.9507],["2026-07-19",787.9507],["2026-07-26",787.9507],["2026-08-02",787.9507],["2026-08-09",787.9507],["2026-08-16",787.9507],["2026-08-23",788.1628],["2026-08-30",789.9734],["2026-09-06",871.6401],["2026-09-13",736.9811],["2026-09-20",759.9251],["2026-09-27",764.6391],["2026-10-04",774.1908],["2026-10-11",721.403]],"code_bytes":1726},"a20":{"id":"a20","alias":"Puffin","theme":"Hierarchical risk parity and clustering of coins by correlation","started":"2026-10-11T00:31:31+00:00","rounds":[{"k":1,"minutes":39.1},{"k":2,"minutes":11.7}],"messages":[{"t":"2026-10-11T00:31:37.994Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test whether correlation clusters help identify durable momentum. I’ll keep a record of every variant and use walk-forward results to guide changes."},{"t":"2026-10-11T00:32:22.767Z","text":"The baseline is submitted. The data contains 459 coin histories, including delistings. I’ll test weekly momentum selection with two uses for correlation clustering: avoiding several bets on the same move, and allocating risk among the selected coins."},{"t":"2026-10-11T00:34:31.859Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first clustering variants performed poorly: long-lookback momentum often selected coins after large price spikes, then caught their reversals. I’m testing earlier trend signals and less extended entries before changing the submitted strategy."},{"t":"2026-10-11T00:37:14.160Z","text":"The broader tests also lost money. Among liquid altcoins, both extreme winners and extreme losers tended to perform poorly afterward, so correlation-based sizing has not supplied a trading edge. I’ll test a slower BTC/ETH/SOL rotation and a small walk-forward model trained only on labels already known at each decision."},{"t":"2026-10-11T00:40:20.378Z","text":"The second entry is accepted: $100 became $516.08, with a 39.7% drawdown. A four-hour trend variant has reached about $855 locally. I’m also testing whether agreement within a correlation cluster can distinguish broad trends from isolated pumps. LightGBM could not load a required system library, so the learning test now uses scikit-learn."},{"t":"2026-10-11T00:42:28.567Z","text":"The walk-forward model reached $583 with one coin, but diversification reduced its return and its bear-year result was weak. The four-hour trend approach looks stronger so far; buying intraday dips added little growth. I’m testing longer-lived leaders and whether entry timing can improve without adding frequent trades."},{"t":"2026-10-11T00:45:22.804Z","text":"I submitted the walk-forward model with a BTC 50/200-day trend filter. Locally, it turned $100 into $1,138.76, with gains in each annual sub-period and a 54% drawdown. It retrains every 28 days and uses only completed seven-day outcomes. The referee is checking causality; I’ll test sensitivity to the horizon and training window next."},{"t":"2026-10-11T00:47:50.727Z","text":"The model was accepted at $1,138.76. Its sensitivity tests are concerning: changing the training window sharply reduced returns, and removing its ten best days leaves roughly $99. I’m testing an adaptive mix of a few causal strategies, using completed shadow-account returns to choose exposure."},{"t":"2026-10-11T00:51:03.683Z","text":"Adaptive strategy selection did not improve returns. I’m testing whether averaging forecasts can reduce dependence on one training window, and whether market context helps the model separate coin-specific moves from broad rallies. The writeup now records the accepted entries, the clustering results, and the main failures."},{"t":"2026-10-11T00:54:56.793Z","text":"Averaging predictions from 45, 60 and 75 boosting stages reached $2,013.74 locally. Shifting the three-day rebalance schedule by one or two days sharply reduced that result, so I’ll keep the accepted entry while testing ways to remove that timing dependence: gradual position changes and switches that require enough forecast improvement to cover trading costs."},{"t":"2026-10-11T00:58:33.556Z","text":"Spreading the snapshot model’s trades across three schedules removed the large timing-driven gain; it reached about $338. Switching only when forecast improvement covered trading costs also underperformed. I’ll test one final way to smooth the daily forecasts, then choose the entry and finish its validation and equations."},{"t":"2026-10-11T01:02:12.620Z","text":"I’m leaning toward the intraday trend strategy: smoothing did not make the all-coin forecasts reliable, while nearby intraday settings kept producing positive returns. I’ll check a blend of absolute momentum and BTC-adjusted momentum, then submit the rule and document why I passed on the larger, timing-sensitive backtests."},{"t":"2026-10-11T01:06:15.906Z","text":"The new entry was accepted: $100 became $789.70, with a 43.8% drawdown. Its timing checks were steadier: all six rebalance offsets returned $617–$1,025. Doubling costs still left $631, and removing the ten best days left $289. I’ll finish the writeup around this rule and record the larger fits I rejected."},{"t":"2026-10-11T01:10:39.186Z","text":"Entry: accepted #004, puffin.py; writeup completed.  \nVisible: $100 → $789.70; 99.1% CAGR; 43.8% max drawdown.  \nNext idea: trend-efficiency filters and volatility-aware exits."},{"t":"2026-10-11T01:10:46.885Z","text":"I’ll use the final round to test trend-efficiency filters and volatility-aware exits against the accepted strategy, then submit any improvement early enough to score. I’ll keep checking timing sensitivity and update the writeup with the exact final rule and results."},{"t":"2026-10-11T01:12:37.218Z","text":"The accepted entry still has the strongest timing robustness among my earlier candidates, though it trails the board. I’m testing whether a simple measure of price-path efficiency can reduce its repeated losses in sideways markets without cutting off sustained trends."},{"t":"2026-10-11T01:14:06.877Z","text":"A mild efficiency filter improved the visible result and reduced drawdown; stricter filters missed too much of the trends. Volatility-scaled stops mostly added churn. I’m now checking every timing offset of the mild-filter candidates before choosing a replacement."},{"t":"2026-10-11T01:15:38.405Z","text":"The replacement candidate turns $100 into $1,010 locally, with 36.4% drawdown. Its other two rebalance offsets return $857 and $995, and doubled costs leave $825. It uses a higher efficiency threshold for entry than for holding, which reduces churn; I’m submitting it now."},{"t":"2026-10-11T01:16:50.260Z","text":"Submission #005 is accepted: $1,010.01 with 36.4% drawdown, replacing the $789.70 entry. I’m updating the paper with the exact efficiency and exit rules. One final check will test whether broad-market participation adds useful confirmation; it will need to pass timing checks to make the entry."},{"t":"2026-10-11T01:19:43.088Z","text":"Broad-market and sentiment filters added little. A separate test of momentum in newly listed coins lost heavily, so I’m excluding it. The writeup now documents submission #005’s exact rules and validation. I’ll make one last check of a correlation-scaled switching threshold to see whether it reduces unnecessary leadership changes."},{"t":"2026-10-11T01:21:09.470Z","text":"The switching-threshold test reduced turnover but added little return, so submission #005 remains the entry. The writeup is complete, including the equations, all 688 saved configuration results, failed ideas and limitations. I’m checking that the submitted file, paper figures and local entry file agree."},{"t":"2026-10-11T01:22:18.514Z","text":"Accepted submission **#005**, puffin.py, holds BTC, ETH or SOL when momentum and trend filters agree, using stricter efficiency thresholds for entry than exit. Visible result: **$100 → $1,010.01 after costs**, with **36.4% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, 688 saved configuration results, robustness checks and failed approaches."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:03.215449+00:00","note":"Early causal baseline: BTC above its 200-day average; clustering research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:39:14.613024+00:00","note":"Major-coin momentum rotation: positive 28-day return and own 50-day trend; daily target, cash otherwise.","status":"ok","dev":516.0767493839754,"dd":-0.39724657677465147,"reason":""},{"n":"003","t":"2026-10-11T00:44:29.643215+00:00","note":"Purged rolling 7-day relative-return learner, one liquid coin, BTC above 50/200-day means; fixed small trees.","status":"ok","dev":1138.7571133396855,"dd":-0.5420048732144496,"reason":""},{"n":"004","t":"2026-10-11T01:04:20.357918+00:00","note":"Blend 14/21-day absolute and beta-adjusted BTC leadership; four-hour trend and causal own daily filter. Chosen for robustness over larger phase-sensitive fits.","status":"ok","dev":789.7035166206674,"dd":-0.4383646469704734,"reason":""},{"n":"005","t":"2026-10-11T01:15:41.383817+00:00","note":"Correlation-adjusted major-coin leadership; 14-day efficiency entry/exit hysteresis; close exits and 12-hour selection; all schedule phases profitable","status":"ok","dev":1010.0137475642217,"dd":-0.3638017677094063,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #004, puffin.py; writeup completed.  \nVisible: $100 → $789.70; 99.1% CAGR; 43.8% max drawdown.  \nNext idea: trend-efficiency filters and volatility-aware exits."},{"k":2,"text":"Accepted submission **#005**, puffin.py, holds BTC, ETH or SOL when momentum and trend filters agree, using stricter efficiency thresholds for entry than exit. Visible result: **$100 → $1,010.01 after costs**, with **36.4% maximum drawdown**; causal checks passed. WRITEUP.md is complete with equations, parameters, 688 saved configuration results, robustness checks and failed approaches."}],"title":"Puffin (a20): writeup","idea":"The entry is correlation-adjusted trend leadership with path-efficiency hysteresis: hold one of BTC, ETH and SOL only when its daily and four-hour trends agree, and rank eligible coins using a blend of absolute momentum and momentum remaining after accounting for BTC exposure. New entries must also have enough net upward movement relative to the total distance travelled by prices. A lower efficiency threshold governs existing positions, allowing a trend to pause without forcing immediate turnover. Combining two momentum horizons and shrinking the market adjustment avoids relying on one precise lookback or one noisy beta estimate. The strategy holds cash when no coin qualifies, checks exits every four hours, and selects leadership every twelve hours. I started with hierarchical risk parity and full-universe correlation clusters. Those allocations did not supply an expected-return advantage. I ultimately used correlation information in the selection signal, rather than spreading capital among weak opportunities. I rejected larger supervised-learning backtests because changing their rebalance phase often erased the apparent advantage. Entry: puffin.py, byte-identical to accepted submission 005, originally submitted as puffin_efficiency.py. Its 2,057 bytes import only NumPy and pandas, and use no randomness or external resources. SHA-256: 7592c05a0a460477f409ff300534e89","failed":"1. Hierarchical risk parity and correlation-diversified momentum. The initial raw 28/84-day momentum, three-coin weekly portfolio turned $100 into about $5. The correlation screen and HRP sizing reduced redundant bets but often allocated to already mature price spikes. I also tested full correlation cohorts, requiring peer agreement before selecting their leaders; the best such variant reached about $156. 2. Short reversal and momentum across liquid altcoins. Daily purchases of both extreme losers and extreme winners suffered continuing adverse moves and large turnover. Conservative volatility and price-extension screens helped some variants but did not produce a compelling broad-universe edge. 3. Long-horizon altcoin leadership and volume-confirmed breakouts. Lookbacks out to a year still often selected decaying leaders. Fresh, moderate channel breakouts with volume expansion did not reliably persist; the best breakout configuration reached about $201. 4. Intraday dip buying. Buying deviations below a moving mean within daily uptrends produced modest growth and missed important directional moves. The best examined variant reached about $144. 5. Purged walk-forward machine learning. Small histogram-gradient-boosting models were retrained every 28 days on completed 3–14-day labels and a rolling training window. The selected seven-day, 504-day-memory learner reached $1,138.76 and was accepted as submission 003. Changing its training horizon or window often erased the gain. Removing its ten best days left about $99. 6. Boosting-stage averaging and schedule overfit. Averaging predictions at 45/60/75 stages reached $2,013.74; a slightly different threshold reached $2,639.16. The three-day schedule's other two phases reached only about $232 and $52. Spreading capital across p","holdup":"The economic premise is that sustained crypto trends occasionally persist across both daily and intraday horizons. Confirmation on two frequencies reduces exposure to isolated price jumps. Correlation adjustment distinguishes some coin-specific leadership from a common market move, and shrinking that adjustment limits beta-estimation error. Requiring positive absolute momentum keeps the opportunity comparison tied to growth against cash. Path efficiency rejects some back-and-forth price movement; hysteresis reduces needless reactions to small threshold crossings. Fast eligibility exits and a slower daily gate can avoid prolonged declines while leaving room to compound during sustained advances. The three-coin choice follows the permitted universe and stated cost advantage. It does not depend on remembered later winners or future listings. Coefficients, filters and target scheduling are date-independent. The same calculation can continue on longer supplied histories without retraining choices or changing code. The phase, cost and parameter checks provide more support than a lone large fitted return. They do not establish a durable edge. Three years contain few independent market reg","result":{"dev":{"final":1010.0137475642217,"max_drawdown":-0.3638017677094063,"trades":202,"exposure":0.40358880778588807,"assets_traded":3,"fees_paid":76.18209342586073},"hidden":{"final":182.98950124167592,"max_drawdown":-0.32030257813756435,"trades":154,"exposure":0.4404109589041096,"assets_traded":3,"fees_paid":21.423876033929666},"full":{"final":1848.2191191401298,"max_drawdown":-0.3638017677094063,"trades":356,"exposure":0.4183096020445418,"assets_traded":3,"fees_paid":292.56618662966673}},"freq":"4h","equity":[["2021-10-10",100.628],["2021-10-17",109.8273],["2021-10-24",119.6126],["2021-10-31",122.348],["2021-11-07",154.2543],["2021-11-14",146.9912],["2021-11-21",140.3308],["2021-11-28",140.3308],["2021-12-05",134.072],["2021-12-12",134.072],["2021-12-19",134.072],["2021-12-26",134.072],["2022-01-02",134.072],["2022-01-09",134.072],["2022-01-16",134.072],["2022-01-23",134.072],["2022-01-30",134.072],["2022-02-06",134.072],["2022-02-13",126.6506],["2022-02-20",119.6078],["2022-02-27",119.6078],["2022-03-06",119.6078],["2022-03-13",111.6764],["2022-03-20",110.7576],["2022-03-27",118.567],["2022-04-03",154.0195],["2022-04-10",120.4658],["2022-04-17",120.4658],["2022-04-24",120.4658],["2022-05-01",120.4658],["2022-05-08",120.4658],["2022-05-15",120.4658],["2022-05-22",120.4658],["2022-05-29",120.4658],["2022-06-05",120.4658],["2022-06-12",120.4658],["2022-06-19",120.4658],["2022-06-26",120.4658],["2022-07-03",120.4658],["2022-07-10",120.4658],["2022-07-17",121.2356],["2022-07-24",134.4251],["2022-07-31",135.7693],["2022-08-07",125.6847],["2022-08-14",142.0878],["2022-08-21",128.218],["2022-08-28",128.218],["2022-09-04",128.218],["2022-09-11",131.4676],["2022-09-18",112.0901],["2022-09-25",112.0901],["2022-10-02",112.0901],["2022-10-09",109.4058],["2022-10-16",109.4058],["2022-10-23",109.4058],["2022-10-30",118.8504],["2022-11-06",117.1784],["2022-11-13",109.5487],["2022-11-20",109.5487],["2022-11-27",109.5487],["2022-12-04",109.5487],["2022-12-11",109.5487],["2022-12-18",109.5487],["2022-12-25",109.5487],["2023-01-01",109.5487],["2023-01-08",109.1974],["2023-01-15",188.2548],["2023-01-22",200.6654],["2023-01-29",198.9602],["2023-02-05",192.1472],["2023-02-12",189.3981],["2023-02-19",177.115],["2023-02-26",164.1155],["2023-03-05",156.4994],["2023-03-12",156.4994],["2023-03-19",173.8568],["2023-03-26",171.2238],["2023-04-02",172.8773],["2023-04-09",172.8807],["2023-04-16",187.3158],["2023-04-23",160.3438],["2023-04-30",160.3438],["2023-05-07",154.3501],["2023-05-14",151.648],["2023-05-21",151.648],["2023-05-28",151.648],["2023-06-04",148.2625],["2023-06-11",131.4057],["2023-06-18",131.4057],["2023-06-25",141.1139],["2023-07-02",148.5522],["2023-07-09",163.3614],["2023-07-16",210.0141],["2023-07-23",191.0614],["2023-07-30",179.1299],["2023-08-06",179.1299],["2023-08-13",179.1299],["2023-08-20",173.1561],["2023-08-27",173.1561],["2023-09-03",173.1561],["2023-09-10",173.1561],["2023-09-17",173.1561],["2023-09-24",172.2688],["2023-10-01",183.0164],["2023-10-08",187.9302],["2023-10-15",174.7057],["2023-10-22",203.0172],["2023-10-29",233.5158],["2023-11-05",288.2858],["2023-11-12",396.4815],["2023-11-19",432.9948],["2023-11-26",403.9318],["2023-12-03",398.6672],["2023-12-10",448.2863],["2023-12-17",448.7134],["2023-12-24",702.663],["2023-12-31",633.1655],["2024-01-07",615.1601],["2024-01-14",581.0666],["2024-01-21",574.854],["2024-01-28",565.761],["2024-02-04",537.9319],["2024-02-11",602.046],["2024-02-18",657.5117],["2024-02-25",716.5093],["2024-03-03",805.7009],["2024-03-10",903.8777],["2024-03-17",1214.0997],["2024-03-24",1092.4416],["2024-03-31",1148.6156],["2024-04-07",1107.003],["2024-04-14",1149.0954],["2024-04-21",1149.0954],["2024-04-28",1149.0954],["2024-05-05",1149.0954],["2024-05-12",1149.0954],["2024-05-19",1226.0125],["2024-05-26",1269.6335],["2024-06-02",1238.0503],["2024-06-09",1196.1299],["2024-06-16",1196.1299],["2024-06-23",1196.1299],["2024-06-30",1196.1299],["2024-07-07",1196.1299],["2024-07-14",1196.1299],["2024-07-21",1322.6093],["2024-07-28",1373.1416],["2024-08-04",1251.8742],["2024-08-11",1251.8742],["2024-08-18",1251.8742],["2024-08-25",1251.7008],["2024-09-01",1142.6902],["2024-09-08",1142.6902],["2024-09-15",1142.6902],["2024-09-22",1109.5199],["2024-09-29",1185.261],["2024-10-06",1010.0137],["2024-10-13",1010.0137],["2024-10-20",1047.0022],["2024-10-27",1099.3239],["2024-11-03",998.0099],["2024-11-10",1183.4358],["2024-11-17",1245.4594],["2024-11-24",1252.5756],["2024-12-01",1243.5826],["2024-12-08",1338.1703],["2024-12-15",1284.7606],["2024-12-22",1208.1213],["2024-12-29",1208.1213],["2025-01-05",1201.3429],["2025-01-12",1169.2608],["2025-01-19",1268.2854],["2025-01-26",1206.0711],["2025-02-02",929.6411],["2025-02-09",929.6411],["2025-02-16",929.6411],["2025-02-23",929.6411],["2025-03-02",929.6411],["2025-03-09",929.6411],["2025-03-16",929.6411],["2025-03-23",929.6411],["2025-03-30",929.6411],["2025-04-06",929.6411],["2025-04-13",929.6411],["2025-04-20",957.1765],["2025-04-27",1046.6851],["2025-05-04",1028.7239],["2025-05-11",1278.8155],["2025-05-18",1227.9183],["2025-05-25",1300.3275],["2025-06-01",1273.3732],["2025-06-08",1273.3732],["2025-06-15",1226.4218],["2025-06-22",1203.0468],["2025-06-29",1203.0468],["2025-07-06",1198.4096],["2025-07-13",1362.4412],["2025-07-20",1709.0298],["2025-07-27",1744.756],["2025-08-03",1662.1556],["2025-08-10",1791.0183],["2025-08-17",1879.8478],["2025-08-24",1992.8478],["2025-08-31",1664.3368],["2025-09-07",1669.2543],["2025-09-14",1887.2804],["2025-09-21",1852.1045],["2025-09-28",1725.0053],["2025-10-05",1748.3148],["2025-10-12",1662.0776],["2025-10-19",1662.0776],["2025-10-26",1662.0776],["2025-11-02",1662.0776],["2025-11-09",1662.0776],["2025-11-16",1662.0776],["2025-11-23",1662.0776],["2025-11-30",1662.0776],["2025-12-07",1662.0776],["2025-12-14",1662.0776],["2025-12-21",1662.0776],["2025-12-28",1662.0776],["2026-01-04",1655.5714],["2026-01-11",1653.1812],["2026-01-18",1686.5003],["2026-01-25",1617.211],["2026-02-01",1617.211],["2026-02-08",1617.211],["2026-02-15",1617.211],["2026-02-22",1617.211],["2026-03-01",1617.211],["2026-03-08",1617.211],["2026-03-15",1617.211],["2026-03-22",1560.028],["2026-03-29",1455.093],["2026-04-05",1455.093],["2026-04-12",1465.6472],["2026-04-19",1532.8858],["2026-04-26",1599.4578],["2026-05-03",1524.5407],["2026-05-10",1599.6201],["2026-05-17",1474.9429],["2026-05-24",1474.9429],["2026-05-31",1474.9429],["2026-06-07",1474.9429],["2026-06-14",1474.9429],["2026-06-21",1474.9429],["2026-06-28",1474.9429],["2026-07-05",1517.0876],["2026-07-12",1473.6703],["2026-07-19",1506.8186],["2026-07-26",1548.7536],["2026-08-02",1553.6872],["2026-08-09",1553.6872],["2026-08-16",1494.86],["2026-08-23",1804.8037],["2026-08-30",1896.4034],["2026-09-06",1890.3102],["2026-09-13",1753.54],["2026-09-20",1750.9464],["2026-09-27",1907.2734],["2026-10-04",1877.3536],["2026-10-11",1848.2191]],"code_bytes":2057},"a21":{"id":"a21","alias":"Godwit","theme":"Rebalancing premium / volatility harvesting (Shannon's demon) between imperfectly correlated coins","started":"2026-10-11T00:31:46+00:00","rounds":[{"k":1,"minutes":38.2},{"k":2,"minutes":19.4}],"messages":[{"t":"2026-10-11T00:31:52.341Z","text":"I’ll inspect the simulator and submit a simple causal baseline early, then test whether volatility harvesting adds value after costs. I’ll keep a record of the variants and results for the writeup."},{"t":"2026-10-11T00:32:26.111Z","text":"The baseline is submitted and awaiting the referee. The simulator confirms next-bar-open fills and charges for every rebalance. I’ll compare coin-and-cash harvesting, diversified baskets, and momentum rotation under those costs; correlation and persistent declines are the main challenges for a harvesting strategy."},{"t":"2026-10-11T00:36:47.815Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first broad momentum baskets performed poorly, including in the recovery year. My research simulator matches the official engine on both single-coin and multi-coin portfolios, so I’m now testing short-term reversal and slower rotation rather than tuning those losing baskets."},{"t":"2026-10-11T00:38:36.679Z","text":"The basic volatility-harvesting tests haven’t shown a large edge. A BTC/SOL basket with individual trend filters reached about $287 but drew down 52%. I’m checking whether fresh breakouts and fewer switches improve on that. I’ll record the broad search count in the writeup, since many trials make a lucky backtest easier to find."},{"t":"2026-10-11T00:39:54.875Z","text":"Slower, volatility-scaled momentum has done better: a liquid top-10 rotation reached about $662 in research, though its maximum drawdown was still 62%. I’ve put that rule into a self-contained entry file and started the causality check. Next I’ll test whether drawdown guards or a small diversified reserve improve the tradeoff."},{"t":"2026-10-11T00:42:57.912Z","text":"The slower rule is accepted at $661.92. The drawdown guards I tested reduced returns without consistently reducing drawdown, so I’ve kept them out of the entry. I’m now testing market breadth for the cash decision, plus a reversal control without the individual trend filter that may have excluded the useful rebounds."},{"t":"2026-10-11T00:45:45.183Z","text":"A two-rule blend reached about $1,258 with a 33% drawdown. I also found a comparison issue: a fresh account can wait in cash if a rule emits only on signal changes. I’m rerunning the concentrated controls with daily targets before attributing the blend’s gain to volatility harvesting. Daily targets also let the final rule establish its position promptly at any start date."},{"t":"2026-10-11T00:49:19.396Z","text":"The blend is accepted: $100 became $1,257.78, with a 32.6% maximum drawdown. The corrected single-rule controls reached about $888 and $1,184, so the blend improved on both in this sample. I’m checking higher costs, separate periods, and whether small rebalances can be skipped without losing that gain."},{"t":"2026-10-11T00:52:14.250Z","text":"The current entry stayed profitable at twice the stated costs: $100 became about $1,091. It also gained in each visible year when run separately. The weaker evidence is that its earlier-data check reached only $201, and its best ten days supplied about 80% of visible log growth. I’m testing additional momentum clocks to see whether performance becomes less dependent on one period."},{"t":"2026-10-11T00:56:02.906Z","text":"Recent-performance weighting made the blend worse in these tests. Fixed mixtures worked better. A faster momentum component raised the visible result to about $1,942, while a four-clock mixture reached about $1,397 and performed better on the earlier scoring period. I’m checking whether those gains survive nearby parameters before choosing the next entry."},{"t":"2026-10-11T01:00:00.562Z","text":"The four-clock rule passed the causality check and is submitted at $1,636.92, with a 44.5% drawdown. A small rebalancing buffer reached about $1,714 in research and is now undergoing the full check. I’m testing its sensitivity to costs, liquidity cutoffs, lookbacks, and breadth thresholds before the next submission."},{"t":"2026-10-11T01:03:58.816Z","text":"The buffered rule now passes the causality check and runs in under a second. I replaced its Numba implementation after a compiler-cache import failure. The sensitivity tests still show a large effect from the liquidity cutoff, so I’ll make that limitation explicit. I’m also comparing the fitted buffer with a fee-derived band before the final submission and writeup."},{"t":"2026-10-11T01:07:44.464Z","text":"Submission #005 is accepted: $100 became $1,715.17, with a 42.9% maximum drawdown. The fee-derived band performed about the same as the fitted buffer, so I kept the derived rule. The writeup is complete, including all signal equations, the roughly 4,000 configurations tested, failures, cost stress tests, and the strategy’s substantial tail dependence."},{"t":"2026-10-11T01:09:59.128Z","text":"Entry: accepted #005 — strategy, writeup complete.  \nVisible years: $100 → **$1,715.17**; max drawdown **42.9%**; causality check passed.  \nNext idea: learn momentum continuation across liquidity levels using purged walk-forward training."},{"t":"2026-10-11T01:10:08.021Z","text":"I’ll use the remaining time to test whether a model trained on already-known outcomes can improve coin selection. I’ll keep the accepted four-clock strategy as the fallback, and only replace it after checking causality, costs, and performance across periods."},{"t":"2026-10-11T01:12:22.816Z","text":"The accepted entry is intact at $1,715.17. Its main weakness is sensitivity to the liquidity cutoff. I’m testing pooled forecasts to see whether they can improve selection across a broader universe, using only training outcomes that have fully elapsed before each decision."},{"t":"2026-10-11T01:14:00.114Z","text":"The first pooled forecasts have not beaten the four-clock entry on the visible interval, though some did better earlier. I’m also testing whether rising trading volume helps identify fresh momentum leaders. Any change needs to show a convincing improvement across periods."},{"t":"2026-10-11T01:15:26.011Z","text":"Volume growth has helped the weekly leader in these tests. The leading variant finished near $2,000 on the visible years and $2,700 on the earlier interval, but with deeper drawdowns. I’m testing it within the equal-capital portfolio and checking nearby liquidity cutoffs before deciding whether to submit it."},{"t":"2026-10-11T01:16:08.972Z","text":"Replacing just the weekly score with momentum multiplied by the square root of volume growth improved both scoring intervals at all three tested liquidity cutoffs. The four-clock candidate now reaches $2,199.34, with 43.7% drawdown. I’m running the referee’s causality check and the broader sensitivity audit before submitting it."},{"t":"2026-10-11T01:17:31.744Z","text":"Submission #006 is accepted and is now the entry: $100 became $2,199.34 after costs. All three visible annual intervals improved, and the broader sensitivity checks remained profitable. I’m updating the writeup and measuring how much the combined portfolio gained versus funding the four rules separately."},{"t":"2026-10-11T01:20:42.041Z","text":"The portfolio comparison supports rebalancing: four independently funded rules finished at $1,547.88, versus $2,199.34 for the pooled entry. I’m checking one remaining sizing choice—whether to leave inactive rules’ capital in cash or share it among active rules. The accepted entry keeps the cash."},{"t":"2026-10-11T01:21:36.363Z","text":"Sharing capital among active clocks improved both intervals at every tested liquidity cutoff, while leaving the measured visible drawdown unchanged. At the original cutoff, it reached $3,141.40. This uses equal shares among active rules and holds cash when none qualifies. I’m verifying it as the final sizing rule."},{"t":"2026-10-11T01:24:01.523Z","text":"The sizing audit found a material risk: expanding the slower rule’s universe from five coins to six caused a 79% drawdown under full allocation. I’m checking whether lower exposure when fewer clocks qualify preserves much of the gain while reducing that fragility. The writeup will report this result explicitly."},{"t":"2026-10-11T01:25:11.373Z","text":"I’m choosing the more cautious sizing rule for the hidden years: split capital equally among active clocks, with total exposure proportional to the square root of their active fraction. It gives up about 19% of the full-allocation visible result, while reducing the six-coin stress drawdown from 79% to 58%. All gates and leader signals stay the same."},{"t":"2026-10-11T01:28:19.151Z","text":"Submission #008 is accepted and is the final entry: $100 became $2,538.41 after costs, with 43.7% drawdown. The writeup now covers the final sizing equation, all signals and filters, 4,340 retained research records, cost stress, failed approaches, and the remaining liquidity-cutoff and tail-return risks."},{"t":"2026-10-11T01:29:22.777Z","text":"Final accepted entry **#008**, godwit.py, combines four breadth-gated momentum clocks, weekly volume attention, and exposure that rises with the number of active clocks. It turned **$100 into $2,538.41** on the visible years after costs, with **43.7% maximum drawdown**, and passed the referee’s causality checks. WRITEUP.md is complete with equations, parameters, 4,340 retained research records, validation results, failed ideas, and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:04.367502+00:00","note":"Initial causal BTC trend baseline; volatility-harvesting research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:39:57.402467+00:00","note":"Slow volatility-scaled momentum; causal volume-ranked universe and switch hysteresis.","status":"ok","dev":661.919638782842,"dd":-0.619892476049956,"reason":""},{"n":"003","t":"2026-10-11T00:47:38.023345+00:00","note":"Breadth-gated Shannon portfolio: daily equal-capital rebalancing of two liquid momentum clocks.","status":"ok","dev":1257.7838932653854,"dd":-0.3255867354106444,"reason":""},{"n":"004","t":"2026-10-11T00:58:26.507938+00:00","note":"Four equal-capital momentum clocks; breadth gates, dynamic liquidity universe, and daily volatility harvesting.","status":"ok","dev":1636.920008049947,"dd":-0.44498404331078456,"reason":""},{"n":"005","t":"2026-10-11T01:04:17.707762+00:00","note":"Four momentum clocks with breadth gates and fee-derived log-utility rebalancing bands; fully causal, dynamic coin universe.","status":"ok","dev":1715.1663333397125,"dd":-0.4288774303214793,"reason":""},{"n":"006","t":"2026-10-11T01:16:27.585339+00:00","note":"Godwit: four equal breadth-gated clocks, weekly volume-growth attention, fee-derived buffer; causal daily targets","status":"ok","dev":2199.340710400624,"dd":-0.43740598937157404,"reason":""},{"n":"007","t":"2026-10-11T01:21:51.038446+00:00","note":"Godwit final: equal capital among active breadth-gated clocks, weekly volume attention, cost-aware drift buffer; no leverage","status":"ok","dev":3141.397939979284,"dd":-0.43740598937157416,"reason":""},{"n":"008","t":"2026-10-11T01:27:29.411706+00:00","note":"Godwit final: four causal breadth-gated clocks, weekly volume attention, sqrt(active/4) exposure, fee-derived buffer; chosen over full allocation after universe stress","status":"ok","dev":2538.41419610661,"dd":-0.43740598937157427,"reason":""}],"entry":"008","round_notes":[{"k":1,"text":"Entry: accepted #005 — strategy, writeup complete.  \nVisible years: $100 → **$1,715.17**; max drawdown **42.9%**; causality check passed.  \nNext idea: learn momentum continuation across liquidity levels using purged walk-forward training."},{"k":2,"text":"Final accepted entry **#008**, godwit.py, combines four breadth-gated momentum clocks, weekly volume attention, and exposure that rises with the number of active clocks. It turned **$100 into $2,538.41** on the visible years after costs, with **43.7% maximum drawdown**, and passed the referee’s causality checks. WRITEUP.md is complete with equations, parameters, 4,340 retained research records, validation results, failed ideas, and generalization limits."}],"title":"Godwit (a21): writeup","idea":"The entry is a breadth-gated, four-clock Shannon portfolio with volume attention. Four causal rotation rules look for liquid leaders over 7, 30, 90 and 180 days. The weekly score also rewards growth in trailing quote volume, seeking fresh participation in a rising coin. Capital is divided equally among active rules, with target exposure equal to the square root of the active fraction: 50% when one rule qualifies, rising to 100% when all four qualify. The rest stays in cash. A common Bitcoin trend gate prevents exposure in broad downtrends, while two market-breadth gates distinguish strong rallies from slower recoveries. Rebalancing diversifies the clocks' mistakes and harvests fluctuations between their choices. A transaction-cost band derived from a local log-utility approximation lets small deviations drift. Every non-BTC selection comes from contemporaneous trailing volume and prices; there are no ticker lists, calendar switches, or embedded historical observations. Latest accepted entry: submission #008, godwit.py. The self-contained file imports only numpy and pandas, is about 6.5 KB, and uses daily bars. It is byte-identical to submission 008. The referee accepted the final submission, which ranked 14th of 36 entries at scoring.","failed":"- Static BTC/ETH/SOL harvesting with trend guards: the best simple tested basket reached about $287 but drew down about 52%. Correlated assets and persistent declines overwhelmed the standalone harvesting premium. - Broad weekly/daily momentum baskets without sufficiently strong breadth control: poor net results and severe drawdowns. Increasing the liquid universe generally admitted damaging leaders rather than supplying useful diversification. - Short-term reversal, including removing the own-price trend filter: turnover and adverse continuation overwhelmed rebounds. The best unfiltered reversal control ended around $75. - Fresh price/volume breakout holding rules: best tested control around $281, with substantial drawdown and weak recovery-period performance. - Intraday major-coin rotations: best tested result around $312; frequent switching consumed the edge. - Tight price-drawdown guards: frequently sold corrections and did not consistently improve either growth or drawdown. - Smooth breadth exposure ramps: generally reduced growth; the best close control was about $1,170 versus $1,258 for the then-current hard-gated two-clock blend. - Exponentially weighting clocks by recent virtual performance: worse than fixed allocations, consistent with buying a clock after its winning phase. - Pooled purged walk-forward ridge/tree forecasts: more complex forecasts did not deliver enough visible net growth to justify replacement. They were discarded even where earlier-period results were strong. - Volume attention over broad top-100 universes: some configurations nearly exhausted the account despite the market gate. Attention is used only in the seasoned top-10 weekly rule. Extending it to every clock weakened visible performance. - New-listing overlays: occasional gains but he","holdup":"The proposed mechanism uses persistent relative trends, changing liquidity leadership, broad participation and the geometry of rebalancing. None requires a particular coin, date or remembered market event. Rising volume can distinguish fresh short-horizon participation from a stale price advance; the square root limits how much this proxy affects the weekly rule. Short clocks can recognize leadership before long clocks, while long clocks can keep exposure through short corrections. When choices differ, equal capital limits dependence on a single estimate and can earn excess growth from their relative fluctuations. Exposure falls when fewer clocks qualify, reducing the risk of assigning all capital to a solitary weak-regime forecast. The fee-derived band prices turnover explicitly. Dense targets make the same code usable for a fresh account at an arbitrary start. Evidence is conditional rather than conclusive: all three visible annual intervals were profitable, the additional earlier interval was profitable, execution-cost stress tests remained profitable, and cutoff checks established causal implementation. Against that stand a very large search, substantial liquidity-cutoff sensit","result":{"dev":{"final":2538.41419610661,"max_drawdown":-0.43740598937157427,"trades":1053,"exposure":0.3713135899779583,"assets_traded":32,"fees_paid":250.4693885501674},"hidden":{"final":321.5663214261542,"max_drawdown":-0.3046994497052603,"trades":705,"exposure":0.3696275673079411,"assets_traded":16,"fees_paid":40.63869320546174},"full":{"final":8162.685152979319,"max_drawdown":-0.43740598937157404,"trades":1758,"exposure":0.37063955024679046,"assets_traded":39,"fees_paid":1282.0477459898214}},"freq":"1d","equity":[["2021-10-17",98.387],["2021-10-24",114.5356],["2021-10-31",170.9744],["2021-11-07",185.9305],["2021-11-14",182.0663],["2021-11-21",243.8755],["2021-11-28",345.4505],["2021-12-05",308.291],["2021-12-12",282.2306],["2021-12-19",278.5556],["2021-12-26",320.1455],["2022-01-02",305.5831],["2022-01-09",305.5831],["2022-01-16",305.5831],["2022-01-23",305.5831],["2022-01-30",305.5831],["2022-02-06",305.5831],["2022-02-13",305.5831],["2022-02-20",305.5831],["2022-02-27",305.5831],["2022-03-06",305.5831],["2022-03-13",305.5831],["2022-03-20",305.5831],["2022-03-27",305.5831],["2022-04-03",305.5831],["2022-04-10",305.5831],["2022-04-17",305.5831],["2022-04-24",305.5831],["2022-05-01",305.5831],["2022-05-08",305.5831],["2022-05-15",305.5831],["2022-05-22",305.5831],["2022-05-29",305.5831],["2022-06-05",305.5831],["2022-06-12",305.5831],["2022-06-19",305.5831],["2022-06-26",305.5831],["2022-07-03",305.5831],["2022-07-10",305.5831],["2022-07-17",305.5831],["2022-07-24",305.5831],["2022-07-31",305.5831],["2022-08-07",305.5831],["2022-08-14",305.5831],["2022-08-21",305.5831],["2022-08-28",305.5831],["2022-09-04",305.5831],["2022-09-11",305.5831],["2022-09-18",305.5831],["2022-09-25",305.5831],["2022-10-02",305.5831],["2022-10-09",305.5831],["2022-10-16",305.5831],["2022-10-23",305.5831],["2022-10-30",305.5831],["2022-11-06",305.5831],["2022-11-13",305.5831],["2022-11-20",305.5831],["2022-11-27",305.5831],["2022-12-04",305.5831],["2022-12-11",305.5831],["2022-12-18",305.5831],["2022-12-25",305.5831],["2023-01-01",305.5831],["2023-01-08",305.5831],["2023-01-15",303.5315],["2023-01-22",320.2722],["2023-01-29",411.284],["2023-02-05",433.8551],["2023-02-12",384.7895],["2023-02-19",442.7269],["2023-02-26",380.097],["2023-03-05",336.1515],["2023-03-12",313.1645],["2023-03-19",623.9537],["2023-03-26",570.5807],["2023-04-02",675.2791],["2023-04-09",617.0223],["2023-04-16",648.1015],["2023-04-23",577.8058],["2023-04-30",592.3756],["2023-05-07",588.4358],["2023-05-14",567.0705],["2023-05-21",570.4424],["2023-05-28",574.4632],["2023-06-04",603.7801],["2023-06-11",606.0013],["2023-06-18",606.0013],["2023-06-25",606.0013],["2023-07-02",606.0013],["2023-07-09",606.0013],["2023-07-16",584.4532],["2023-07-23",608.9503],["2023-07-30",605.7122],["2023-08-06",601.8263],["2023-08-13",601.8263],["2023-08-20",592.1067],["2023-08-27",592.1067],["2023-09-03",592.1067],["2023-09-10",592.1067],["2023-09-17",592.1067],["2023-09-24",592.1067],["2023-10-01",592.1067],["2023-10-08",592.1067],["2023-10-15",592.1067],["2023-10-22",658.1751],["2023-10-29",698.6659],["2023-11-05",843.8212],["2023-11-12",982.2657],["2023-11-19",1009.2275],["2023-11-26",990.1664],["2023-12-03",1056.221],["2023-12-10",1284.1883],["2023-12-17",1509.4069],["2023-12-24",1920.1923],["2023-12-31",1739.4816],["2024-01-07",1552.0903],["2024-01-14",1649.5035],["2024-01-21",1544.6611],["2024-01-28",1532.2982],["2024-02-04",1503.4189],["2024-02-11",1665.7533],["2024-02-18",1686.5631],["2024-02-25",1658.6893],["2024-03-03",1845.1718],["2024-03-10",2474.7331],["2024-03-17",2481.9942],["2024-03-24",2423.9052],["2024-03-31",2687.6428],["2024-04-07",2446.902],["2024-04-14",2117.6582],["2024-04-21",2150.488],["2024-04-28",2138.147],["2024-05-05",2145.3083],["2024-05-12",2123.0519],["2024-05-19",2249.2853],["2024-05-26",2820.7052],["2024-06-02",2809.7529],["2024-06-09",2536.1327],["2024-06-16",2460.6718],["2024-06-23",2501.0352],["2024-06-30",2501.0352],["2024-07-07",2501.0352],["2024-07-14",2501.0352],["2024-07-21",2501.0352],["2024-07-28",2501.0352],["2024-08-04",2501.0352],["2024-08-11",2501.0352],["2024-08-18",2501.0352],["2024-08-25",2501.0352],["2024-09-01",2501.0352],["2024-09-08",2501.0352],["2024-09-15",2501.0352],["2024-09-22",2501.0352],["2024-09-29",2629.4627],["2024-10-06",2538.4142],["2024-10-13",2538.4142],["2024-10-20",2557.3791],["2024-10-27",2390.8802],["2024-11-03",2462.1304],["2024-11-10",2711.1072],["2024-11-17",4300.4118],["2024-11-24",4956.806],["2024-12-01",5411.713],["2024-12-08",6660.6039],["2024-12-15",5664.4406],["2024-12-22",5204.5638],["2024-12-29",5102.7227],["2025-01-05",5930.8915],["2025-01-12",5747.3579],["2025-01-19",6791.5209],["2025-01-26",6473.7178],["2025-02-02",6228.059],["2025-02-09",5738.4515],["2025-02-16",6104.3437],["2025-02-23",5903.8479],["2025-03-02",5582.3717],["2025-03-09",5582.3717],["2025-03-16",5582.3717],["2025-03-23",5582.3717],["2025-03-30",5582.3717],["2025-04-06",5582.3717],["2025-04-13",5582.3717],["2025-04-20",5582.3717],["2025-04-27",5582.3717],["2025-05-04",5582.3717],["2025-05-11",5979.8309],["2025-05-18",5631.436],["2025-05-25",5388.1026],["2025-06-01",5291.1142],["2025-06-08",5140.9178],["2025-06-15",5013.2414],["2025-06-22",5013.2414],["2025-06-29",5013.2414],["2025-07-06",5013.2414],["2025-07-13",5223.4742],["2025-07-20",5739.2135],["2025-07-27",6016.4233],["2025-08-03",5155.7001],["2025-08-10",5843.42],["2025-08-17",5935.4027],["2025-08-24",5863.6986],["2025-08-31",5265.808],["2025-09-07",5190.3478],["2025-09-14",5818.0038],["2025-09-21",5725.1683],["2025-09-28",5340.1455],["2025-10-05",5902.3981],["2025-10-12",5875.1019],["2025-10-19",5666.0361],["2025-10-26",5666.0361],["2025-11-02",5666.0361],["2025-11-09",5666.0361],["2025-11-16",5666.0361],["2025-11-23",5666.0361],["2025-11-30",5666.0361],["2025-12-07",5666.0361],["2025-12-14",5666.0361],["2025-12-21",5666.0361],["2025-12-28",5666.0361],["2026-01-04",5666.0361],["2026-01-11",5666.0361],["2026-01-18",5666.0361],["2026-01-25",5666.0361],["2026-02-01",5666.0361],["2026-02-08",5666.0361],["2026-02-15",5666.0361],["2026-02-22",5666.0361],["2026-03-01",5666.0361],["2026-03-08",5666.0361],["2026-03-15",5666.0361],["2026-03-22",5666.0361],["2026-03-29",5666.0361],["2026-04-05",5666.0361],["2026-04-12",5666.0361],["2026-04-19",5666.0361],["2026-04-26",5666.0361],["2026-05-03",5666.0361],["2026-05-10",5666.0361],["2026-05-17",5666.0361],["2026-05-24",5666.0361],["2026-05-31",5666.0361],["2026-06-07",5666.0361],["2026-06-14",5666.0361],["2026-06-21",5666.0361],["2026-06-28",5666.0361],["2026-07-05",5666.0361],["2026-07-12",5666.0361],["2026-07-19",5666.0361],["2026-07-26",5666.0361],["2026-08-02",5666.0361],["2026-08-09",5666.0361],["2026-08-16",5666.0361],["2026-08-23",6089.8275],["2026-08-30",6022.6372],["2026-09-06",6598.5667],["2026-09-13",6884.0823],["2026-09-20",8574.1228],["2026-09-27",10121.0612],["2026-10-04",8786.6316],["2026-10-11",8162.6852]],"code_bytes":6419},"a22":{"id":"a22","alias":"Dunlin","theme":"Fractional differentiation and stationarity-preserving features for prediction","started":"2026-10-11T00:32:01+00:00","rounds":[{"k":1,"minutes":37.7},{"k":2,"minutes":17.6}],"messages":[{"t":"2026-10-11T00:32:07.383Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test fractional-differenced signals against simpler momentum rules. I’ll track the variants and their tradeoffs in the writeup."},{"t":"2026-10-11T00:33:02.517Z","text":"The BTC trend baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. I’m now testing dynamic coin selection using recent volume and price history, including a fractional-differenced trend signal with its price-level dependence removed."},{"t":"2026-10-11T00:34:27.492Z","text":"Buying the strongest recent performers has lost money across the first 28 tests, including both fractional-differencing settings. I’m testing the reverse signal next and checking returns from the actual next-bar fills."},{"t":"2026-10-11T00:36:02.224Z","text":"The reverse signals also lose money when they select extreme movers. I’m testing a smaller, more liquid universe and a pooled model trained only on outcomes already known at each retraining point. Its inputs are return and volume ratios, not coin names or raw prices."},{"t":"2026-10-11T00:37:17.486Z","text":"Restricting momentum to the most liquid coins reduces the losses, but those variants still trail the BTC baseline. The prediction model is training now. I’m also checking fractional-difference signals on four-hour BTC, ETH and SOL bars, where trading costs are lower."},{"t":"2026-10-11T00:39:00.140Z","text":"The three-day model is the first approach to beat the baseline: its five-coin version reaches about $409 with a 37% drawdown under the BTC 200-day filter. The seven-day model performs much worse, so I’m treating the shorter-horizon result cautiously and testing nearby settings."},{"t":"2026-10-11T00:41:07.873Z","text":"The model’s $409 result depends strongly on which day the three-day rebalance cycle starts; the other two offsets reach about $83 and $182. I’ll keep the baseline as the accepted entry while testing smoother allocations and stateful entries that hold a trend until it breaks."},{"t":"2026-10-11T00:44:40.528Z","text":"I found a tie-breaking issue in the research code: equal tree predictions could select different coins depending on how excluded values were sorted. I’m fixing the ranking to break ties by trailing volume. I’m also testing a basket-aware target, \\(K\\log(1+R/K)\\), to value large gains and losses at the weight each coin would actually hold."},{"t":"2026-10-11T00:46:10.845Z","text":"I’ve submitted a simpler interim entry: daily rotation into the strongest 14-day performer among BTC, ETH and SOL, gated by BTC’s 200-day trend. The local test reaches about $652 with a 46% drawdown. It avoids the unstable coin selection seen in the broader prediction model."},{"t":"2026-10-11T00:46:45.896Z","text":"The interim submission was rejected because I tried to modify a read-only array returned by pandas. I’ve fixed that and am rerunning the referee checks. The original BTC baseline remains the accepted entry."},{"t":"2026-10-11T00:50:23.383Z","text":"The fractional filter has a useful result: requiring both BTC’s 200-day trend and a positive fractional-differenced trend lifts the rotation strategy to about $926, with a 33% drawdown and gains in all three visible yearly segments. I’m checking nearby filter settings before submitting it."},{"t":"2026-10-11T00:54:57.251Z","text":"The fractional strategy is accepted at $925.55. Nearby settings retain most of its growth, and doubled trading costs still leave about $787. One important weakness is upside concentration: removing the ten best days reduces the result to about $191. I’m recording that in the writeup while testing an adaptive mixture of signals."},{"t":"2026-10-11T00:58:57.489Z","text":"The adaptive mixtures stay below the current entry, around $768–$777 for the variants using 14-day momentum. Broadening the coin universe also fails to improve it. I’m now testing the same daily signals updated at each four-hour close, with a switching buffer to limit churn."},{"t":"2026-10-11T01:02:24.867Z","text":"The four-hour version passes the causal check and reaches about $964 with a 31% drawdown. That is a small gain over the daily entry, and it makes 260 trades instead of 162. I’m checking higher costs before deciding whether it earns its added complexity. Volatility-adaptive memory did not improve the daily strategy."},{"t":"2026-10-11T01:05:04.368Z","text":"A 1% momentum lead requirement before switching raises the result to about $1,032 and reduces trades from 162 to 134. The 0.2%, 0.5% and 2% buffers also improve on the unbuffered entry. With doubled costs, the 1% version retains about $903. I’m checking and submitting that revision."},{"t":"2026-10-11T01:08:58.047Z","text":"Submission #004 is accepted: $100 became $1,031.99, with a 33.4% drawdown and 134 trades. The writeup now covers the final equations, parameter search, subperiods, cost sensitivity and failed ideas. The final parameter checks still show gains across nearby settings."},{"t":"2026-10-11T01:09:46.174Z","text":"Entry: accepted **#004** — strategy.py, WRITEUP.md.  \nVisible years: **$100 → $1,031.99**; maximum drawdown **33.4%**; causal checks passed.  \nNext idea: purged walk-forward tail-risk prediction using fractional features."},{"t":"2026-10-11T01:09:53.750Z","text":"I’ll use the remaining time to test a causal downside-risk model against the accepted strategy. I’ll keep the current entry unless the new model shows a consistent improvement, then finish the writeup and verify the submitted file."},{"t":"2026-10-11T01:12:26.725Z","text":"The accepted file still matches submission 004 exactly. Its cost checks are encouraging: doubling fees leaves about $903, and quadrupling them leaves $690. I’m now testing whether a model trained only on past downside episodes can improve when the strategy holds cash."},{"t":"2026-10-11T01:14:11.654Z","text":"The downside models cut drawdowns, but they also missed too much upside. Across 24 exposure rules, the best visible result was about $978 versus the accepted strategy’s $1,032. I’ll keep the simpler entry while checking whether penalizing risk in the coin ranking gives a more consistent improvement."},{"t":"2026-10-11T01:16:03.000Z","text":"Risk penalties also reduced returns, so the learned model will stay out of the entry. A separate test of steadier trends across larger coin universes performed poorly. The useful result so far remains concentrated rotation among the three low-cost majors, with the fractional filter deciding when to move to cash."},{"t":"2026-10-11T01:18:37.080Z","text":"Changing the momentum memory or using ETH and SOL to define the market regime did not give a consistent gain. I’m making one final comparison: intraday exits with entry and exit buffers to reduce repeated trades near a threshold. Then I’ll lock the entry and update the writeup with the final-round results."},{"t":"2026-10-11T01:20:49.538Z","text":"The buffered four-hour version is a credible improvement: about $1,360 versus $1,032, with drawdown falling from 33.4% to 28.6%. It also improved all three fresh-account yearly slices and both cost stress tests. I’m checking nearby settings and causality before making it the final submission."},{"t":"2026-10-11T01:22:43.140Z","text":"Submission 005 is accepted and is now the final entry: $100 grew to $1,360.39 on the visible years, with 28.6% maximum drawdown and 144 trades. Causality checks passed, and nearby buffer settings also improved on the daily version. I’m finishing the equations, search history and limitations in the writeup."},{"t":"2026-10-11T01:26:23.705Z","text":"The writeup is complete with every final signal equation, about 780 explored variants, the failed approaches and the cost checks. The final rule leaves $1,178 at twice the stated costs and $883 at four times costs. I also verified that `strategy.py` exactly matches the accepted submission."},{"t":"2026-10-11T01:27:22.975Z","text":"Accepted submission **005** is the final strategy: four-hour BTC/ETH/SOL rotation using fractional market memory and entry, exit and switching buffers. It turned **$100 into $1,360.39 (×13.604)** on the visible years, with **28.6% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with the equations, approximately 780 explored variants, robustness results, failed approaches and limits of the evidence for unseen years."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:20.961008+00:00","note":"Initial causal BTC 200-day trend baseline; research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:45:49.853768+00:00","note":"Liquid-major 14-day momentum rotation with causal BTC 200-day market gate; simpler interim entry after rejecting fragile broad-universe predictions.","status":"rejected","dev":null,"dd":null,"reason":"weights() failed on the market cut at 2022-08-17 00:00:00:\n  File \"<frozen runpy>\", line 88, in _run_code\n  File \"the simulator\", line 221, in <module>\n    main()\n  File \"the simulator"},{"n":"003","t":"2026-10-11T00:51:21.062168+00:00","note":"Fractional market-inertia gate plus liquid-major rotation. Local causal check passed: 25.55, 33.4% drawdown; nearby orders, kernel widths and 10-18 day signals retained most growth.","status":"ok","dev":925.5458914941535,"dd":-0.3339326439036744,"reason":""},{"n":"004","t":"2026-10-11T01:05:09.071789+00:00","note":"Fractional market inertia with 1% momentum switching buffer. Buffer neighbourhood improved growth and lowered turnover; doubled costs leave 02.57. Causal daily spot-only rules.","status":"ok","dev":1031.9928079783965,"dd":-0.3339326439036745,"reason":""},{"n":"005","t":"2026-10-11T01:21:56.929171+00:00","note":"Final: four-hour fractional market inertia, volatility-scaled market entry and exit buffers, coin trend buffers, and a 2 percent rotation buffer. Nearby settings improved all visible yearly slices and","status":"ok","dev":1360.389934939222,"dd":-0.2863936065111844,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted **#004** — strategy.py, WRITEUP.md.  \nVisible years: **$100 → $1,031.99**; maximum drawdown **33.4%**; causal checks passed.  \nNext idea: purged walk-forward tail-risk prediction using fractional features."},{"k":2,"text":"Accepted submission **005** is the final strategy: four-hour BTC/ETH/SOL rotation using fractional market memory and entry, exit and switching buffers. It turned **$100 into $1,360.39 (×13.604)** on the visible years, with **28.6% maximum drawdown**, and passed causality checks. WRITEUP.md is complete with the equations, approximately 780 explored variants, robustness results, failed approaches and limits of the evidence for unseen years."}],"title":"Dunlin (a22): fractional market inertia with buffered intraday rotation","idea":"The final accepted entry is submission 005, strategy.py, with FREQ = \"4h\". It rotates among the expressly permitted BTC, ETH and SOL, holding one coin at 100% or cash. Completed daily history supplies a zero-mass fractional BTC feature, a slow BTC trend, a volatility scale, and coin trend references. Each four-hour close updates those features with the current price. Separate entry and exit thresholds prevent small price reversals from repeatedly paying costs. A challenger also needs a 2% advantage in fourteen-day log momentum before replacing the held coin. There is no leverage, shorting, yield, fitted predictor, date-specific exception, external data, or other named-coin selection. On the visible years, the referee reports $100 → $1,360.39, ×13.604, 138.7% CAGR, and 28.6% maximum drawdown, after all stated costs. These results do not establish that the million-dollar goal is attainable.","failed":"- Extreme broad-universe momentum and fractional ranks: trailing-volume universes, 7–90-day returns, fractional orders 0.4/0.7, volatility normalization, spike caps and smaller baskets often chased exhausted moves. The final round's 48 persistence variants used positive momentum times directional efficiency, clipped daily-return sums, or a regression slope times R^2, on 14/28-day windows; volume ranks 20/50/100; and one/three positions with cost-scaled switch buffers. Even the best finished below $89. Increasing apparent smoothness did not make broad-universe continuation profitable after costs. - Extreme-loser reversal: negative past 1–14-day returns and negative fractional scores mostly selected continuing declines; turnover compounded the damage. Four-hour fractional reversal in the majors also failed. - Pooled walk-forward mean prediction: a shallow histogram gradient booster used return horizons, fractional features, volatility, volume, range position, drawdown, age and market context. Training windows were 730 days, refits every 60 bars, seed 22, depth 3, minimum leaf 300, with labels purged for next-open execution. A three-day model reached about $409 in one configuration, but other three-day rebalance offsets gave about $83 and $182. Seven-day prediction was poor. Stable tie handling did not fix this cadence dependence. - Basket-aware prediction targets: K\\log(1+R/K), with K=5 or 10, did not reliably improve outcomes. LightGBM's offline runtime library was unavailable; sklearn was used and no packages were downloaded. - Final-round downside prediction: classifiers estimated the label Y_{i,s}=\\mathbf1\\{\\min_{1\\le j\\le h}L_{i,s+j}/O_{i,s+1}<1-a\\}, with (h,a)=(7,0.10) or (14,0.15). They pooled seasoned coins from trailing-volume top 100, used stationary return-memo","holdup":"The hypothesis is persistence of market demand and relative strength. The slow market average avoids sustained declines; the fractional return-memory feature responds when a formerly positive path deteriorates while that average is still positive. The selected major must show its own trend and relative momentum. The two kinds of entry/exit bands and the switch buffer recognize that paying costs to react to tiny, reversible signal changes is economically unattractive. Four-hour evaluation supplies more exit opportunities while those bands limit churn. These mechanisms do not require the identity of a past winning altcoin. Price-unit invariance, causal context availability, the long-only constraint and cost avoidance are structural properties independent of this sample. Signals use return changes, relative price gaps and a past-volatility scale. They can adjust to different price levels without retraining. The permitted majors have the lowest stated cost tier, and the strategy is small, deterministic and easy to rerun. Its yearly slices, fee stress results and parameter neighbourhood are more useful evidence than its leaderboard position. However, profitability is not guaranteed by s","result":{"dev":{"final":1360.389934939222,"max_drawdown":-0.2863936065111844,"trades":144,"exposure":0.3555352798053528,"assets_traded":3,"fees_paid":93.31051539064266},"hidden":{"final":180.05178188792482,"max_drawdown":-0.29251512389588574,"trades":112,"exposure":0.3461187214611872,"assets_traded":3,"fees_paid":14.026884259693755},"full":{"final":2449.406318482053,"max_drawdown":-0.29251512389588563,"trades":256,"exposure":0.3517707192405988,"assets_traded":3,"fees_paid":284.13083704509046}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",118.1349],["2021-10-31",120.8365],["2021-11-07",152.3486],["2021-11-14",145.1752],["2021-11-21",134.9987],["2021-11-28",134.9987],["2021-12-05",134.9987],["2021-12-12",134.9987],["2021-12-19",134.9987],["2021-12-26",134.9987],["2022-01-02",134.9987],["2022-01-09",134.9987],["2022-01-16",134.9987],["2022-01-23",134.9987],["2022-01-30",134.9987],["2022-02-06",134.9987],["2022-02-13",134.9987],["2022-02-20",134.9987],["2022-02-27",134.9987],["2022-03-06",134.9987],["2022-03-13",134.9987],["2022-03-20",134.9987],["2022-03-27",134.9987],["2022-04-03",134.9987],["2022-04-10",134.9987],["2022-04-17",134.9987],["2022-04-24",134.9987],["2022-05-01",134.9987],["2022-05-08",134.9987],["2022-05-15",134.9987],["2022-05-22",134.9987],["2022-05-29",134.9987],["2022-06-05",134.9987],["2022-06-12",134.9987],["2022-06-19",134.9987],["2022-06-26",134.9987],["2022-07-03",134.9987],["2022-07-10",134.9987],["2022-07-17",134.9987],["2022-07-24",134.9987],["2022-07-31",134.9987],["2022-08-07",134.9987],["2022-08-14",134.9987],["2022-08-21",134.9987],["2022-08-28",134.9987],["2022-09-04",134.9987],["2022-09-11",134.9987],["2022-09-18",134.9987],["2022-09-25",134.9987],["2022-10-02",134.9987],["2022-10-09",134.9987],["2022-10-16",134.9987],["2022-10-23",134.9987],["2022-10-30",134.9987],["2022-11-06",134.9987],["2022-11-13",134.9987],["2022-11-20",134.9987],["2022-11-27",134.9987],["2022-12-04",134.9987],["2022-12-11",134.9987],["2022-12-18",134.9987],["2022-12-25",134.9987],["2023-01-01",134.9987],["2023-01-08",134.9987],["2023-01-15",172.2481],["2023-01-22",183.6035],["2023-01-29",186.5316],["2023-02-05",178.3628],["2023-02-12",171.0154],["2023-02-19",172.2559],["2023-02-26",148.5373],["2023-03-05",136.175],["2023-03-12",136.175],["2023-03-19",160.2845],["2023-03-26",157.8571],["2023-04-02",158.5146],["2023-04-09",164.0191],["2023-04-16",178.134],["2023-04-23",163.9544],["2023-04-30",153.1439],["2023-05-07",150.1867],["2023-05-14",144.7545],["2023-05-21",144.7545],["2023-05-28",144.7545],["2023-06-04",154.4658],["2023-06-11",138.4008],["2023-06-18",138.4008],["2023-06-25",150.3532],["2023-07-02",162.1287],["2023-07-09",178.2914],["2023-07-16",229.2078],["2023-07-23",208.523],["2023-07-30",206.4011],["2023-08-06",206.4011],["2023-08-13",206.1062],["2023-08-20",199.6379],["2023-08-27",199.6379],["2023-09-03",199.6379],["2023-09-10",199.6379],["2023-09-17",199.6379],["2023-09-24",199.6379],["2023-10-01",199.6379],["2023-10-08",198.8257],["2023-10-15",198.8257],["2023-10-22",237.3634],["2023-10-29",273.0217],["2023-11-05",337.0577],["2023-11-12",463.5578],["2023-11-19",506.2484],["2023-11-26",473.2376],["2023-12-03",480.7488],["2023-12-10",522.367],["2023-12-17",522.8647],["2023-12-24",818.7803],["2023-12-31",737.7981],["2024-01-07",705.0633],["2024-01-14",707.8563],["2024-01-21",700.2881],["2024-01-28",668.5066],["2024-02-04",695.4245],["2024-02-11",735.1232],["2024-02-18",780.8769],["2024-02-25",839.6341],["2024-03-03",944.1524],["2024-03-10",1026.5317],["2024-03-17",1384.4136],["2024-03-24",1245.6893],["2024-03-31",1393.4809],["2024-04-07",1251.3417],["2024-04-14",1180.6631],["2024-04-21",1180.6631],["2024-04-28",1208.4552],["2024-05-05",1156.5389],["2024-05-12",1156.5389],["2024-05-19",1290.1948],["2024-05-26",1408.7528],["2024-06-02",1373.7089],["2024-06-09",1354.0431],["2024-06-16",1267.2396],["2024-06-23",1267.2396],["2024-06-30",1267.2396],["2024-07-07",1267.2396],["2024-07-14",1267.2396],["2024-07-21",1470.456],["2024-07-28",1526.6371],["2024-08-04",1391.8139],["2024-08-11",1391.8139],["2024-08-18",1391.8139],["2024-08-25",1403.7831],["2024-09-01",1392.6834],["2024-09-08",1392.6834],["2024-09-15",1392.6834],["2024-09-22",1392.6834],["2024-09-29",1408.2152],["2024-10-06",1360.3899],["2024-10-13",1360.3899],["2024-10-20",1447.9764],["2024-10-27",1541.4626],["2024-11-03",1374.4451],["2024-11-10",1557.7128],["2024-11-17",1682.6946],["2024-11-24",1691.494],["2024-12-01",1677.352],["2024-12-08",1804.9325],["2024-12-15",1668.0524],["2024-12-22",1534.6784],["2024-12-29",1534.6784],["2025-01-05",1584.3717],["2025-01-12",1503.793],["2025-01-19",1847.7144],["2025-01-26",1757.077],["2025-02-02",1407.1014],["2025-02-09",1407.1014],["2025-02-16",1407.1014],["2025-02-23",1407.1014],["2025-03-02",1407.1014],["2025-03-09",1407.1014],["2025-03-16",1407.1014],["2025-03-23",1407.1014],["2025-03-30",1407.1014],["2025-04-06",1407.1014],["2025-04-13",1407.1014],["2025-04-20",1407.1014],["2025-04-27",1451.3559],["2025-05-04",1438.5139],["2025-05-11",1896.6219],["2025-05-18",1821.1358],["2025-05-25",1912.9841],["2025-06-01",1751.2939],["2025-06-08",1779.9166],["2025-06-15",1730.7498],["2025-06-22",1681.7921],["2025-06-29",1653.4281],["2025-07-06",1529.2049],["2025-07-13",1799.0452],["2025-07-20",2256.7006],["2025-07-27",2303.8755],["2025-08-03",2115.2609],["2025-08-10",2352.2916],["2025-08-17",2468.9586],["2025-08-24",2293.0802],["2025-08-31",2207.3171],["2025-09-07",2207.3171],["2025-09-14",2337.9605],["2025-09-21",2294.3847],["2025-09-28",2136.9343],["2025-10-05",2240.8584],["2025-10-12",2070.5209],["2025-10-19",2070.5209],["2025-10-26",2070.5209],["2025-11-02",2070.5209],["2025-11-09",2070.5209],["2025-11-16",2070.5209],["2025-11-23",2070.5209],["2025-11-30",2070.5209],["2025-12-07",2070.5209],["2025-12-14",2070.5209],["2025-12-21",2070.5209],["2025-12-28",2070.5209],["2026-01-04",2070.5209],["2026-01-11",2070.5209],["2026-01-18",2070.5209],["2026-01-25",2070.5209],["2026-02-01",2070.5209],["2026-02-08",2070.5209],["2026-02-15",2070.5209],["2026-02-22",2070.5209],["2026-03-01",2070.5209],["2026-03-08",2070.5209],["2026-03-15",2070.5209],["2026-03-22",2070.5209],["2026-03-29",2070.5209],["2026-04-05",2070.5209],["2026-04-12",2070.5209],["2026-04-19",2070.5209],["2026-04-26",2070.5209],["2026-05-03",2070.5209],["2026-05-10",2070.5209],["2026-05-17",2070.5209],["2026-05-24",2070.5209],["2026-05-31",2070.5209],["2026-06-07",2070.5209],["2026-06-14",2070.5209],["2026-06-21",2070.5209],["2026-06-28",2070.5209],["2026-07-05",2070.5209],["2026-07-12",2070.5209],["2026-07-19",2070.5209],["2026-07-26",2070.5209],["2026-08-02",2070.5209],["2026-08-09",2070.5209],["2026-08-16",2070.5209],["2026-08-23",2244.8779],["2026-08-30",2333.6613],["2026-09-06",2326.1632],["2026-09-13",2294.985],["2026-09-20",2339.7938],["2026-09-27",2606.2909],["2026-10-04",2565.4054],["2026-10-11",2449.4063]],"code_bytes":1960},"a23":{"id":"a23","alias":"Lapwing","theme":"Hurst exponent, variance ratios and fractal measures to switch between trend and reversion","started":"2026-10-11T00:32:16+00:00","rounds":[{"k":1,"minutes":38.5},{"k":2,"minutes":18.1}],"messages":[{"t":"2026-10-11T00:32:25.048Z","text":"I’ll inspect the simulator and submit a simple baseline early, then test whether variance ratios and path roughness can improve a causal momentum strategy. I’ll track the variants and failures in the writeup."},{"t":"2026-10-11T00:32:53.482Z","text":"The simulator fills at the next open and checks causality by rerunning truncated data. The BTC baseline is awaiting the referee. I’ll compare liquid-coin momentum with measures of path efficiency and return persistence, checking separate periods as well as the full result."},{"t":"2026-10-11T00:34:25.860Z","text":"The BTC baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first momentum tests are losing heavily after selecting recent winners. I’m checking their trade timing before drawing a conclusion from those results."},{"t":"2026-10-11T00:36:28.018Z","text":"The trade timing checks out. In this sample, the most extreme recent winners and losers both tend to perform poorly immediately afterward. I’m widening the tests to smoother trends and using variance ratios as conditional evidence, rather than assuming that every persistent price path is worth buying."},{"t":"2026-10-11T00:40:35.438Z","text":"The daily forecasting model hasn’t improved on the BTC baseline. I’m now testing a different holding rule: enter a fresh breakout and keep the position until its trend or trailing stop fails. That may avoid rotating into coins after their largest move has already happened."},{"t":"2026-10-11T00:44:57.854Z","text":"The most stable improvement so far is major-coin trend rotation: roughly $330 from $100, with about 35% drawdown. A trend-and-pullback rule reached roughly $1,400, but its result depends heavily on one period. I’m testing whether a causal allocator can switch between these families using only their recent simulated performance."},{"t":"2026-10-11T00:48:42.593Z","text":"I’ve submitted the adaptive mixture. Locally, it turned $100 into about $887 with 48% drawdown, improved on the baseline, and held up better across periods than the single-coin pullback rule. The referee’s causality check is pending. Next I’ll check costs and nearby settings before considering more concentration."},{"t":"2026-10-11T00:51:32.182Z","text":"The adaptive entry passed the referee: $100 became $886.54. Adding a filter that excludes experts with negative returns over the past month improved all three visible-year blocks in the local test, reaching about $1,569 with 42% drawdown. I’ll check nearby month lengths and higher exposure before submitting it."},{"t":"2026-10-11T00:55:51.545Z","text":"I’ve submitted the monthly performance filter with daily responses to exits; it reached about $2,079 locally. The stress tests also exposed a limitation in the previous entry: delaying fills by an extra day removed its profit, and its ten best days accounted for essentially all its growth. Doubling costs still left it profitable."},{"t":"2026-10-11T00:57:35.343Z","text":"The new entry is accepted: $100 became $2,078.66, with 42.3% drawdown, placing tenth at acceptance. I’m checking exposure and trading cadence now. The aim is to improve growth while responding promptly to exits and avoiding unnecessary turnover."},{"t":"2026-10-11T00:59:29.119Z","text":"The accepted entry remains profitable at twice the stated costs: $100 becomes about $1,323. Removing its ten best days from the compounded return leaves about $201. Larger cash-funded exposure improved the local result without much change in drawdown; I’m checking nearby exposure and concentration settings."},{"t":"2026-10-11T01:04:29.716Z","text":"The fourth candidate is submitted: it reached about $4,559 locally with 48% drawdown, and neighboring exposure settings gave similar results. I also verified the 15 internal paper accounts against the referee’s simulator; their equity paths match to floating-point precision. I’m finishing the stress checks and updating the writeup."},{"t":"2026-10-11T01:06:40.246Z","text":"The fourth entry is accepted: $100 became $4,559.30, ranking seventh at acceptance. Nearby parameter checks ranged from roughly $2,369 to $4,784, and doubling costs still produced about $2,606. The writeup will flag the main limitations: gains depend heavily on a few strong days, and an extra day of fill delay cuts the result to about $139."},{"t":"2026-10-11T01:10:46.972Z","text":"Entry: accepted #004, lapwing.py; writeup.  \nVisible years: **$100 → $4,559.30**, 48.1% max drawdown; causality check passed.  \nNext idea: test independent bear-market reversal experts with walk-forward checks."},{"t":"2026-10-11T01:10:53.374Z","text":"I’ll test whether an independent reversal expert improves the current strategy in difficult markets, then leave time for scoring and the writeup. The accepted $4,559 entry will remain in place unless the new tests support replacing it."},{"t":"2026-10-11T01:12:52.121Z","text":"The accepted strategy’s gain is concentrated in fast pullback trades: an extra day of execution delay cuts the visible result sharply. I’m testing a separate lower-wick reversal signal that performed better in the bear-market segment, to see whether it adds useful diversification."},{"t":"2026-10-11T01:14:07.188Z","text":"The reversal additions improve the full visible result only modestly, while weakening the bear-market and earlier-history results. I’m also checking all three starting phases of the pullback experts’ three-day refresh cycle; this will show whether the result depends on one arbitrary schedule."},{"t":"2026-10-11T01:15:34.657Z","text":"The schedule check exposed a weakness: the two alternate phases finish at about $375 and $873, versus $4,559 for the accepted phase. Averaging phases and changing the refresh interval also reduced returns. I’ll document that sensitivity and finish the diversification checks before deciding on the final entry."},{"t":"2026-10-11T01:18:23.814Z","text":"The extra reversal experts and risk penalties haven’t produced a clear improvement across periods. Volatility limits reduce drawdown but give up substantial growth. I’m finishing checks on the liquidity universe and whether asset-level variance ratios help more than the market-wide prior."},{"t":"2026-10-11T01:21:12.716Z","text":"A two-day minimum holding rule avoids the global schedule and returns $3,895, with stronger bear-market gains but a worse overall drawdown. I’m testing whether blending it with the accepted strategy improves the portfolio. A blend would spread trades across two timing rules."},{"t":"2026-10-11T01:22:29.131Z","text":"The equal blend improves the visible result to $5,530, with drawdown falling from 48.1% to 44.6%. It also improves the earlier-history result and reduces its drawdown from 66.9% to 51.2%. I’m submitting the blend, then updating the equations and final validation results."},{"t":"2026-10-11T01:24:49.872Z","text":"Submission 005 is accepted: $100 → $5,530.30, with 44.6% maximum drawdown. With execution costs doubled, it finishes at $3,284.95; an extra day of delay cuts that to $221.47. I’m retaining this entry and documenting the timing sensitivity, search count, and exact rules."},{"t":"2026-10-11T01:28:23.826Z","text":"The final entry is byte-for-byte identical to the accepted snapshot, and the writeup now covers both timing rules and the blend. Changing the scheduled half’s phase gives $1,528 or $2,297: better than the original pool’s alternate phases, but still a substantial sensitivity. Those results are included in the writeup."},{"t":"2026-10-11T01:28:52.134Z","text":"My final entry is **accepted submission 005**, lapwing.py, which blends two fee-aware trend/pullback expert pools with different timing rules and a variance-ratio prior. Visible result: **$100 → $5,530.30 (55.30×), with 44.6% maximum drawdown**. WRITEUP.md is complete with equations, 769 logged evaluations, results, failed ideas, and limitations. Performance remains sensitive to execution timing and rare large gains."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:38.428473+00:00","note":"Initial causal BTC trend baseline; research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:48:38.148210+00:00","note":"Persistence-conditioned, cost-aware allocation between major-coin trends and liquid-coin pullbacks; 120-day memory, broad posterior.","status":"ok","dev":886.5425573980995,"dd":-0.476482333126377,"reason":""},{"n":"003","t":"2026-10-11T00:55:45.937697+00:00","note":"Causal monthly net-performance gate for experts, daily response to exits; unchanged broad fractal prior and 120-day growth memory.","status":"ok","dev":2078.6594258813266,"dd":-0.4231361174129207,"reason":""},{"n":"004","t":"2026-10-11T01:02:47.737393+00:00","note":"Final round-one candidate: monthly expert loss gate, tempered quarter-growth allocation, daily targets, cash-funded exposure capped at 100%; nearby settings remain profitable.","status":"ok","dev":4559.298352394875,"dd":-0.4808866690538278,"reason":""},{"n":"005","t":"2026-10-11T01:22:31.148853+00:00","note":"Equal blend of scheduled and entry-timed pullback expert pools; causal fee-aware allocation and weak variance-ratio prior. Diversifies refresh timing and improves earlier, bear and full visible result","status":"ok","dev":5530.300523925712,"dd":-0.44556562840118774,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #004, lapwing.py; writeup.  \nVisible years: **$100 → $4,559.30**, 48.1% max drawdown; causality check passed.  \nNext idea: test independent bear-market reversal experts with walk-forward checks."},{"k":2,"text":"My final entry is **accepted submission 005**, lapwing.py, which blends two fee-aware trend/pullback expert pools with different timing rules and a variance-ratio prior. Visible result: **$100 → $5,530.30 (55.30×), with 44.6% maximum drawdown**. WRITEUP.md is complete with equations, 769 logged evaluations, results, failed ideas, and limitations. Performance remains sensitive to execution timing and rare large gains."}],"title":"Lapwing (a23): timing-diversified persistence-conditioned expert allocation","idea":"Crypto can reward a continuing trend and then abruptly reward buying a pullback. A single cross-sectional momentum rank was particularly bad in my experiments: it repeatedly bought terminal price spikes. Lapwing instead runs two small, generic families of paper strategies, including their next-open execution costs, and allocates to their recent performance. One family rotates among BTC, ETH and SOL in established trends. The other buys liquid coins whose weekly trend is positive but whose latest daily move is weak. A trailing variance ratio shifts the prior between the two families. An expert with a negative trailing-month net return is suspended. Cash is an explicit competing expert. Two independently evaluated pools use different pullback timing: a fixed three-bar refresh and a rule tied to each position's entry. The final portfolio trades the equal average of their targets. This spreads decisions across two timing rules. Every statistic, including the paper strategies' profits, is computed causally. There is no fitted asset identifier, fixed list of other coins, date trigger, precomputed price, external model, or random component.","failed":"- Unfiltered strongest-coin momentum lost almost everything after repeatedly buying sharp speculative peaks. Simple breadth/BTC gates helped but did not establish a useful edge. - Buying the largest recent losers without a confirmed continuing trend also failed net of turnover costs; apparent cheapness often preceded another decline. - Variance-ratio gates on major coins and individual-coin variance-ratio/path-efficiency ranks did not reliably rescue weak strategies. Persistence alone says nothing about profitable entry price. - A direct forecast \\mu_L+\\beta(\\mathrm{VR}(2)-1)r_t was too noisy and costly. - The causal histogram-boosting model used only prior observable labels and no asset IDs, but its best tested allocation underperformed the BTC baseline. LightGBM could not load its system OpenMP dependency, so the executable experiment used scikit-learn's histogram model. - Breakout entries with trailing range stops worked much better in the earlier development interval than in the visible years and were discarded. - Generic new-listing trades and confirmed-shock reversals were mostly unprofitable. One seven-observation-age listing variant was profitable visibly but weak earlier, so it was not added to the expert pool. - The largest single pullback result, about $1,394, had roughly 80% drawdown and was heavily concentrated in one block. It was not submitted by itself. - Concentrating the mixture on one top expert and aggressively penalizing expert drawdown did not consistently improve the retained broad allocation. - Slow actual portfolio refreshes at five or seven bars were poor unless a separate target-change trigger preserved fast exits. Daily targets were retained for a simpler execution rule. - A few all-cash subperiod variants could not be scored because the simu","holdup":"The proposed mechanism is recurring price discovery and persistent versus reversing order flow, not a forecast about a named token. The liquid universe updates as assets enter and leave. Momentum experts can exploit a continuing major-coin rally, while pullback experts can exploit a different kind of market without requiring a manually scheduled regime change. Within each timing pool, the strategy observes recent net success, suspends recently losing experts, and cash can absorb allocation when neither family provides useful exposure. The equal blend spreads timing risk across a fixed refresh and a position-dependent rule; it improved the first two annual blocks and earlier history while sacrificing some of the final visible block's return. It does not remove all sensitivity to entry timing. The weak, aggregated persistence prior avoids giving a statistically fragile single-coin Hurst estimate complete control. This argument is a hypothesis. Expert returns are correlated, past profits can reverse, crypto liquidity and transaction costs can change, and the mixture still has material drawdown. The rare-gain concentration, severe delay sensitivity, and earlier 51.2% drawdown are pract","result":{"dev":{"final":5530.300523925707,"max_drawdown":-0.44556562840118796,"trades":5066,"exposure":0.42005179138446436,"assets_traded":339,"fees_paid":968.6764152642933},"hidden":{"final":27.39409832371975,"max_drawdown":-0.7730024624203292,"trades":2694,"exposure":0.4453988497904344,"assets_traded":280,"fees_paid":20.985050065995342},"full":{"final":1514.244745452861,"max_drawdown":-0.8115659859293975,"trades":7767,"exposure":0.43018506226965497,"assets_traded":457,"fees_paid":2131.0385517145087}},"freq":"1d","equity":[["2021-10-17",111.261],["2021-10-24",112.1096],["2021-10-31",108.7961],["2021-11-07",132.9543],["2021-11-14",124.2858],["2021-11-21",112.5973],["2021-11-28",99.2718],["2021-12-05",99.2718],["2021-12-12",99.2718],["2021-12-19",99.2718],["2021-12-26",100.3609],["2022-01-02",102.5469],["2022-01-09",102.5469],["2022-01-16",102.5469],["2022-01-23",102.5469],["2022-01-30",102.5469],["2022-02-06",102.5469],["2022-02-13",97.6116],["2022-02-20",92.4193],["2022-02-27",92.4193],["2022-03-06",91.9571],["2022-03-13",91.9571],["2022-03-20",94.1393],["2022-03-27",128.0222],["2022-04-03",200.8043],["2022-04-10",193.3115],["2022-04-17",193.1652],["2022-04-24",193.1652],["2022-05-01",193.1652],["2022-05-08",193.1652],["2022-05-15",193.1652],["2022-05-22",193.1652],["2022-05-29",193.1652],["2022-06-05",193.1652],["2022-06-12",193.1652],["2022-06-19",193.1652],["2022-06-26",193.1652],["2022-07-03",193.1652],["2022-07-10",193.1652],["2022-07-17",193.1652],["2022-07-24",193.1512],["2022-07-31",190.7327],["2022-08-07",181.1631],["2022-08-14",188.4959],["2022-08-21",161.8458],["2022-08-28",161.5972],["2022-09-04",161.5972],["2022-09-11",161.5972],["2022-09-18",157.4816],["2022-09-25",157.4816],["2022-10-02",157.4816],["2022-10-09",156.5916],["2022-10-16",156.5916],["2022-10-23",156.5916],["2022-10-30",165.1323],["2022-11-06",186.7751],["2022-11-13",148.0697],["2022-11-20",148.0697],["2022-11-27",148.0697],["2022-12-04",148.0697],["2022-12-11",148.0697],["2022-12-18",152.9896],["2022-12-25",152.9896],["2023-01-01",152.9896],["2023-01-08",152.9896],["2023-01-15",255.3427],["2023-01-22",251.7247],["2023-01-29",322.6012],["2023-02-05",349.569],["2023-02-12",346.6298],["2023-02-19",371.396],["2023-02-26",374.1895],["2023-03-05",385.9612],["2023-03-12",388.1918],["2023-03-19",507.5595],["2023-03-26",494.8312],["2023-04-02",940.8016],["2023-04-09",855.1401],["2023-04-16",930.2558],["2023-04-23",878.1002],["2023-04-30",803.4512],["2023-05-07",768.3739],["2023-05-14",760.0768],["2023-05-21",760.0768],["2023-05-28",760.0768],["2023-06-04",716.8081],["2023-06-11",716.8081],["2023-06-18",716.8081],["2023-06-25",678.3832],["2023-07-02",826.63],["2023-07-09",1145.314],["2023-07-16",993.9368],["2023-07-23",906.8635],["2023-07-30",874.9862],["2023-08-06",860.834],["2023-08-13",863.9916],["2023-08-20",851.7281],["2023-08-27",851.7281],["2023-09-03",851.7281],["2023-09-10",851.7281],["2023-09-17",851.7281],["2023-09-24",780.9565],["2023-10-01",780.8469],["2023-10-08",849.8956],["2023-10-15",802.4766],["2023-10-22",1000.2201],["2023-10-29",1078.9354],["2023-11-05",1445.0887],["2023-11-12",1917.8629],["2023-11-19",1993.3143],["2023-11-26",1984.4351],["2023-12-03",2134.4211],["2023-12-10",2395.5604],["2023-12-17",2312.7636],["2023-12-24",3381.0052],["2023-12-31",3193.3503],["2024-01-07",2938.3575],["2024-01-14",3018.7807],["2024-01-21",2950.4924],["2024-01-28",2950.4924],["2024-02-04",2898.3],["2024-02-11",3497.4468],["2024-02-18",3749.69],["2024-02-25",3961.0837],["2024-03-03",4879.5432],["2024-03-10",5975.3047],["2024-03-17",6198.8942],["2024-03-24",5033.663],["2024-03-31",5761.9394],["2024-04-07",5792.2869],["2024-04-14",5208.41],["2024-04-21",5240.6682],["2024-04-28",5258.9317],["2024-05-05",5258.9317],["2024-05-12",5366.3724],["2024-05-19",5390.1851],["2024-05-26",6368.6417],["2024-06-02",6727.042],["2024-06-09",7600.5857],["2024-06-16",6879.9135],["2024-06-23",6879.9135],["2024-06-30",6879.9135],["2024-07-07",6879.9135],["2024-07-14",6879.9135],["2024-07-21",6701.5238],["2024-07-28",7045.5939],["2024-08-04",5804.1726],["2024-08-11",5804.1726],["2024-08-18",5804.1726],["2024-08-25",5804.1726],["2024-09-01",5804.0488],["2024-09-08",5804.0488],["2024-09-15",5790.6398],["2024-09-22",5649.9008],["2024-09-29",5708.2608],["2024-10-06",5540.6132],["2024-10-13",5498.629],["2024-10-20",5543.8133],["2024-10-27",5225.3202],["2024-11-03",4874.4479],["2024-11-10",6128.647],["2024-11-17",5411.6847],["2024-11-24",6053.1123],["2024-12-01",6143.3748],["2024-12-08",5989.157],["2024-12-15",5760.0154],["2024-12-22",5282.2618],["2024-12-29",5926.4154],["2025-01-05",5147.9639],["2025-01-12",4326.0144],["2025-01-19",4719.9056],["2025-01-26",4607.9619],["2025-02-02",3802.3074],["2025-02-09",3798.5051],["2025-02-16",3798.5051],["2025-02-23",3798.5051],["2025-03-02",3798.5051],["2025-03-09",3798.5051],["2025-03-16",3798.5051],["2025-03-23",3798.5051],["2025-03-30",3782.4674],["2025-04-06",3782.4674],["2025-04-13",3782.4674],["2025-04-20",3782.4674],["2025-04-27",3822.009],["2025-05-04",3671.1346],["2025-05-11",4658.3658],["2025-05-18",3548.5513],["2025-05-25",3551.6602],["2025-06-01",3025.1186],["2025-06-08",2800.6637],["2025-06-15",2582.4891],["2025-06-22",2590.8016],["2025-06-29",2689.6771],["2025-07-06",2392.7415],["2025-07-13",2075.1613],["2025-07-20",1930.6971],["2025-07-27",2010.731],["2025-08-03",1826.2663],["2025-08-10",2280.3187],["2025-08-17",2366.6852],["2025-08-24",2557.4388],["2025-08-31",2342.9187],["2025-09-07",2327.1094],["2025-09-14",2728.1122],["2025-09-21",2695.4712],["2025-09-28",2419.9009],["2025-10-05",2399.3695],["2025-10-12",2312.6216],["2025-10-19",2312.6216],["2025-10-26",2312.6216],["2025-11-02",2312.6216],["2025-11-09",2312.6216],["2025-11-16",2312.6216],["2025-11-23",2312.6216],["2025-11-30",2312.6216],["2025-12-07",2312.6216],["2025-12-14",2306.4777],["2025-12-21",2306.4777],["2025-12-28",2306.4777],["2026-01-04",2306.3531],["2026-01-11",2300.5595],["2026-01-18",2284.0228],["2026-01-25",2272.7109],["2026-02-01",2272.7109],["2026-02-08",2272.7109],["2026-02-15",2272.7109],["2026-02-22",2272.7109],["2026-03-01",2272.7109],["2026-03-08",2272.7109],["2026-03-15",2272.7109],["2026-03-22",2098.5828],["2026-03-29",2098.5828],["2026-04-05",2098.5828],["2026-04-12",2233.2248],["2026-04-19",2881.0525],["2026-04-26",2352.0493],["2026-05-03",2673.8349],["2026-05-10",2442.0193],["2026-05-17",1945.6884],["2026-05-24",1916.0741],["2026-05-31",1807.5259],["2026-06-07",1807.5259],["2026-06-14",1807.5259],["2026-06-21",1807.5259],["2026-06-28",1807.5259],["2026-07-05",1821.1093],["2026-07-12",1800.3397],["2026-07-19",1800.064],["2026-07-26",1800.4397],["2026-08-02",1775.8672],["2026-08-09",1753.8383],["2026-08-16",1582.6367],["2026-08-23",1654.516],["2026-08-30",1770.9585],["2026-09-06",1724.8166],["2026-09-13",1658.7184],["2026-09-20",1548.6907],["2026-09-27",1684.6881],["2026-10-04",1659.8318],["2026-10-11",1514.2447]],"code_bytes":9552},"a24":{"id":"a24","alias":"Dotterel","theme":"Entropy and information-theoretic signals (predictability as a gate for trading)","started":"2026-10-11T00:32:31+00:00","rounds":[{"k":1,"minutes":38.0},{"k":2,"minutes":7.7}],"messages":[{"t":"2026-10-11T00:32:38.617Z","text":"I’ll inspect the simulator and data, submit an early causal baseline, then test whether information-based gates improve momentum selection after costs. I’ll track the variants and document the final rules in `WRITEUP.md`."},{"t":"2026-10-11T00:33:11.186Z","text":"The BTC baseline is submitted. I’m testing whether persistent buying across several days distinguishes durable momentum from one-day spikes. The strategy will select coins from trailing liquidity ranks and use only information available at each daily close."},{"t":"2026-10-11T00:34:52.399Z","text":"The baseline is accepted: $100 became $230.64, with a 25.7% drawdown. Broad momentum selection performed badly in the first tests: it repeatedly bought coins after sharp price and volume spikes, then caught their reversals. I’m testing tighter liquidity ranks and persistence filters to reduce that exposure."},{"t":"2026-10-11T00:37:00.042Z","text":"Rotating among BTC, ETH and SOL has been more resilient than the wider coin strategies. A trend rule weighted by directional information turned $100 into roughly $568 in the visible period. I’m testing it alongside a wider-universe approach that requires orderly trends before committing capital."},{"t":"2026-10-11T00:41:00.758Z","text":"The second entry is accepted, and it passed the causal checks: $100 became $567.75. The online return models tested so far have not earned enough to cover frequent trading costs. I’m focusing next on slower decisions and a daily market gate that controls exposure."},{"t":"2026-10-11T00:43:36.261Z","text":"A range-volatility version of the information gate reached $1,046 with a 29.3% drawdown. It estimates a roughly monthly trend, trades only when the drift estimate clears a confidence threshold, and sizes exposure with capped half-Kelly. I’m checking nearby parameters before submitting it."},{"t":"2026-10-11T00:45:14.500Z","text":"The third entry is accepted at $1,046.21. Nearby confidence thresholds and lookback lengths remained profitable, but returns vary enough that parameter choice is a real risk. I’m now checking finer bars and robustness measures to see how much the result depends on signal timing or a few large moves."},{"t":"2026-10-11T00:48:57.388Z","text":"The accepted entry still grew to $821 at double costs and $204 after removing its ten best days. Much of its profit came from SOL trends. A faster drift estimate has reached $1,320 locally; I’m validating it before the next submission."},{"t":"2026-10-11T00:51:13.470Z","text":"The fourth entry is accepted at $1,319.74, and the equations and search history are now in `WRITEUP.md`. It uses fewer trades than the prior entry and remains above $1,000 at double costs. I’m testing entry timing next, especially whether waiting for a modest pullback reduces buying after a sharp rise."},{"t":"2026-10-11T00:52:41.875Z","text":"Waiting for pullbacks reduced returns in these tests. Finer bars also increased turnover without improving results. The four-hour entry remains stronger on both measures, so I’m keeping it as the checkpoint while comparing a few sizing rules and recording the failed variants."},{"t":"2026-10-11T00:54:44.584Z","text":"Using entropy as an entry gate, then ranking by risk-adjusted drift, improved the local result to $1,557 with a 27.2% drawdown. It also improved all three visible-year subperiods. I’m checking the simplified code and its robustness results before submitting it."},{"t":"2026-10-11T00:55:55.126Z","text":"The fifth entry is accepted: $100 became $1,556.53, with a 27.2% drawdown and 239 trades. Double costs leave $1,238.64. I’m testing whether requiring stronger evidence of relative outperformance reduces unproductive switches between coins."},{"t":"2026-10-11T00:57:57.583Z","text":"The stronger switching test reduced returns, so I’m retaining the simpler rule. I’m now testing the original information-theoretic idea more directly: whether the drift model’s forecasts beat a zero-drift model on recently observed returns. That measures past predictive performance separately from confidence in the current estimate."},{"t":"2026-10-11T00:59:07.696Z","text":"Requiring positive recent forecast information cut returns sharply; the filter tended to confirm a trend after much of the move had passed. I’m keeping that finding in the writeup. The entry still uses uncertainty about drift direction as its gate, with the limits of that confidence estimate stated explicitly."},{"t":"2026-10-11T01:00:03.213Z","text":"A final simplification reached $1,673.38 locally with the same 27.2% drawdown: rank eligible assets by estimated log drift and leave risk control to the entropy gate and sizing rule. Double costs leave $1,315.78. I’ll submit this version and update the writeup; the extra entropy filters did not improve it."},{"t":"2026-10-11T01:01:36.881Z","text":"The sixth entry is accepted at $1,673.38. Both the causal checks and stress tests are complete. Resampling the visible returns gives a wide range of outcomes, and removing SOL targets leaves about $257. I’m documenting that dependence and the scale of the search before closing the round."},{"t":"2026-10-11T01:03:34.746Z","text":"The accepted file and writeup now match. I’m using the remaining round time to test a separate exit rule: a trailing stop that requires price recovery before re-entry. This will show whether it can preserve gains when price weakens faster than the drift estimate."},{"t":"2026-10-11T01:06:29.416Z","text":"The trailing stop added less than 1%, and the online calibration filter reduced returns. I’m retaining the accepted strategy. One remaining idea is to estimate changing drift with a Kalman filter; I’ll run a small comparison before closing the round."},{"t":"2026-10-11T01:07:37.180Z","text":"The Kalman variants stayed below the accepted entry, so I’m retaining submission #006. The search now totals 1,200 local evaluations. I’ll finish the writeup with those results and keep the submitted code unchanged for the round."},{"t":"2026-10-11T01:09:24.029Z","text":"The current entry remains the best version tested. I’m making one final execution check at hourly resolution, using the same physical lookbacks and sizing rule. I’ll retain submission #006 unless the result gives a clear reason to change it."},{"t":"2026-10-11T01:10:32.741Z","text":"Entry: #006 accepted; strategy.py and WRITEUP.md.  \nVisible years: $100 → **$1,673.38** after costs; **27.2% drawdown**, 254 trades.  \nNext idea: Bayesian change detection to adapt drift windows."},{"t":"2026-10-11T01:10:39.546Z","text":"I’ll verify the accepted entry and writeup, then test a small set of regime-change rules. I’ll leave enough time to submit and check any improvement before the workspace seals."},{"t":"2026-10-11T01:12:24.978Z","text":"The accepted entry and writeup agree: $1,673.38 from $100, with a 27.22% drawdown. I’m testing whether Bayesian detection of drift changes can improve the slow trend estimator. Any replacement will need to justify its added complexity in cost and subperiod checks."},{"t":"2026-10-11T01:13:57.176Z","text":"Bayesian change detection underperformed: its best variant reached about $783 and had a deeper drawdown. I’ll keep the accepted estimator and test one final alternative: stronger confidence thresholds for broad-coin selection, to reduce the chance of buying noisy winners."},{"t":"2026-10-11T01:15:30.843Z","text":"Higher confidence thresholds did not rescue broad-coin selection; all tested variants lost money. Less frequent decisions and larger switching margins also weakened the current strategy. I’m checking its remaining sensitivity and documenting these failures before the final verification."},{"t":"2026-10-11T01:16:50.798Z","text":"The accepted rule stayed profitable with execution delayed by an extra four hours, but wealth fell to $957 and drawdown rose to 37%. Its five overlapping two-year windows ended between $302 and $1,351. I’ll add these checks to the writeup: both timing and market regime matter substantially."},{"t":"2026-10-11T01:18:10.653Z","text":"Final verification passed: the causal prefix checks passed, and the result remains $1,673.38. The local strategy matches accepted submission #006 exactly. The completed writeup includes all equations, 1,258 configuration evaluations, failed ideas, and the cost, timing and regime checks."},{"t":"2026-10-11T01:18:16.482Z","text":"Final entry: accepted submission **#006**, strategy.py, using directional entropy to gate four-hour BTC/ETH/SOL trend rotation with capped half-Kelly sizing. Visible result: **$100 → $1,673.38**, with **27.22% maximum drawdown**. Causal checks passed, and the file matches the accepted entry exactly. WRITEUP.md is complete with equations, 1,258 evaluations, failed approaches, robustness results, and the material selection-bias and execution-timing limitations."}],"submissions":[{"n":"001","t":"2026-10-11T00:32:49.270028+00:00","note":"Initial entry: BTC 200-day trend filter; entropy research follows.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T00:38:22.953008+00:00","note":"Directional KL information weighted trend rotation; BTC/ETH/SOL only, causal 14-day signal and 100-day market gate.","status":"ok","dev":567.7506065246486,"dd":-0.40134174907970555,"reason":""},{"n":"003","t":"2026-10-11T00:43:42.283479+00:00","note":"4h Gaussian drift information gate, Parkinson range variance, capped half-Kelly; BTC/ETH/SOL, parameter-neighbourhood checked.","status":"ok","dev":1046.2095638548703,"dd":-0.29299956143890493,"reason":""},{"n":"004","t":"2026-10-11T00:49:21.730641+00:00","note":"Lag-corrected drift confidence with exact EWMA effective sample size, directional KL information and half-Kelly; neighbourhood and double-cost checks passed.","status":"ok","dev":1319.7375254883773,"dd":-0.3218536396564152,"reason":""},{"n":"005","t":"2026-10-11T00:54:55.750457+00:00","note":"Entropy gate at directional confidence 0.70, lag-corrected drift growth ranking and capped half-Kelly; simpler 56-line implementation, causal and double-cost checks passed.","status":"ok","dev":1556.5285338791193,"dd":-0.27218211202801934,"reason":""},{"n":"006","t":"2026-10-11T01:00:04.722637+00:00","note":"Final simplified entropy-gated log-drift rotation: exact EWMA effective sample size, capped half-Kelly, no extra variance haircut in ranking; causal and robustness checks passed.","status":"ok","dev":1673.3750112694097,"dd":-0.27218211202801923,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: #006 accepted; strategy.py and WRITEUP.md.  \nVisible years: $100 → **$1,673.38** after costs; **27.2% drawdown**, 254 trades.  \nNext idea: Bayesian change detection to adapt drift windows."},{"k":2,"text":"Final entry: accepted submission **#006**, strategy.py, using directional entropy to gate four-hour BTC/ETH/SOL trend rotation with capped half-Kelly sizing. Visible result: **$100 → $1,673.38**, with **27.22% maximum drawdown**. Causal checks passed, and the file matches the accepted entry exactly. WRITEUP.md is complete with equations, 1,258 evaluations, failed approaches, robustness results, and the material selection-bias and execution-timing limitations."}],"title":"Dotterel (a24): writeup","idea":"Trade the strongest positive drift passing an information gate among BTC, ETH and SOL, or hold cash. The entry uses four-hour bars, a lag-corrected combination of two exponential return averages, and range-based volatility. Convert the drift estimate into a Gaussian directional confidence score using the actual effective sample size of the averaging kernel. Low entropy is useful only when its direction is positive: require low Bernoulli directional entropy relative to an uninformative 50/50 direction, then rank eligible assets on estimated logarithmic drift. Hold the asset with the best score, give the incumbent a modest switching advantage, and size it with capped half-Kelly. This is a generic trend-following rule with an information gate, rather than an assertion that estimated probabilities are calibrated. Broad altcoin momentum and reversal performed poorly after costs; the permitted liquid majors supplied more stable trends.","failed":"- Ranking the hottest coins by 14–63-day returns or simple directional entropy in a broad trailing-volume universe. Many selections entered mature pumps and suffered sudden reversals. Some versions nearly exhausted the account. - Replacing returns with volatility-normalized momentum, shortening the momentum window, relaxing listing-age requirements, skipping the top rank, and imposing median-volume, volatility, shock or orderly-trend filters. These mitigated some failures but generally did not beat the liquid-major model. - Daily reversal among trailing-liquid coins. Losers often continued falling, and daily turnover paid too much in costs. Macro filters did not rescue the family. - Causal pooled ridge regression trained on rolling 180/365/730-day windows, with return, relative momentum, volatility, volume, candle and sentiment features. Most predictions were too weak net of costs; high thresholds largely reverted to cash. Training used only labels observable by the current close. - Strong market breadth gates and daily information/Kelly selection across all coins. Requiring broad participation helped some trend variants but missed important moves; the daily Kelly variants performed poorly. - Wealth-based online selection or mixing of multiple trend experts. Chasing which expert had recently won introduced switching and did not improve the stationary rule. - One-hour and fifteen-minute versions with physically matched lookbacks. More frequent confidence crossings raised turnover; the best tested results were below the four-hour strategy. No five-minute strategy was needed after this evidence. - Looser exit confidence, confirmation delays, posterior smoothing, continuous confidence sizing, extra sentiment, signed-volume entropy and short trend guards. These added complex","holdup":"Argue only from data up to 2024-10-09 and from general principles. Don't refer to anything later. The economic hypothesis is that persistent changes in demand create trends longer than short-term noise. The rule estimates their direction rather than forecasting an event or choosing a historical winning coin by name. Drift ranking, a directional information gate, and variance-based sizing rotate among the same three permissible liquid assets. Capped sizing, cash when evidence is weak, and limited refresh frequency address variance drag and trading friction. The lag correction reacts to changing drift while its effective sample size accounts for the added estimation noise. Behavior is entirely causal and identical under historical prefix truncation. The case for persistence is modest: the rule survives higher costs, works in the small previsible sample, and has profitable neighboring parameters. There is no evidence here that it can earn the required 10,000× or that crypto trends will remain exploitable. Heavy tails, correlated crashes, model probability miscalibration, dependence on one asset's strong trends, execution timing sensitivity and approximately 1,200 explored variants all","result":{"dev":{"final":1673.3750112694097,"max_drawdown":-0.27218211202801923,"trades":254,"exposure":0.4060713501947241,"assets_traded":3,"fees_paid":134.89530149893565},"hidden":{"final":174.92810951581703,"max_drawdown":-0.36822944534426894,"trades":214,"exposure":0.43082191780821916,"assets_traded":3,"fees_paid":31.05180161811378},"full":{"final":2927.203272323672,"max_drawdown":-0.3682294453442685,"trades":468,"exposure":0.41596615542903487,"assets_traded":3,"fees_paid":654.5083903254023}},"freq":"4h","equity":[["2021-10-10",100.0444],["2021-10-17",109.1904],["2021-10-24",117.2376],["2021-10-31",113.9012],["2021-11-07",139.1275],["2021-11-14",125.2341],["2021-11-21",124.791],["2021-11-28",124.791],["2021-12-05",116.6913],["2021-12-12",116.6913],["2021-12-19",116.6913],["2021-12-26",116.6913],["2022-01-02",116.6913],["2022-01-09",116.6913],["2022-01-16",116.6913],["2022-01-23",116.6913],["2022-01-30",116.6913],["2022-02-06",117.3357],["2022-02-13",119.061],["2022-02-20",117.7221],["2022-02-27",117.7221],["2022-03-06",114.9465],["2022-03-13",114.1454],["2022-03-20",112.2373],["2022-03-27",116.8499],["2022-04-03",153.1398],["2022-04-10",124.8937],["2022-04-17",124.8937],["2022-04-24",124.8937],["2022-05-01",124.8937],["2022-05-08",124.8937],["2022-05-15",124.8937],["2022-05-22",124.8937],["2022-05-29",124.8937],["2022-06-05",124.8937],["2022-06-12",124.8937],["2022-06-19",124.8937],["2022-06-26",118.0977],["2022-07-03",118.0977],["2022-07-10",118.0977],["2022-07-17",121.0906],["2022-07-24",143.9863],["2022-07-31",147.8399],["2022-08-07",145.2264],["2022-08-14",162.2473],["2022-08-21",152.5685],["2022-08-28",152.5685],["2022-09-04",152.5685],["2022-09-11",151.6774],["2022-09-18",140.3062],["2022-09-25",140.3062],["2022-10-02",140.3062],["2022-10-09",140.3062],["2022-10-16",140.3062],["2022-10-23",140.3062],["2022-10-30",157.9867],["2022-11-06",149.1833],["2022-11-13",133.4728],["2022-11-20",133.4728],["2022-11-27",133.4728],["2022-12-04",133.4728],["2022-12-11",133.4728],["2022-12-18",132.1085],["2022-12-25",132.1085],["2023-01-01",132.1085],["2023-01-08",136.0911],["2023-01-15",234.619],["2023-01-22",250.0861],["2023-01-29",264.3481],["2023-02-05",222.6187],["2023-02-12",217.2681],["2023-02-19",230.8748],["2023-02-26",219.556],["2023-03-05",219.556],["2023-03-12",219.556],["2023-03-19",258.428],["2023-03-26",254.5142],["2023-04-02",255.5742],["2023-04-09",253.3566],["2023-04-16",280.178],["2023-04-23",264.836],["2023-04-30",244.9831],["2023-05-07",241.959],["2023-05-14",241.959],["2023-05-21",241.959],["2023-05-28",241.959],["2023-06-04",240.6996],["2023-06-11",233.0968],["2023-06-18",233.0968],["2023-06-25",253.2272],["2023-07-02",255.8414],["2023-07-09",276.8558],["2023-07-16",355.9203],["2023-07-23",320.5389],["2023-07-30",305.6075],["2023-08-06",305.6075],["2023-08-13",305.6075],["2023-08-20",305.6075],["2023-08-27",305.6075],["2023-09-03",305.6075],["2023-09-10",305.6075],["2023-09-17",305.6075],["2023-09-24",295.2356],["2023-10-01",309.87],["2023-10-08",318.1896],["2023-10-15",302.2027],["2023-10-22",379.5589],["2023-10-29",436.5787],["2023-11-05",538.9762],["2023-11-12",741.2577],["2023-11-19",809.5227],["2023-11-26",717.5743],["2023-12-03",757.4553],["2023-12-10",817.715],["2023-12-17",773.5273],["2023-12-24",1183.9857],["2023-12-31",1066.8826],["2024-01-07",972.1372],["2024-01-14",1028.9761],["2024-01-21",1002.6697],["2024-01-28",1002.6697],["2024-02-04",986.7009],["2024-02-11",1001.8115],["2024-02-18",1064.1639],["2024-02-25",1132.1106],["2024-03-03",1199.93],["2024-03-10",1242.3786],["2024-03-17",1675.5116],["2024-03-24",1506.4668],["2024-03-31",1711.8154],["2024-04-07",1523.1984],["2024-04-14",1508.7827],["2024-04-21",1508.7827],["2024-04-28",1508.7827],["2024-05-05",1508.7827],["2024-05-12",1508.7827],["2024-05-19",1577.3793],["2024-05-26",1722.3272],["2024-06-02",1679.4828],["2024-06-09",1635.6627],["2024-06-16",1635.6627],["2024-06-23",1635.6627],["2024-06-30",1635.6627],["2024-07-07",1595.7259],["2024-07-14",1595.7259],["2024-07-21",1873.9998],["2024-07-28",1945.5989],["2024-08-04",1806.0681],["2024-08-11",1806.0681],["2024-08-18",1806.0681],["2024-08-25",1796.2522],["2024-09-01",1776.1481],["2024-09-08",1776.1481],["2024-09-15",1767.6591],["2024-09-22",1696.9728],["2024-09-29",1818.9513],["2024-10-06",1673.375],["2024-10-13",1673.375],["2024-10-20",1746.6334],["2024-10-27",1859.4019],["2024-11-03",1731.0472],["2024-11-10",1949.3398],["2024-11-17",1951.4569],["2024-11-24",2060.7186],["2024-12-01",2122.3512],["2024-12-08",2283.7787],["2024-12-15",2067.3731],["2024-12-22",1960.8602],["2024-12-29",1960.8602],["2025-01-05",1912.3757],["2025-01-12",1849.7496],["2025-01-19",2252.3451],["2025-01-26",2141.859],["2025-02-02",1987.9072],["2025-02-09",1987.9072],["2025-02-16",1987.9072],["2025-02-23",1987.9072],["2025-03-02",1987.9072],["2025-03-09",1987.9072],["2025-03-16",1987.9072],["2025-03-23",1987.9072],["2025-03-30",1987.9072],["2025-04-06",1987.9072],["2025-04-13",1928.576],["2025-04-20",1795.1613],["2025-04-27",1930.2323],["2025-05-04",1792.439],["2025-05-11",2340.8439],["2025-05-18",2247.6775],["2025-05-25",2358.1854],["2025-06-01",2360.2562],["2025-06-08",2352.532],["2025-06-15",2346.4992],["2025-06-22",2346.4992],["2025-06-29",2346.4992],["2025-07-06",2245.8499],["2025-07-13",2524.4836],["2025-07-20",3166.6817],["2025-07-27",3232.8793],["2025-08-03",2968.2086],["2025-08-10",3270.7102],["2025-08-17",3432.9282],["2025-08-24",3312.0173],["2025-08-31",3031.0796],["2025-09-07",2914.7576],["2025-09-14",3295.4625],["2025-09-21",3234.0403],["2025-09-28",3160.4858],["2025-10-05",3178.3997],["2025-10-12",3082.6831],["2025-10-19",3082.6831],["2025-10-26",3082.6831],["2025-11-02",3082.6831],["2025-11-09",3082.6831],["2025-11-16",3082.6831],["2025-11-23",3082.6831],["2025-11-30",3082.6831],["2025-12-07",3082.6831],["2025-12-14",3024.9053],["2025-12-21",3024.9053],["2025-12-28",3024.9053],["2026-01-04",2960.3544],["2026-01-11",2968.3032],["2026-01-18",2956.4325],["2026-01-25",2834.9685],["2026-02-01",2834.9685],["2026-02-08",2834.9685],["2026-02-15",2834.9685],["2026-02-22",2834.9685],["2026-03-01",2834.9685],["2026-03-08",2721.9382],["2026-03-15",2677.4289],["2026-03-22",2561.566],["2026-03-29",2522.9081],["2026-04-05",2522.9081],["2026-04-12",2442.4467],["2026-04-19",2480.2577],["2026-04-26",2548.9114],["2026-05-03",2491.2124],["2026-05-10",2723.829],["2026-05-17",2492.5229],["2026-05-24",2492.5229],["2026-05-31",2492.5229],["2026-06-07",2492.5229],["2026-06-14",2492.5229],["2026-06-21",2492.5229],["2026-06-28",2492.5229],["2026-07-05",2514.0859],["2026-07-12",2464.8523],["2026-07-19",2490.7106],["2026-07-26",2475.8612],["2026-08-02",2332.4575],["2026-08-09",2324.4089],["2026-08-16",2321.7838],["2026-08-23",2718.8206],["2026-08-30",2826.3481],["2026-09-06",2817.267],["2026-09-13",2673.8133],["2026-09-20",2736.8561],["2026-09-27",3048.5777],["2026-10-04",3000.754],["2026-10-11",2927.2033]],"code_bytes":1947},"a25":{"id":"a25","alias":"Whimbrel","theme":"Spectral and wavelet decomposition of price cycles","started":"2026-10-11T01:16:02+00:00","rounds":[{"k":1,"minutes":39.5},{"k":2,"minutes":16.3}],"messages":[{"t":"2026-10-11T01:16:10.226Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test whether wavelet cycle signals improve returns across separate visible periods. I’ll record the variants tried and their limits in the writeup."},{"t":"2026-10-11T01:16:52.823Z","text":"The visible data contains 459 assets through the stated cutoff. The first strategy selects three liquid coins with positive 7- and 21-day momentum, trades every three days, and moves to cash when market breadth weakens. This gives me a baseline for testing the cycle signals."},{"t":"2026-10-11T01:17:33.512Z","text":"The BTC trend baseline is accepted: $100 became $230.64 with a 25.7% maximum drawdown. The first altcoin momentum test lost heavily. I’m now comparing momentum with buying short dips inside longer uptrends, including their results in each visible year."},{"t":"2026-10-11T01:18:51.974Z","text":"Raw momentum and unconditional dip buying both perform poorly; extreme moves often keep losing or reverse sharply. I’m testing cycle phase signals with liquidity and volatility screens, alongside intraday tests on BTC, ETH and SOL to see whether the pattern differs in the majors."},{"t":"2026-10-11T01:21:20.216Z","text":"The simple cycle and trend tests on the majors have reached only about $255 so far, while a walk-forward model using wavelet and candle features also failed after costs. I’m testing entries around volatility compression and newly listed markets next, with holding rules that allow successful positions to run."},{"t":"2026-10-11T01:23:46.807Z","text":"A Haar wavelet restart signal with persistent holding improved the accepted entry to $733.55: a fresh short upswing must occur within a longer upswing. It traded 69 times, though drawdown reached 63.1%. I’m checking nearby settings and whether the gain depends on a few fortunate positions."},{"t":"2026-10-11T01:27:14.910Z","text":"The gains are concentrated in a few sustained rallies, and nearby wavelet settings often do much worse. That makes the current result vulnerable to selection bias. I’m testing broader market filters and an online ensemble that chooses among cycle settings using only returns already observed."},{"t":"2026-10-11T01:31:21.809Z","text":"The online ensemble underperformed the single wavelet rule. Hourly volume bursts showed modest gains in all three visible years. I’m testing that idea across the daily coin universe: a price impulse with unusually high volume, followed by a capped holding period."},{"t":"2026-10-11T01:33:09.498Z","text":"The latest entry is accepted at $4,337.87, with a 59.4% maximum drawdown and rank 12 of 40. It tightens the stop and exits a declining short phase after a position has advanced 50% from its signal price. I’m checking earlier data, parameter sensitivity and doubled costs before making further changes."},{"t":"2026-10-11T01:36:01.566Z","text":"The current rule also turned $100 into about $1,100 on the earlier 2019–2021 data, but changing its cycle lengths causes large losses on the visible years. I’m testing whether rotating out of a stalled short cycle into a stronger fresh upswing reduces that sensitivity."},{"t":"2026-10-11T01:40:09.820Z","text":"The latest accepted entry is $6,951.42, with a 46.2% drawdown and rank 10 of 40. Earlier profit protection made all three fresh-cash annual tests profitable. I’m testing rolling Fourier estimates as a way to choose cycle length, then will finish the cost checks and writeup."},{"t":"2026-10-11T01:41:00.060Z","text":"The Fourier-based cycle selector has underperformed so far. The final checks are running on the accepted Haar strategy, including doubled costs and eight extra random data cutoffs."},{"t":"2026-10-11T01:43:26.012Z","text":"The accepted rule still returns $5,826.71 with doubled costs, and all eight additional prefix checks passed. Its gains are concentrated: removing the ten best daily portfolio returns leaves $413.34. I’ll report that arithmetic sensitivity, the cycle-length fragility and the roughly 2,000 research evaluations in the writeup."},{"t":"2026-10-11T01:46:30.266Z","text":"The writeup is complete, and `entry.py` matches accepted submission #004 byte for byte. I’m using the remaining time for one focused test: learn which phase restarts succeed from completed earlier outcomes, while preserving the same causal holding rules."},{"t":"2026-10-11T01:49:24.045Z","text":"The learned restart filter failed to improve returns, so the accepted entry remains unchanged. I’m making one final check of whether a small volatility-scaled confirmation threshold helps the short phase signal; its purpose is to test sensitivity to marginal crossings."},{"t":"2026-10-11T01:52:17.230Z","text":"The confirmation thresholds also underperformed. One final experiment is comparing forecasts of maximum upside with forecasts of terminal returns, since the strategy exits after a successful move. The accepted entry and completed writeup remain in place."},{"t":"2026-10-11T01:55:00.653Z","text":"The last forecast tests also underperformed. I’m keeping submission #004 at $6,951.42, currently rank 13 of 40. The writeup includes the failed approaches and sensitivity checks; the next research step is to train on outcomes produced by the actual exit rules."},{"t":"2026-10-11T01:55:31.191Z","text":"Entry: Accepted #004; strategy and writeup.  \nVisible: **$100 → $6,951.42** (69.51×); max drawdown **46.2%**; rank 13/40.  \nNext idea: Learn restart quality from outcomes under the strategy’s actual exit rules."},{"t":"2026-10-11T01:55:38.952Z","text":"I’ll keep the accepted Haar strategy as the fallback while testing whether forecasts of its actual trade outcomes can improve restart selection. I’ll allow time for any submission to score and update the writeup to match the final accepted entry."},{"t":"2026-10-11T01:57:30.938Z","text":"The accepted strategy remains $100 → $6,951 on the visible years, with a 46.2% maximum drawdown. Earlier forecasts struggled to select useful restarts. I’m testing labels based on this strategy’s actual exit rules, which better match the trades it would take."},{"t":"2026-10-11T01:58:36.595Z","text":"The outcome models failed to improve the entry: useful score changes or filters sharply reduced returns. I’m now checking whether volume acceleration and the smoothness of the longer trend help rank candidates. Any replacement will also need credible results across the historical windows."},{"t":"2026-10-11T02:00:17.582Z","text":"The ranking checks also favored the existing rule. I’m testing two final ideas: a causal oscillator for detecting cycle restarts, and a market gate based on breadth or the selected coin’s strength. The writeup already records the entry’s sensitivity to cycle length and profit-protection timing."},{"t":"2026-10-11T02:03:45.268Z","text":"The oscillator, alternative market gates and partial profit-taking rules all underperformed. I’m retaining submission #004. Before finishing the writeup, I’m checking fresh two-year starting windows—the horizon used for final ranking—and the effect of delaying execution by an extra bar."},{"t":"2026-10-11T02:07:54.720Z","text":"The two-year checks had a median 12.31× return; only 2 of 37 overlapping windows reached 40×. An extra day of execution delay reduced visible wealth to $1,388. I’ve documented both limits and confirmed that `entry.py` is byte-identical to accepted submission #004."},{"t":"2026-10-11T02:11:51.531Z","text":"Accepted submission **#004** remains the final entry: causal 4/16-day Haar restarts, a BTC trend gate, one liquid coin at a time, and tighter exits after a 40% peak gain. Visible result: **$100 → $6,951.42 (69.51×)** with **46.2% maximum drawdown**. Causality checks passed, and the file matches the accepted archive. WRITEUP.md is complete with equations, 2,310 evaluations, failed approaches, and evidence of parameter and execution sensitivity."}],"submissions":[{"n":"001","t":"2026-10-11T01:16:57.990177+00:00","note":"Early causal BTC trend baseline; multiscale cross-sectional research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:22:58.311899+00:00","note":"Causal Haar phase restart: fresh 4-day upswing inside a 16-day upswing, persistent holding, BTC trend gate.","status":"ok","dev":733.5533144671215,"dd":-0.6308138198974544,"reason":""},{"n":"003","t":"2026-10-11T01:32:16.267477+00:00","note":"Haar phase restart, persistent winner; tighten protection after a 50% advance and exit a declining short phase.","status":"ok","dev":4337.866897451517,"dd":-0.5936120019209014,"reason":""},{"n":"004","t":"2026-10-11T01:39:00.915172+00:00","note":"Earlier phase protection after a 40% advance; all three visible annual windows profitable from fresh cash.","status":"ok","dev":6951.4164191892205,"dd":-0.46206709763047626,"reason":""}],"entry":"004","round_notes":[{"k":1,"text":"Entry: Accepted #004; strategy and writeup.  \nVisible: **$100 → $6,951.42** (69.51×); max drawdown **46.2%**; rank 13/40.  \nNext idea: Learn restart quality from outcomes under the strategy’s actual exit rules."},{"k":2,"text":"Accepted submission **#004** remains the final entry: causal 4/16-day Haar restarts, a BTC trend gate, one liquid coin at a time, and tighter exits after a 40% peak gain. Visible result: **$100 → $6,951.42 (69.51×)** with **46.2% maximum drawdown**. Causality checks passed, and the file matches the accepted archive. WRITEUP.md is complete with equations, 2,310 evaluations, failed approaches, and evidence of parameter and execution sensitivity."}],"title":"Whimbrel (a25): causal Haar restarts and phase-aware profit protection","idea":"Buy a fresh short upswing inside an existing longer upswing, then let the position run. Two trailing Haar contrasts of log prices identify these nested movements. Candidates come from a changing, liquid universe; their longer contrast is ranked relative to daily volatility. A BTC trend filter controls market exposure. The strategy requests one fully invested spot position at a time and otherwise cash. Once the selected coin's observed price peak exceeds its entry signal price by 40%, protection tightens: exit on either a 10% closing drawdown or a negative short contrast. Before that, permit a 30% closing drawdown and use a slower trend exit. The distinctive choice is to combine phase restart for entry, persistence for holding, and a different exit phase after success. Constantly buying the highest recent return performed very badly in this data. The entry is the accepted submission #004, phase40.py; entry.py is an identical copy. It imports only numpy and pandas, contains no dates or embedded observations, and uses no asset name other than BTC. Research used only the supplied data ending at the stated cutoff. The final source was checked byte-for-byte against the accepted archive, so a resubmission was unnecessary. final_manifest.json records its SHA-256 hash and the four accepted entries.","failed":"- Raw cross-sectional momentum: choosing the strongest recent returns, even with risk normalization and liquidity screens, repeatedly bought exhausted advances. Most variants lost nearly all capital. The first three-coin, three-day rotation trial ended at $9.10. - Unconditional and trend-conditioned short reversal: extreme losers frequently continued falling; fees and permanent declines overwhelmed rebounds. Low-volatility selection alone also failed to beat the basic BTC trend entry. - New listings: generic age and volume screens occasionally captured large gains but had severe losses; the best initial listing-age variant was only about $204.74 with an 86% drawdown. - Harmonic extrapolation: causal rolling ridge/least-squares fits of linear trend plus sine/cosine terms at periods 8–64 did not produce a useful net entry advantage. The best tested group was about $133.03. Endpoint phase estimates were unreliable. - Intraday cycles and burst trading: most dip/phase rules lost. Hourly positive return shocks accompanied by high volume did better, up to about $280.29, but their edge was too small. Applying the burst idea to daily altcoins failed badly. - Walk-forward nonlinear forecasting: a pooled histogram gradient-boosted model trained only on matured next-open labels, using returns, Haar levels/changes, candle features, volume, breadth and sentiment, failed after costs. Its best portfolio variant was about $63.66. It is not used in the entry. - Recent-performance expert selection: selecting/averaging among 36 causal Haar policies using 90-, 180- or 365-day historical equity changes often switched into late winners. Even the best version, about $557.22, lagged the simpler persistent policy. - Rotation, structural stops and normalized protection: stronger fresh candidate r","holdup":"There is an economic hypothesis, rather than a guarantee. Demand can accumulate over weeks; a short setback and recovery can offer a new entry inside that persistent demand. The longer Haar contrast measures that background, while the short crossing avoids continuously chasing the current biggest daily return. Persistent holding reduces costs and permits positively skewed outcomes. Ending a short phase after substantial progress attempts to preserve a successful impulse before its reversal. The BTC gate removes much market-wide downside in the observed sample, and the universe automatically admits new liquid coins and drops inactive ones without recognizing their names. The algorithm does not require the same winning coins to recur. Every calculation uses a bounded trailing price/volume window or a forward-updated reference state. No Fourier phase estimated using later observations, full-sample normalization, future asset survivor list, calendar exception, or future return label enters the accepted weights. Tests in different historical environments and increased costs support the possibility of a repeatable effect. The objections are serious: the search was extensive; only one sma","result":{"dev":{"final":6951.4164191892205,"max_drawdown":-0.46206709763047626,"trades":73,"exposure":0.47992700729927007,"assets_traded":34,"fees_paid":472.85360876770113},"hidden":{"final":31.73899073594815,"max_drawdown":-0.8562645649229627,"trades":45,"exposure":0.48904109589041095,"assets_traded":23,"fees_paid":7.39774635741591},"full":{"final":2198.2256191872225,"max_drawdown":-0.881899474965606,"trades":119,"exposure":0.4835706462212486,"assets_traded":56,"fees_paid":1002.5758499737753}},"freq":"1d","equity":[["2021-10-17",94.782],["2021-10-24",107.5655],["2021-10-31",102.5804],["2021-11-07",129.5129],["2021-11-14",160.4763],["2021-11-21",207.02],["2021-11-28",165.6833],["2021-12-05",148.3466],["2021-12-12",148.3466],["2021-12-19",148.3466],["2021-12-26",148.3466],["2022-01-02",148.3466],["2022-01-09",148.3466],["2022-01-16",148.3466],["2022-01-23",148.3466],["2022-01-30",148.3466],["2022-02-06",148.3466],["2022-02-13",148.3466],["2022-02-20",148.3466],["2022-02-27",148.3466],["2022-03-06",148.3466],["2022-03-13",148.3466],["2022-03-20",148.3466],["2022-03-27",158.2841],["2022-04-03",199.851],["2022-04-10",187.6631],["2022-04-17",187.6631],["2022-04-24",187.6631],["2022-05-01",187.6631],["2022-05-08",187.6631],["2022-05-15",187.6631],["2022-05-22",187.6631],["2022-05-29",187.6631],["2022-06-05",187.6631],["2022-06-12",187.6631],["2022-06-19",187.6631],["2022-06-26",187.6631],["2022-07-03",187.6631],["2022-07-10",187.6631],["2022-07-17",187.6631],["2022-07-24",187.6631],["2022-07-31",187.6631],["2022-08-07",187.6631],["2022-08-14",187.6631],["2022-08-21",187.6631],["2022-08-28",187.6631],["2022-09-04",187.6631],["2022-09-11",187.6631],["2022-09-18",187.6631],["2022-09-25",187.6631],["2022-10-02",187.6631],["2022-10-09",187.6631],["2022-10-16",187.6631],["2022-10-23",187.6631],["2022-10-30",187.6631],["2022-11-06",188.5869],["2022-11-13",283.6771],["2022-11-20",283.6771],["2022-11-27",283.6771],["2022-12-04",283.6771],["2022-12-11",283.6771],["2022-12-18",283.6771],["2022-12-25",283.6771],["2023-01-01",283.6771],["2023-01-08",283.6771],["2023-01-15",283.6771],["2023-01-22",277.7084],["2023-01-29",306.7164],["2023-02-05",303.588],["2023-02-12",346.5686],["2023-02-19",324.4472],["2023-02-26",285.1202],["2023-03-05",253.7816],["2023-03-12",218.0558],["2023-03-19",311.8973],["2023-03-26",277.6017],["2023-04-02",282.1303],["2023-04-09",277.0302],["2023-04-16",300.3987],["2023-04-23",275.8251],["2023-04-30",289.842],["2023-05-07",286.0517],["2023-05-14",265.4971],["2023-05-21",268.7411],["2023-05-28",260.8109],["2023-06-04",280.0233],["2023-06-11",273.9931],["2023-06-18",273.9931],["2023-06-25",265.9836],["2023-07-02",256.1992],["2023-07-09",283.5839],["2023-07-16",271.1087],["2023-07-23",337.4405],["2023-07-30",390.3843],["2023-08-06",379.4304],["2023-08-13",373.6365],["2023-08-20",290.1718],["2023-08-27",290.1718],["2023-09-03",290.1718],["2023-09-10",290.1718],["2023-09-17",290.1718],["2023-09-24",290.1718],["2023-10-01",290.1718],["2023-10-08",260.4327],["2023-10-15",245.1684],["2023-10-22",299.8395],["2023-10-29",323.4626],["2023-11-05",433.7038],["2023-11-12",576.1583],["2023-11-19",565.9917],["2023-11-26",591.6059],["2023-12-03",599.8686],["2023-12-10",936.9848],["2023-12-17",1229.6802],["2023-12-24",1605.5216],["2023-12-31",1501.4319],["2024-01-07",1305.6039],["2024-01-14",1371.546],["2024-01-21",1288.8685],["2024-01-28",1372.6042],["2024-02-04",1415.0244],["2024-02-11",1704.1906],["2024-02-18",1739.8196],["2024-02-25",1656.952],["2024-03-03",1818.4704],["2024-03-10",2297.1722],["2024-03-17",4454.0518],["2024-03-24",3094.3762],["2024-03-31",2925.3324],["2024-04-07",3844.7094],["2024-04-14",4403.4475],["2024-04-21",5878.261],["2024-04-28",5962.0667],["2024-05-05",6389.3696],["2024-05-12",6278.8269],["2024-05-19",7030.5174],["2024-05-26",11032.1641],["2024-06-02",10398.6211],["2024-06-09",8489.5108],["2024-06-16",8132.311],["2024-06-23",6624.3531],["2024-06-30",6624.3531],["2024-07-07",6624.3531],["2024-07-14",6624.3531],["2024-07-21",8340.448],["2024-07-28",7410.5934],["2024-08-04",7476.5849],["2024-08-11",7476.5849],["2024-08-18",7476.5849],["2024-08-25",7607.6019],["2024-09-01",6942.7494],["2024-09-08",6942.7494],["2024-09-15",6942.7494],["2024-09-22",7046.4409],["2024-09-29",7418.9162],["2024-10-06",6983.8618],["2024-10-13",6925.9468],["2024-10-20",6280.8762],["2024-10-27",5358.2209],["2024-11-03",5172.4915],["2024-11-10",5373.5959],["2024-11-17",8940.6914],["2024-11-24",9777.8785],["2024-12-01",8750.4869],["2024-12-08",8177.395],["2024-12-15",8581.9304],["2024-12-22",6491.8558],["2024-12-29",6509.1218],["2025-01-05",7000.3231],["2025-01-12",6858.8473],["2025-01-19",6657.8212],["2025-01-26",7156.246],["2025-02-02",9198.7484],["2025-02-09",8686.8327],["2025-02-16",9128.5478],["2025-02-23",7249.9297],["2025-03-02",6084.1722],["2025-03-09",6084.1722],["2025-03-16",6084.1722],["2025-03-23",6084.1722],["2025-03-30",6084.1722],["2025-04-06",6084.1722],["2025-04-13",6084.1722],["2025-04-20",6084.1722],["2025-04-27",6376.2928],["2025-05-04",6686.7276],["2025-05-11",7796.1035],["2025-05-18",6435.3325],["2025-05-25",7305.6927],["2025-06-01",6334.3936],["2025-06-08",5885.7253],["2025-06-15",6289.7058],["2025-06-22",5366.7462],["2025-06-29",5613.0989],["2025-07-06",4940.9629],["2025-07-13",4459.0709],["2025-07-20",3946.2707],["2025-07-27",4278.6117],["2025-08-03",3980.9225],["2025-08-10",4332.1804],["2025-08-17",4494.4172],["2025-08-24",4525.0279],["2025-08-31",4149.5952],["2025-09-07",4149.5952],["2025-09-14",4036.4182],["2025-09-21",3666.5411],["2025-09-28",2931.9065],["2025-10-05",2784.9786],["2025-10-12",1777.4906],["2025-10-19",1777.4906],["2025-10-26",1777.4906],["2025-11-02",1684.3358],["2025-11-09",1684.3358],["2025-11-16",1684.3358],["2025-11-23",1684.3358],["2025-11-30",1684.3358],["2025-12-07",1684.3358],["2025-12-14",1684.3358],["2025-12-21",1684.3358],["2025-12-28",1684.3358],["2026-01-04",1684.3358],["2026-01-11",1684.3358],["2026-01-18",1614.6788],["2026-01-25",1614.6788],["2026-02-01",1614.6788],["2026-02-08",1614.6788],["2026-02-15",1614.6788],["2026-02-22",1614.6788],["2026-03-01",1614.6788],["2026-03-08",1614.6788],["2026-03-15",1614.6788],["2026-03-22",1614.6788],["2026-03-29",1614.6788],["2026-04-05",1614.6788],["2026-04-12",1614.6788],["2026-04-19",1454.3442],["2026-04-26",1692.8795],["2026-05-03",1798.1761],["2026-05-10",1792.1836],["2026-05-17",1944.9922],["2026-05-24",1862.8854],["2026-05-31",1862.8854],["2026-06-07",1862.8854],["2026-06-14",1862.8854],["2026-06-21",1862.8854],["2026-06-28",1862.8854],["2026-07-05",1862.8854],["2026-07-12",1862.8854],["2026-07-19",1862.8854],["2026-07-26",1862.8854],["2026-08-02",1862.8854],["2026-08-09",1862.8854],["2026-08-16",1862.8854],["2026-08-23",2058.9265],["2026-08-30",2050.931],["2026-09-06",2269.9185],["2026-09-13",2154.6058],["2026-09-20",2256.5631],["2026-09-27",2290.8252],["2026-10-04",2330.1808],["2026-10-11",2198.2256]],"code_bytes":2150},"a26":{"id":"a26","alias":"Redshank","theme":"Genetic programming or evolutionary search for trading rules, with honest validation","started":"2026-10-11T01:16:13+00:00","rounds":[{"k":1,"minutes":39.8},{"k":2,"minutes":12.6}],"messages":[{"t":"2026-10-11T01:16:19.980Z","text":"I’ll inspect the simulator and data, submit an initial strategy, then test an evolutionary search with time-based validation. I’ll keep the rules causal and record the variants and failures in the writeup."},{"t":"2026-10-11T01:17:12.974Z","text":"The data includes 459 assets, including delisted coins. I’ll test rules that rotate into liquid coins with rising prices and volume, using rank retention to reduce trading costs. I’ll use the earlier data and first two visible years for selection, then test the final visible year separately."},{"t":"2026-10-11T01:19:30.433Z","text":"The initial BTC trend entry is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first momentum basket failed badly; the data also shows weak next-day returns after recent price spikes. I’m expanding the search to dip-buying rules within longer trends and combinations of price and volume signals."},{"t":"2026-10-11T01:21:51.060Z","text":"The first search evaluated 694 rule variants. Its best training rules lost much of their advantage in the reserved year, so I’m keeping the BTC entry for now. I’ll test whether similar rules work more reliably on BTC, ETH and SOL at shorter bar lengths, where trading costs are lower."},{"t":"2026-10-11T01:24:31.064Z","text":"The best four-hour rule turned $100 into $1,250 in the exact simulator and passed the causal checks, with a 32.5% maximum drawdown. It combines long-term trend strength, relative momentum and volatility, and stays in cash when prices weaken. Its reserved-year test returned about 3.9×; I’ll submit it while testing alternatives."},{"t":"2026-10-11T01:27:11.900Z","text":"The four-hour entry is accepted. The hourly search produced weaker candidates, so I’m continuing with the broader daily universe and nonlinear signal combinations. I’m also checking how much the submitted rule depends on costs and a few unusually profitable days."},{"t":"2026-10-11T01:29:36.269Z","text":"The submitted rule remains profitable with doubled costs: $100 becomes about $724. Removing its ten best days leaves about $246, so a few rallies account for much of its growth. The daily search has found stronger training results with a liquid, long-term leader held through small rank changes; its reserved-year test is still pending."},{"t":"2026-10-11T01:32:15.917Z","text":"I found a detail worth fixing: a fully invested, single-coin position can repeat its target without rebalancing costs. Doing so lets a fresh account enter the current signal immediately; with hold-only rows, it may wait for the next switch. I’ll verify the effect in the exact simulator before changing the entry."},{"t":"2026-10-11T01:34:40.477Z","text":"The daily search failed its reserved-year test: large training gains were followed by severe losses and drawdowns above 80% in the final year. I’ll keep the four-hour entry. Next I’ll test a causal combination of rules that shifts toward those earning money recently and moves to cash when they’re all losing."},{"t":"2026-10-11T01:38:18.407Z","text":"The complete-target version passed the causal checks and improved the exact result to $1,299.95, with the same 32.5% maximum drawdown. I’ll submit that fix. The next checks vary each parameter by ±20% and test the same time windows on finer bars to assess sensitivity."},{"t":"2026-10-11T01:42:50.795Z","text":"All 20 parameter-neighbour tests remained profitable, ending between about $808 and $1,430. Finer bars produced weaker results, consistent with more frequent exits and higher turnover. I’ll test one final variation: keep the four-hour decisions, but use hourly prices for an earlier exit during sharp drops."},{"t":"2026-10-11T01:44:19.675Z","text":"A 10% hourly exit ranked best in the hybrid’s training tests. Its simulated visible result is $1,398.80, while its reserved-year return remains close to the original rule’s 3.9×. It adds only five round trips over three years; the exact referee check is running now."},{"t":"2026-10-11T01:49:28.883Z","text":"The hybrid is accepted at $1,398.80 and a 33.0% maximum drawdown. Doubled costs leave about $801; removing its ten best days leaves about $296. I’m finishing the writeup with the equations, search counts and failed validation results. Those failures remain the strongest reason to treat the visible gains cautiously."},{"t":"2026-10-11T01:51:39.683Z","text":"The final four-hour expert test produced a stronger candidate: $100 became about $1,693, with a 30.3% maximum drawdown and fewer trades. It had a slightly higher training score and a 4.6× reserved-year return. I’ll verify the self-contained implementation before replacing the accepted hybrid."},{"t":"2026-10-11T01:53:57.106Z","text":"The expert selector is accepted as submission 005: $100 became $1,693.26, with a 30.3% maximum drawdown. Doubled costs leave about $1,158. It will be this round’s entry; I’m updating the writeup with the virtual-account equations and the final results."},{"t":"2026-10-11T01:56:00.186Z","text":"Entry: entry_meta.py, accepted #005; writeup completed.  \nVisible years: **$100 → $1,693.26**, 30.3% max drawdown; rank **28/40**.  \nNext idea: reduce redundant experts and test selector stability across regimes."},{"t":"2026-10-11T01:56:07.908Z","text":"I’ll verify the accepted entry, then test whether fewer experts and a volatility-based switching margin improve its stability. I’ll keep the current rule unless the training and validation results support a change, and update the writeup for the final submission."},{"t":"2026-10-11T01:58:33.592Z","text":"The accepted entry is still `entry_meta.py`: $1,693.26 from $100, with 30.3% maximum drawdown. I’m testing a limited set of smaller expert pools and switching margins using the original training windows. The final visible year will be a validation check."},{"t":"2026-10-11T02:00:04.471Z","text":"Smaller pools improved the training score but weakened final-year validation. The best new candidate returned $348 in that year versus $461 for the accepted entry, with more trades. I’m keeping the sixteen-expert selector and checking one last transfer of its evolved signal to the daily, data-selected coin universe."},{"t":"2026-10-11T02:03:18.078Z","text":"The daily transfer failed validation: its training winner lost money in the final visible year, and the daily expert selector did worse. I’m rejecting that family. The final test is a small fixed-seed evolutionary search over the selector’s score formula; the expert trading rules remain frozen."},{"t":"2026-10-11T02:04:08.886Z","text":"The final evolutionary search kept the existing score as its training winner. All eight nearby selector parameter changes remained profitable, with visible wealth between $1,144 and $1,974. I’ve also verified that `entry_meta.py` is byte-for-byte identical to accepted submission #005, so no resubmission is needed. I’m adding these checks and the rejected ideas to the writeup."},{"t":"2026-10-11T02:08:38.024Z","text":"Accepted submission **#005**, entry_meta.py, remains the final entry. It selects among sixteen BTC/ETH/SOL trend experts using causal, fee-adjusted virtual performance, holding one coin or cash. Visible results were **$100 → $1,693.26**, with **30.3% maximum drawdown**. Final-round experiments supported retaining it. WRITEUP.md is complete with equations, 8,686 evaluated configurations, validation results, failures and limitations. Hidden-year performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T01:18:56.702866+00:00","note":"Initial causal Bitcoin trend entry; evolutionary research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:24:32.832737+00:00","note":"Evolved four-hour strength/volatility interaction; selected on earlier periods, separate final-year validation.","status":"ok","dev":1250.0035634656278,"dd":-0.32470485747005273,"reason":""},{"n":"003","t":"2026-10-11T01:38:20.426032+00:00","note":"Complete single-asset targets establish fresh accounts immediately; unchanged holdings incur no rebalancing turnover.","status":"ok","dev":1299.9503858611438,"dd":-0.32470485747005307,"reason":""},{"n":"004","t":"2026-10-11T01:44:36.328799+00:00","note":"Evolved four-hour allocation on complete hourly groups, with a 10% hourly drawdown exit; training-selected and causality checked.","status":"ok","dev":1398.8001486733212,"dd":-0.3301944834519951,"reason":""},{"n":"005","t":"2026-10-11T01:53:02.920042+00:00","note":"Sixteen GP-derived major-coin experts; causal fee-adjusted virtual growth selects the active rule, with cash and switching hysteresis.","status":"ok","dev":1693.262902324796,"dd":-0.3033075763635019,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: entry_meta.py, accepted #005; writeup completed.  \nVisible years: **$100 → $1,693.26**, 30.3% max drawdown; rank **28/40**.  \nNext idea: reduce redundant experts and test selector stability across regimes."},{"k":2,"text":"Accepted submission **#005**, entry_meta.py, remains the final entry. It selects among sixteen BTC/ETH/SOL trend experts using causal, fee-adjusted virtual performance, holding one coin or cash. Visible results were **$100 → $1,693.26**, with **30.3% maximum drawdown**. Final-round experiments supported retaining it. WRITEUP.md is complete with equations, 8,686 evaluated configurations, validation results, failures and limitations. Hidden-year performance remains unknown."}],"title":"Redshank (a26): causal selection among evolved trend experts","idea":"The accepted entry is entry_meta.py, submission 005: a four-hour, long-only selector among sixteen deterministic BTC/ETH/SOL trading experts. Twelve experts vary a small genetic-programming expression combining absolute trend, relative strength and volatility. Four use a simpler volatility score in rising trends. Every expert runs a causal virtual account with actual next-open timing and the specified fees. A second layer selects the expert with the strongest recent virtual log growth, penalizes volatility, uses a slower growth term, and requires a meaningful advantage before switching. It holds that expert's selected coin or cash. This combines evolved rules with online selection from outcomes that are already known. Broad daily searches generated spectacular training profits followed by failed validation, and were rejected. Runtime computes the frozen rules and their causal virtual histories; it performs no evolutionary search.","failed":"- A daily top-three momentum/volume basket with loose filters turned $100 into approximately $0.0016 in the exact engine. Extreme recent winners often reversed and turnover added losses. - Linear daily evolution favoured low volatility, but strong training candidates often lost in the reserved year. - Nonlinear daily evolution found earlier training multiples near 60× and 83×, then approximately 3.2× and 4.6× in later training windows. Its training winner returned only about $43 in the reserved year, with a drawdown above 80%. I rejected it. - Causal pooled online ridge regression used purged 3-, 7- or 14-day forward labels and decayed sufficient statistics. Its leading training configuration returned about $61 in the reserved year. - Selecting among twenty-five daily experts by their past fee-adjusted virtual profits did not repair a weak pool. Its training winner was roughly flat in the reserved year with large drawdowns. - Requiring altcoins to outperform Bitcoin over trailing returns did not rescue the daily families; several nearly lost all capital in validation. - Hourly GP rules were weaker. Some produced redundant zero-valued expressions that depended on tie ordering; I rejected such rules. - Tighter hourly exits, exit cooldowns and finer-frequency feature recomputation generally reduced growth. - Pruning correlated four-hour experts improved training but weakened validation; volatility-dependent switching margins supplied no consistent improvement. - Transferring the evolved signal to the daily broad-coin universe, with liquidity and trend filters, still produced failed validation. Its causal meta-selector did not rescue it. - The final evolutionary selector search found no training improvement over the already accepted expression.","holdup":"The hypothesis is persistence of trends and activity, rather than successful historical altcoin identities or calendar events. Emerging relative winners can qualify before their 128-day absolute trend recovers. Weak shorter trends and retreats from recent peaks remove exposure. Slower decisions and selector hysteresis amortize costs. Selecting from past net virtual performance can reduce reliance on a single parameter choice, while the risk penalty and cash alternative reduce exposure after poor recent outcomes. The permitted major-coin universe avoids learned preferences for obscure tickers and receives the specified cheaper costs. Profitable earlier visible windows, positive reserved-year performance, profitable neighbours and doubled-cost performance support trying the rule. Its small reduced expression is easier to audit than the unstable daily combinations. Uncertainty remains substantial: only three visible years, thousands of trials, an unusually profitable training quarter and a selector whose experts are highly correlated. Meta-selection can lag regime changes or chase an expert after its profitable episode ends. False trends, correlated declines and next-open gaps can cau","result":{"dev":{"final":1693.262902324796,"max_drawdown":-0.3033075763635019,"trades":380,"exposure":0.35355839416058393,"assets_traded":3,"fees_paid":239.55166537206685},"hidden":{"final":199.5895341730257,"max_drawdown":-0.3360141992818253,"trades":294,"exposure":0.34200913242009134,"assets_traded":3,"fees_paid":58.168432936483896},"full":{"final":3379.575539074713,"max_drawdown":-0.33601419928182474,"trades":674,"exposure":0.34894121942314715,"assets_traded":3,"fees_paid":1224.4961611492256}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",109.4318],["2021-10-24",129.8987],["2021-10-31",138.9533],["2021-11-07",167.8735],["2021-11-14",169.2093],["2021-11-21",169.2093],["2021-11-28",169.2093],["2021-12-05",164.3755],["2021-12-12",159.3593],["2021-12-19",159.3593],["2021-12-26",159.3593],["2022-01-02",159.3593],["2022-01-09",159.3593],["2022-01-16",159.3593],["2022-01-23",159.3593],["2022-01-30",159.3593],["2022-02-06",159.3593],["2022-02-13",153.7028],["2022-02-20",150.7573],["2022-02-27",150.7573],["2022-03-06",152.4499],["2022-03-13",147.1529],["2022-03-20",146.3432],["2022-03-27",147.1829],["2022-04-03",192.8933],["2022-04-10",171.7542],["2022-04-17",171.7542],["2022-04-24",171.7542],["2022-05-01",171.7542],["2022-05-08",171.7542],["2022-05-15",171.7542],["2022-05-22",171.7542],["2022-05-29",171.7542],["2022-06-05",171.7542],["2022-06-12",171.7542],["2022-06-19",171.7542],["2022-06-26",171.7542],["2022-07-03",171.7542],["2022-07-10",171.7542],["2022-07-17",171.7542],["2022-07-24",195.5249],["2022-07-31",182.617],["2022-08-07",178.8969],["2022-08-14",192.9058],["2022-08-21",181.3981],["2022-08-28",181.3981],["2022-09-04",181.3981],["2022-09-11",181.3981],["2022-09-18",181.3981],["2022-09-25",181.3981],["2022-10-02",181.3981],["2022-10-09",179.7257],["2022-10-16",179.7257],["2022-10-23",174.8988],["2022-10-30",204.9257],["2022-11-06",188.0119],["2022-11-13",180.9808],["2022-11-20",180.9808],["2022-11-27",180.9808],["2022-12-04",180.9808],["2022-12-11",180.9808],["2022-12-18",168.395],["2022-12-25",168.395],["2023-01-01",168.395],["2023-01-08",164.2271],["2023-01-15",272.8221],["2023-01-22",280.3678],["2023-01-29",315.324],["2023-02-05",278.1659],["2023-02-12",277.1578],["2023-02-19",282.5353],["2023-02-26",267.3952],["2023-03-05",247.4648],["2023-03-12",247.4648],["2023-03-19",294.8469],["2023-03-26",287.9124],["2023-04-02",283.8038],["2023-04-09",284.5508],["2023-04-16",338.0752],["2023-04-23",315.5009],["2023-04-30",291.3325],["2023-05-07",291.3325],["2023-05-14",291.3325],["2023-05-21",291.3325],["2023-05-28",281.6731],["2023-06-04",289.7872],["2023-06-11",257.1639],["2023-06-18",257.1639],["2023-06-25",277.1328],["2023-07-02",279.3255],["2023-07-09",299.7752],["2023-07-16",373.73],["2023-07-23",343.0672],["2023-07-30",343.0672],["2023-08-06",343.0672],["2023-08-13",342.1463],["2023-08-20",340.7492],["2023-08-27",340.7492],["2023-09-03",339.0154],["2023-09-10",339.0154],["2023-09-17",337.9132],["2023-09-24",339.4658],["2023-10-01",359.3731],["2023-10-08",369.0217],["2023-10-15",367.3886],["2023-10-22",457.3688],["2023-10-29",526.0778],["2023-11-05",624.9756],["2023-11-12",828.4369],["2023-11-19",784.7174],["2023-11-26",784.7174],["2023-12-03",820.1404],["2023-12-10",948.219],["2023-12-17",881.378],["2023-12-24",1365.4471],["2023-12-31",1286.1292],["2024-01-07",1181.5166],["2024-01-14",1188.3383],["2024-01-21",1127.5236],["2024-01-28",1127.5236],["2024-02-04",1102.5624],["2024-02-11",1085.7622],["2024-02-18",1100.0048],["2024-02-25",1154.0407],["2024-03-03",1382.9324],["2024-03-10",1280.1459],["2024-03-17",1726.4458],["2024-03-24",1565.8008],["2024-03-31",1529.4796],["2024-04-07",1428.9928],["2024-04-14",1278.8911],["2024-04-21",1278.8911],["2024-04-28",1278.8911],["2024-05-05",1278.8911],["2024-05-12",1278.8911],["2024-05-19",1364.4963],["2024-05-26",1698.8626],["2024-06-02",1636.4558],["2024-06-09",1624.7796],["2024-06-16",1594.1365],["2024-06-23",1594.1365],["2024-06-30",1594.1365],["2024-07-07",1594.1365],["2024-07-14",1594.1365],["2024-07-21",1778.509],["2024-07-28",1846.4597],["2024-08-04",1679.4677],["2024-08-11",1679.4677],["2024-08-18",1679.4677],["2024-08-25",1692.5185],["2024-09-01",1644.1895],["2024-09-08",1644.1895],["2024-09-15",1637.0315],["2024-09-22",1647.1627],["2024-09-29",1858.1363],["2024-10-06",1693.2629],["2024-10-13",1693.2629],["2024-10-20",1816.554],["2024-10-27",1963.3321],["2024-11-03",1913.5207],["2024-11-10",2207.9387],["2024-11-17",2588.4665],["2024-11-24",2777.1506],["2024-12-01",2782.4484],["2024-12-08",2733.0702],["2024-12-15",2705.6402],["2024-12-22",2358.4542],["2024-12-29",2358.4542],["2025-01-05",2321.9581],["2025-01-12",2250.4323],["2025-01-19",2942.0025],["2025-01-26",2812.3301],["2025-02-02",2710.9672],["2025-02-09",2710.9672],["2025-02-16",2710.9672],["2025-02-23",2697.8315],["2025-03-02",2697.8315],["2025-03-09",2697.8315],["2025-03-16",2697.8315],["2025-03-23",2697.8315],["2025-03-30",2688.6616],["2025-04-06",2669.9636],["2025-04-13",2669.9636],["2025-04-20",2669.9636],["2025-04-27",2799.541],["2025-05-04",2748.6547],["2025-05-11",3653.1221],["2025-05-18",3742.0312],["2025-05-25",3740.0765],["2025-06-01",3541.1613],["2025-06-08",3512.6542],["2025-06-15",3333.4928],["2025-06-22",3215.5227],["2025-06-29",3204.6852],["2025-07-06",3040.8302],["2025-07-13",3462.9274],["2025-07-20",4286.7987],["2025-07-27",4140.9089],["2025-08-03",3694.3899],["2025-08-10",4010.2175],["2025-08-17",4113.4829],["2025-08-24",3824.3362],["2025-08-31",3827.8654],["2025-09-07",3973.4764],["2025-09-14",4636.1256],["2025-09-21",4547.0757],["2025-09-28",4547.0757],["2025-10-05",4604.6297],["2025-10-12",4492.2759],["2025-10-19",4492.2759],["2025-10-26",4492.2759],["2025-11-02",4433.7899],["2025-11-09",4433.7899],["2025-11-16",4433.7899],["2025-11-23",4433.7899],["2025-11-30",4433.7899],["2025-12-07",4433.7899],["2025-12-14",4205.5869],["2025-12-21",4205.5869],["2025-12-28",4205.5869],["2026-01-04",4207.3708],["2026-01-11",4278.692],["2026-01-18",4353.3638],["2026-01-25",3972.7976],["2026-02-01",3972.7976],["2026-02-08",3972.7976],["2026-02-15",3972.7976],["2026-02-22",3972.7976],["2026-03-01",3972.7976],["2026-03-08",3830.957],["2026-03-15",3909.0945],["2026-03-22",3761.9344],["2026-03-29",3761.9344],["2026-04-05",3761.9344],["2026-04-12",3688.6499],["2026-04-19",3591.3465],["2026-04-26",3567.3175],["2026-05-03",3440.372],["2026-05-10",3732.9468],["2026-05-17",3254.5212],["2026-05-24",3254.5212],["2026-05-31",3254.5212],["2026-06-07",3254.5212],["2026-06-14",3254.5212],["2026-06-21",3254.5212],["2026-06-28",3254.5212],["2026-07-05",3376.3548],["2026-07-12",3363.8615],["2026-07-19",3259.5567],["2026-07-26",3252.5511],["2026-08-02",3212.8029],["2026-08-09",3212.8029],["2026-08-16",3212.8029],["2026-08-23",3779.2746],["2026-08-30",3864.2411],["2026-09-06",3604.0828],["2026-09-13",3441.2757],["2026-09-20",3346.5544],["2026-09-27",3505.7552],["2026-10-04",3401.7792],["2026-10-11",3379.5755]],"code_bytes":5188},"a27":{"id":"a27","alias":"Turnstone","theme":"Reinforcement learning (tabular or linear policies) for allocation, trained walk-forward","started":"2026-10-11T01:18:18+00:00","rounds":[{"k":1,"minutes":40.0},{"k":2,"minutes":13.1}],"messages":[{"t":"2026-10-11T01:18:23.793Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test walk-forward allocation ideas. I’ll track the variants and document the final rules and maths in `WRITEUP.md`."},{"t":"2026-10-11T01:19:01.294Z","text":"The simulator fills at the next open and charges for every rebalance, so turnover matters. I’ll start with liquid-coin momentum and a market trend filter, then test whether walk-forward learning can improve the allocation."},{"t":"2026-10-11T01:20:51.845Z","text":"The first entry is accepted: $100 became $187.45, with a 36.1% maximum drawdown. The initial cross-sectional momentum test lost money, so I’m testing shorter horizons, reversal, breakout and liquidity rules before adding learning."},{"t":"2026-10-11T01:22:14.260Z","text":"The pooled data shows that extreme daily gainers tend to reverse, while moderate drops tend to bounce. Buying the largest losers still performs badly after costs and volatility. I’m testing dip trades within established trends and a walk-forward action-value model to distinguish rebounds from continuing declines."},{"t":"2026-10-11T01:25:23.140Z","text":"The broad-universe strategies have disappointed so far. Rotating among BTC, ETH and SOL on 21-day momentum, with a 30-day BTC trend filter, reached about $934 after costs and gained in all three visible sub-periods. I’ll submit it while testing whether learning can improve on it."},{"t":"2026-10-11T01:28:45.970Z","text":"The major-rotation entry is accepted at $934.25. The first tabular policies stayed mostly in cash during weak trends but underperformed it; the pooled return models also traded too much. I’m testing whether limiting selection to the most liquid coins preserves useful momentum, alongside a four-hour version."},{"t":"2026-10-11T01:31:53.959Z","text":"The trade review found several large losses from buying coins after explosive rallies and holding through sharp reversals. I’m testing limits on how extended a rally can be, plus a pooled table that learns net log returns for return-and-volatility states using only outcomes already realized."},{"t":"2026-10-11T01:35:42.231Z","text":"A breadth filter produced a stronger candidate: enter leaders when more than 60% of the liquid universe is above its 30-day trend, and avoid fresh entries after a daily gain above 10%. The exact simulator gives $8,451.90, with gains in all three sub-periods. Its 70.7% drawdown is substantial; I’m checking causality and nearby parameters before relying on it."},{"t":"2026-10-11T01:38:49.240Z","text":"The breadth strategy is accepted at $8,451.90, currently eighth. Nearby breadth, liquidity and spike thresholds also perform well, but the result is sensitive to momentum length and rebalance cadence. I’m testing volatility sizing and ensembles of nearby rules to reduce dependence on one setting."},{"t":"2026-10-11T01:42:10.371Z","text":"Changing the rebalance phase sharply weakens the result, so the $8,451 backtest looks partly dependent on timing luck. An online ensemble reached about $2,213 and gained in each sub-period. Volatility sizing on the single-rule version reached about $10,138 and reduced drawdown to 51.7%. I’m comparing those two approaches next."},{"t":"2026-10-11T01:46:14.986Z","text":"The volatility-sized entry is accepted at $10,137.54, now seventh. Most of its growth comes from a small number of large winning days: removing the best ten leaves about $277. I’m testing a contextual learner that chooses among rebalance phases using realized outcomes, and documenting this sensitivity in the writeup."},{"t":"2026-10-11T01:49:41.858Z","text":"The latest entry is accepted: $100 became $18,553.86, currently third, with a 61.7% drawdown. It also gained in each annual window and returned about $10,907 with costs doubled. The writeup will make the timing sensitivity and reliance on rare winning days explicit."},{"t":"2026-10-11T01:52:39.911Z","text":"`WRITEUP.md` now contains the equations, parameters, search count, results and failed approaches. I’m using the remaining round time to test whether a trailing exit improves the weaker rebalance phases, which would give better evidence that the rule can work beyond this sample."},{"t":"2026-10-11T01:55:24.198Z","text":"The trailing-exit tests did not fix the rebalance-phase weakness, so I’m keeping submission 005. I’m checking the earlier available history, before the visible scoring window, to see how the same rule behaved in a different market cycle."},{"t":"2026-10-11T01:56:57.100Z","text":"The earlier-cycle check is poor: the selected rule turned $100 into $32.21 from September 2019 through October 2021. That strengthens the overfitting concern, and I’ve recorded it in the writeup. The next idea is an online allocation between this strategy and the liquid-major baseline, rather than relying on the winning phase alone."},{"t":"2026-10-11T01:58:24.641Z","text":"The current entry has a strong visible result, but it lost money in the earlier available cycle. I’ll use this final round to test an online allocation between the broad-universe rule and the more consistent BTC/ETH/SOL baseline, then submit the chosen strategy with enough time for scoring and finish the writeup."},{"t":"2026-10-11T02:00:26.588Z","text":"The weak point is timing: shifting the current rule’s three-day rebalance schedule changes its result dramatically. I’m testing all three schedules together, with an online learner that can move capital toward the BTC/ETH/SOL strategy when broad-coin trading stops working."},{"t":"2026-10-11T02:01:54.344Z","text":"The learner has a much stronger result across the available history: about $1,631 in the scored years and $5,591 in the earlier cycle, versus $18,554 and $32 for the standalone rule. I’m checking the major-coin fallback’s rebalance timing before choosing the final entry; I’m willing to give up visible rank for better evidence of generalization."},{"t":"2026-10-11T02:03:59.981Z","text":"I’m choosing the learner with all three broad-coin schedules, a fallback averaged across all three major-coin schedules, and the lower volatility target. It returned about $1,287 in the scored years and $10,230 in the earlier cycle, with smaller drawdowns than the standalone entry. I’ll submit it now and finish documenting the tradeoff and failed tests."},{"t":"2026-10-11T02:06:51.005Z","text":"Submission #006 is accepted as the final entry: $100 became $1,286.93, with 50.9% maximum drawdown. Its visible rank is now #35. That is a deliberate tradeoff: the new policy earned across both available cycles and avoids committing to the single schedule that produced the much larger fitted result."},{"t":"2026-10-11T02:09:24.732Z","text":"The writeup now covers the full algorithm, all 951 logged configuration evaluations, the retired rule’s failures, and the final validation results. The accepted file matches the working copy exactly. Doubling execution costs left $923 from the initial $100; tripling them left $662. Those checks support cost tolerance, though they do not establish future profitability."},{"t":"2026-10-11T02:11:21.648Z","text":"Submission **#006** is accepted: turnstone.py, an online learner combining three broad-coin momentum schedules with a BTC/ETH/SOL trend fallback. Visible result: **$100 → $1,286.93**, **50.9% maximum drawdown**, rank **35/40** at acceptance. The earlier available cycle returned **$10,229.77**. I chose it over the higher-ranked entry because that rule failed the earlier cycle and depended heavily on rebalance timing. Causal checks passed. WRITEUP.md is complete with equations, 951 logged evaluations, results, failures, and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T01:19:43.805258+00:00","note":"Initial entry: causal BTC trend baseline; research proceeds on walk-forward policies.","status":"ok","dev":187.4511072173092,"dd":-0.3612426554665471,"reason":""},{"n":"002","t":"2026-10-11T01:25:31.220184+00:00","note":"Major rotation: positive 21-day momentum, BTC EMA30 market gate, three-day allocation cadence.","status":"ok","dev":934.2464972245282,"dd":-0.4174503989449275,"reason":""},{"n":"003","t":"2026-10-11T01:36:40.375911+00:00","note":"Broad participation leaders: trend and liquidity eligibility, breadth hysteresis, and avoid chasing daily spikes.","status":"ok","dev":8451.902224833477,"dd":-0.7073773951158551,"reason":""},{"n":"004","t":"2026-10-11T01:42:49.327519+00:00","note":"Breadth leaders with anti-chasing and rounded volatility sizing; causal cuts passed; reduced drawdown.","status":"ok","dev":10137.54410714445,"dd":-0.5173809912170546,"reason":""},{"n":"005","t":"2026-10-11T01:48:44.418749+00:00","note":"Broad-participation momentum, no chase after 10% daily spike, 15% volatility target capped at spot-only exposure.","status":"ok","dev":18553.857978152977,"dd":-0.6168210385718544,"reason":""},{"n":"006","t":"2026-10-11T02:04:05.213741+00:00","note":"Final candidate: fee-aware rolling expert learner, all three breadth phases and phase-averaged BTC/ETH/SOL fallback; lower volatility target; prioritizes both observed cycles.","status":"ok","dev":1286.9267848923214,"dd":-0.5089353956169235,"reason":""}],"entry":"006","round_notes":[{"k":2,"text":"Submission **#006** is accepted: turnstone.py, an online learner combining three broad-coin momentum schedules with a BTC/ETH/SOL trend fallback. Visible result: **$100 → $1,286.93**, **50.9% maximum drawdown**, rank **35/40** at acceptance. The earlier available cycle returned **$10,229.77**. I chose it over the higher-ranked entry because that rule failed the earlier cycle and depended heavily on rebalance timing. Causal checks passed. WRITEUP.md is complete with equations, 951 logged evaluations, results, failures, and generalization limits."}],"title":"Turnstone (a27): writeup","idea":"The final accepted entry is turnstone.py, submission 006: a daily, long-only, full-information online allocation policy. It learns from four fee-paying shadow portfolios: three schedules of a broad-market, liquid-leader strategy and one BTC/ETH/SOL momentum strategy averaged over those same schedules. The broad strategy requires Bitcoin's positive trend, majority market participation, positive own trend, seven-day momentum and a last daily return below 10%; it limits exposure to very volatile leaders. A rolling 90-day exponential-weights learner allocates among the four strategies, using only returns already realized at the decision bar's close. This returns $1,286.93 from $100 in the scored years and $10,229.77 in the earlier available cycle. I deliberately replaced the $18,553.86 standalone entry: it lost money in the earlier cycle and depended severely on one arbitrary rebalance phase. The final policy preserves the original walk-forward learning direction, with a simpler full-information learner rather than the unsuccessful tabular Q-learning prototypes. Submission acceptance is recorded below.","failed":"- Broad-universe textbook momentum. Top recent winners often underwent sharp reversals. The first 21/63-day momentum prototype returned about $65.20. Diversification, longer holding intervals and trend filters helped some variants, but the initial coarse family peaked around $123.95. - Daily loser buying and trend-qualified dips. Large declines sometimes bounced in the pooled average, but extreme losers had adverse tail risk and daily rotation paid too much in costs. Absolute and volatility-normalized loser ranking, candle location and dip-in-uptrend variants usually lost most capital. - Robust mean, Kelly and win-rate rankings. Winsorized drift, drift/variance and the fraction of positive days did not fix reversal risk. The best coarse variant returned about $184.23. - Listings, volume shocks and isolated breakouts. Short listing histories, volume acceleration and breakout-only entry were insufficient without broad participation. Concentrated leaders suffered severe single-day losses. There are no named exclusion lists in the final solution. - High beta and residual quality. Rolling covariance/correlation with BTC, high-beta selection and residual Sharpe ranking performed poorly. Participation filtering, rather than estimated beta, was the useful discovery in that experiment. - Walk-forward pooled linear and tree action-value prediction. Predictors trained on 365 trailing days, purging 1/3/7-day labels, with 28-day refits, usually chased unstable high estimates. Best result was about $115.21. The tree implementation used scikit-learn histogram gradient boosting; installed LightGBM could not load its OpenMP dependency offline, and I did not install packages or access the network. - Tabular Q-learning over BTC/ETH/SOL/cash. All actions' realized price outcomes are observ","holdup":"The proposed broad-coin mechanism is shared capital inflow: participation offers evidence that momentum extends across liquid assets. Liquidity and minimum history reduce exposure to barely traded, newly listed coins. Positive own trend, positive Bitcoin trend and breadth hysteresis reduce exposure during falling markets. The entry ceiling responds to the observed short-horizon reversal after large gains. Volatility sizing limits the allocation to erratic leaders. The major expert accesses a different, cheaper source of trend returns and succeeded when the broad rule failed in the earlier available cycle. These mechanisms require no knowledge of future winners, sectors, listings or events. The learner is the generalization response: include all schedules, observe all expert outcomes net of fees, and update relative allocation using only the most recent quarter. It can reduce a losing specialist's allocation without editing the coin universe or naming a future regime. The original paper-account history is replayed on every call, so this training continues unchanged when new bars arrive. Lower broad-expert exposure and a trade threshold address risk and turnover. The final policy gai","result":{"dev":{"final":1286.9267848923214,"max_drawdown":-0.5089353956169235,"trades":953,"exposure":0.2903538200785068,"assets_traded":124,"fees_paid":136.61629042774754},"hidden":{"final":67.14259334603851,"max_drawdown":-0.6087504399316751,"trades":711,"exposure":0.33597478739843195,"assets_traded":94,"fees_paid":21.639343135229872},"full":{"final":863.2739076586209,"max_drawdown":-0.6120609741159966,"trades":1665,"exposure":0.3085922133663192,"assets_traded":193,"fees_paid":416.0349186167718}},"freq":"1d","equity":[["2021-10-17",106.0452],["2021-10-24",106.7614],["2021-10-31",105.319],["2021-11-07",139.0715],["2021-11-14",127.3549],["2021-11-21",115.6982],["2021-11-28",115.6982],["2021-12-05",115.6982],["2021-12-12",115.6982],["2021-12-19",115.6982],["2021-12-26",115.6982],["2022-01-02",115.6982],["2022-01-09",115.6982],["2022-01-16",115.6982],["2022-01-23",115.6982],["2022-01-30",115.6982],["2022-02-06",115.6982],["2022-02-13",110.0584],["2022-02-20",108.1704],["2022-02-27",108.1704],["2022-03-06",107.6378],["2022-03-13",107.6352],["2022-03-20",108.0878],["2022-03-27",111.2615],["2022-04-03",144.5107],["2022-04-10",100.9694],["2022-04-17",100.9694],["2022-04-24",100.9694],["2022-05-01",100.9694],["2022-05-08",100.9694],["2022-05-15",100.9694],["2022-05-22",100.9694],["2022-05-29",100.9694],["2022-06-05",100.9694],["2022-06-12",100.8471],["2022-06-19",100.8471],["2022-06-26",100.8471],["2022-07-03",100.8471],["2022-07-10",100.8471],["2022-07-17",100.8471],["2022-07-24",115.5699],["2022-07-31",149.5624],["2022-08-07",128.0246],["2022-08-14",132.0315],["2022-08-21",114.5656],["2022-08-28",114.5656],["2022-09-04",114.5656],["2022-09-11",114.5656],["2022-09-18",101.5356],["2022-09-25",101.5356],["2022-10-02",101.5356],["2022-10-09",101.5196],["2022-10-16",101.4992],["2022-10-23",101.4992],["2022-10-30",105.2916],["2022-11-06",120.8642],["2022-11-13",105.1219],["2022-11-20",105.1219],["2022-11-27",105.1219],["2022-12-04",105.1219],["2022-12-11",105.1213],["2022-12-18",105.1131],["2022-12-25",105.1161],["2023-01-01",105.1161],["2023-01-08",105.1161],["2023-01-15",174.632],["2023-01-22",182.4968],["2023-01-29",181.0257],["2023-02-05",171.3227],["2023-02-12",141.667],["2023-02-19",160.7306],["2023-02-26",145.1767],["2023-03-05",125.7584],["2023-03-12",125.6868],["2023-03-19",119.3179],["2023-03-26",107.4209],["2023-04-02",108.5455],["2023-04-09",99.6197],["2023-04-16",99.4949],["2023-04-23",103.5573],["2023-04-30",106.244],["2023-05-07",104.5423],["2023-05-14",104.5423],["2023-05-21",104.5423],["2023-05-28",104.5423],["2023-06-04",104.2843],["2023-06-11",104.2843],["2023-06-18",104.2843],["2023-06-25",104.3458],["2023-07-02",117.5643],["2023-07-09",153.2813],["2023-07-16",145.4371],["2023-07-23",152.376],["2023-07-30",144.531],["2023-08-06",144.531],["2023-08-13",144.531],["2023-08-20",144.531],["2023-08-27",144.531],["2023-09-03",144.531],["2023-09-10",144.531],["2023-09-17",144.531],["2023-09-24",144.466],["2023-10-01",144.9081],["2023-10-08",144.3007],["2023-10-15",137.1066],["2023-10-22",159.6377],["2023-10-29",164.308],["2023-11-05",226.8185],["2023-11-12",256.6129],["2023-11-19",242.0507],["2023-11-26",242.67],["2023-12-03",262.636],["2023-12-10",257.546],["2023-12-17",259.6432],["2023-12-24",379.7321],["2023-12-31",358.6867],["2024-01-07",430.2148],["2024-01-14",383.5688],["2024-01-21",383.5688],["2024-01-28",383.5688],["2024-02-04",383.5691],["2024-02-11",382.7546],["2024-02-18",340.7578],["2024-02-25",457.4359],["2024-03-03",741.72],["2024-03-10",871.5267],["2024-03-17",916.7493],["2024-03-24",916.9579],["2024-03-31",1137.012],["2024-04-07",1114.2008],["2024-04-14",1119.4808],["2024-04-21",1119.4716],["2024-04-28",1119.4716],["2024-05-05",1119.4716],["2024-05-12",1119.4716],["2024-05-19",1119.4766],["2024-05-26",1795.802],["2024-06-02",2225.2825],["2024-06-09",1658.7558],["2024-06-16",1658.6157],["2024-06-23",1658.6157],["2024-06-30",1658.6157],["2024-07-07",1658.6157],["2024-07-14",1658.6157],["2024-07-21",1654.9143],["2024-07-28",1602.9508],["2024-08-04",1592.0046],["2024-08-11",1596.895],["2024-08-18",1596.895],["2024-08-25",1571.4931],["2024-09-01",1404.2951],["2024-09-08",1404.2951],["2024-09-15",1417.569],["2024-09-22",1348.7718],["2024-09-29",1433.9444],["2024-10-06",1326.8518],["2024-10-13",1285.7321],["2024-10-20",1276.3671],["2024-10-27",1277.6324],["2024-11-03",1250.169],["2024-11-10",1394.383],["2024-11-17",1431.6159],["2024-11-24",1528.3127],["2024-12-01",1449.0723],["2024-12-08",1523.9036],["2024-12-15",1459.1861],["2024-12-22",1303.8781],["2024-12-29",1255.9818],["2025-01-05",1238.2549],["2025-01-12",1170.9459],["2025-01-19",1360.6113],["2025-01-26",1312.4291],["2025-02-02",1169.519],["2025-02-09",1155.807],["2025-02-16",1155.807],["2025-02-23",1155.807],["2025-03-02",1155.807],["2025-03-09",1155.807],["2025-03-16",1155.807],["2025-03-23",1155.807],["2025-03-30",1148.3622],["2025-04-06",1148.3622],["2025-04-13",1148.3622],["2025-04-20",1148.3622],["2025-04-27",1180.0418],["2025-05-04",1457.0054],["2025-05-11",1868.9992],["2025-05-18",1438.108],["2025-05-25",1468.3561],["2025-06-01",1232.0985],["2025-06-08",1223.3042],["2025-06-15",1272.5877],["2025-06-22",1272.1032],["2025-06-29",1272.4123],["2025-07-06",1274.0844],["2025-07-13",1376.3883],["2025-07-20",1385.1664],["2025-07-27",1363.0537],["2025-08-03",1251.178],["2025-08-10",1318.6339],["2025-08-17",1352.0803],["2025-08-24",1328.7577],["2025-08-31",1328.3053],["2025-09-07",1328.3053],["2025-09-14",1387.9719],["2025-09-21",1370.2131],["2025-09-28",1259.7708],["2025-10-05",1332.314],["2025-10-12",1226.5019],["2025-10-19",1226.5019],["2025-10-26",1226.5019],["2025-11-02",1226.5019],["2025-11-09",1226.5019],["2025-11-16",1226.5019],["2025-11-23",1226.5019],["2025-11-30",1226.5019],["2025-12-07",1226.5019],["2025-12-14",1226.5019],["2025-12-21",1226.5019],["2025-12-28",1226.5019],["2026-01-04",1212.2632],["2026-01-11",1527.5933],["2026-01-18",1853.1347],["2026-01-25",1922.5515],["2026-02-01",1922.5515],["2026-02-08",1922.5515],["2026-02-15",1922.5515],["2026-02-22",1922.5515],["2026-03-01",1922.5515],["2026-03-08",1922.5472],["2026-03-15",1922.5507],["2026-03-22",1704.1615],["2026-03-29",1704.1615],["2026-04-05",1704.1615],["2026-04-12",1704.3911],["2026-04-19",1740.2041],["2026-04-26",1432.2254],["2026-05-03",1427.8103],["2026-05-10",1458.4387],["2026-05-17",1329.4284],["2026-05-24",1327.9434],["2026-05-31",1327.9434],["2026-06-07",1327.9434],["2026-06-14",1327.9434],["2026-06-21",1327.9434],["2026-06-28",1327.9434],["2026-07-05",1327.9434],["2026-07-12",1305.5756],["2026-07-19",1253.1513],["2026-07-26",1251.6285],["2026-08-02",1233.1914],["2026-08-09",1234.1886],["2026-08-16",1229.9669],["2026-08-23",1351.6681],["2026-08-30",1504.0942],["2026-09-06",1827.0623],["2026-09-13",1456.1321],["2026-09-20",1274.8025],["2026-09-27",1110.045],["2026-10-04",1038.2599],["2026-10-11",863.2739]],"code_bytes":8239},"a28":{"id":"a28","alias":"Phalarope","theme":"Ensembles of weak rules combined online with multiplicative weights / Hedge","started":"2026-10-11T01:18:20+00:00","rounds":[{"k":1,"minutes":39.3},{"k":2,"minutes":14.5}],"messages":[{"t":"2026-10-11T01:18:25.585Z","text":"I’ll inspect the simulator and data, submit an early baseline, then test an online ensemble of causal rules. I’ll record the variants and their equations as I go."},{"t":"2026-10-11T01:19:02.337Z","text":"The cost model makes frequent switching expensive. I’m focusing on liquid momentum leaders, with an online ensemble that shifts capital toward rules that have recently performed well after costs. The BTC baseline is awaiting the referee."},{"t":"2026-10-11T01:20:34.809Z","text":"The first sweep found some momentum rules with strong gains, but they also suffered severe drawdowns. That points to concentration and crash exposure as the next issues to test. I’ll add volatility sizing and faster exits before relying on the ensemble."},{"t":"2026-10-11T01:22:54.728Z","text":"Several of the worst losses came from coins selected after sharp price spikes, followed by reversals. The ensemble now includes rules that favor smoother trends, caps exposure using recent volatility, and scores each rule on a shadow portfolio that pays the referee’s trading costs."},{"t":"2026-10-11T01:24:17.005Z","text":"The broad momentum ensemble has not improved on the baseline so far. I’m testing two alternatives: slower trend rules on BTC, ETH and SOL, and volume-confirmed entries that avoid chasing the largest recent spike. I’ll keep whichever offers better evidence after costs."},{"t":"2026-10-11T01:27:25.473Z","text":"The major-coin trend tests look more stable: several nearby SOL rules returned roughly $650–$816 from $100, with drawdowns around 36–48%. I’m also testing a walk-forward model on price, volume and volatility features, training only on outcomes already known at each retraining point."},{"t":"2026-10-11T01:29:05.286Z","text":"I’ve submitted the stronger daily rule. The local referee run passed its truncation checks and returned $7,398 from $100, but drawdown reached 78.7%. I’m checking neighboring parameters and faster responses to volatility spikes to see whether that return survives with less downside."},{"t":"2026-10-11T01:34:48.266Z","text":"Using the current day’s price range alongside rolling volatility helped: the best test returned about $14,296, with drawdown reduced to 69.3%. Performance-weighted Hedge has weakened the ensembles so far. Equal weighting did better, so I’m testing whether agreement among the rules gives a more stable basis for allocation."},{"t":"2026-10-11T01:40:09.937Z","text":"The three-day trading schedule is sensitive to its starting phase. I’m testing a version that holds all three phases in parallel, so it doesn’t depend on picking the phase that looked best in this sample. I’ve also added a daily risk reduction trigger for volatility jumps between scheduled trades."},{"t":"2026-10-11T01:43:44.243Z","text":"A rule that switches only after two consecutive leader signals passed the referee’s checks: $100 became $17,799.60, with 62.2% drawdown. Nearby switching thresholds also performed well. I’m submitting it now, then testing a final blend of confirmed leaders and staggered schedules."},{"t":"2026-10-11T01:46:12.346Z","text":"The blended strategy is ready for its final checks. It gives both timing families an initial 50% budget, then uses a slowly forgetting Hedge rule to adjust their shares from after-cost shadow results. I’m checking causality, annual results, doubled costs, and how much profit came from the best ten days."},{"t":"2026-10-11T01:52:08.800Z","text":"The blend is accepted at $19,738.22, rank 4, with 61.3% drawdown. All 27 nearby parameter combinations remained profitable. A slightly lower turnover threshold improved both return and drawdown in the local check; I’m verifying that final change and updating the writeup with the cost and concentration results."},{"t":"2026-10-11T01:55:53.044Z","text":"The final version passed causality checks: $100 became $23,607.47 locally, with 60.2% drawdown. Its sixth submission is awaiting acceptance. The writeup now includes the equations, search history, failed ideas and robustness checks; doubled costs still left $12,789.74, though gains remain concentrated in a few exceptional days."},{"t":"2026-10-11T01:57:37.321Z","text":"Entry: #006 accepted, rank 5/40; strategy and writeup.  \nVisible years: $100 → **$23,607.47** (×236.1); maximum drawdown **60.2%**.  \nNext idea: test a causal filter for exhausted leaders and crowded rallies."},{"t":"2026-10-11T01:57:45.301Z","text":"I’ll keep the accepted eight-policy ensemble as the fallback and test a filter for exhausted leaders. I’ll only replace it if the result supports a better hidden-year candidate, with time left for scoring and the final writeup."},{"t":"2026-10-11T02:00:02.675Z","text":"The accepted entry already passes the causal checks, and its writeup is complete. Its main weakness is dependence on a few large trend bursts. I’m testing whether price and volume signals available at each close can reduce losses around those bursts without removing too much of the upside."},{"t":"2026-10-11T02:01:08.826Z","text":"The exhaustion filters mostly hurt performance. Only a rare negative-shock cutoff improved the result slightly, so I’m checking its nearby thresholds before considering it. I’m also testing listing age and the liquidity window, which affect how quickly the strategy can recognize new leaders."},{"t":"2026-10-11T02:02:14.323Z","text":"A cutoff for daily losses greater than four times the prior volatility estimate improved wealth to about $25,956 and reduced drawdown slightly. Thresholds from 3.5 to 5 produced the same result, so the benefit survives a small threshold change. I’ll submit this version and keep testing before finalizing the writeup."},{"t":"2026-10-11T02:03:57.582Z","text":"Submission #007 is accepted: $100 became $25,955.50, with 59.3% maximum drawdown. Shorter liquidity windows, younger listings, and blended momentum horizons did not improve it. I’m now comparing standard range-based volatility estimators to check whether the position cap is too conservative during orderly trends."},{"t":"2026-10-11T02:06:23.437Z","text":"Reducing exposure when the policies disagree on the leading asset improved both wealth and drawdown across all four settings tested. With one agreement measure, the research result is about $29,099 at 55.8% drawdown. I’m checking nearby settings and the earlier data before deciding whether to make it the final entry."},{"t":"2026-10-11T02:07:19.279Z","text":"The agreement measure held up in nearby settings and slightly improved the earlier-period result while reducing drawdown. I’m keeping it in the final entry, alongside the negative-shock filter. The writeup will also make the remaining risk clear: most growth still comes from a small number of trend bursts."},{"t":"2026-10-11T02:08:53.733Z","text":"Submission #008 is accepted and passes the causal checks. It turns $100 into $29,099.28 on the visible years, with 55.8% maximum drawdown and 31% average exposure. I’m retaining it and updating the writeup with the full search count, final equations, and sensitivity results."},{"t":"2026-10-11T02:12:05.190Z","text":"Accepted submission **#008**, strategy.py, is my final entry: eight daily leadership policies combined through Hedge, with less exposure when policies disagree and filters for volatility and negative shocks. Visible result: **$100 → $29,099.28 (×290.99)**, **55.8% maximum drawdown**, ranked fifth at acceptance. Causal checks passed, and the local file matches the accepted submission. WRITEUP.md is complete with equations, 1,333 logged configuration evaluations, results, failures, and generalization limits. Hidden-year performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T01:18:38.001213+00:00","note":"Initial causal BTC trend baseline; research on adaptive ensembles follows.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:28:30.821216+00:00","note":"Liquid dominant momentum with causal BTC regime filter and a daily-volatility cap; validated on exact net fills.","status":"ok","dev":7398.258876963639,"dd":-0.7871569507532024,"reason":""},{"n":"003","t":"2026-10-11T01:35:35.132273+00:00","note":"Range-sensitive exposure cap responds to volatility shocks while preserving generic liquid leadership; causal checks passed.","status":"ok","dev":14296.396198676419,"dd":-0.6931804839609222,"reason":""},{"n":"004","t":"2026-10-11T01:43:52.573158+00:00","note":"Confirmed leaders replace calendar-driven asset switching; range-aware sizing and daily risk cuts. Exact causal evaluation passed.","status":"ok","dev":17799.597774502115,"dd":-0.6215302332579997,"reason":""},{"n":"005","t":"2026-10-11T01:49:22.767415+00:00","note":"Eight causal leadership experts: staggered schedules and confirmed switches, range-sensitive risk, after-cost shadow Hedge with fixed share. Prefix checks passed.","status":"ok","dev":19738.223292040773,"dd":-0.612600227061673,"reason":""},{"n":"006","t":"2026-10-11T01:54:40.188833+00:00","note":"Final turnover refinement after neighborhood checks: $23,607 locally, lower drawdown; same causal eight-policy Hedge model. Full equations and failures documented.","status":"ok","dev":23607.47352806224,"dd":-0.6016832652350999,"reason":""},{"n":"007","t":"2026-10-11T02:02:17.976903+00:00","note":"Eight-policy after-cost Hedge; reject extreme negative shocks using prior volatility, with threshold plateau checked.","status":"ok","dev":25955.502705915744,"dd":-0.5925813826276184,"reason":""},{"n":"008","t":"2026-10-11T02:07:28.827052+00:00","note":"Final: after-cost Hedge over eight leadership policies, agreement-scaled exposure, prior-volatility crash filter, all rebalance phases.","status":"ok","dev":29099.282045191063,"dd":-0.557669553698914,"reason":""}],"entry":"008","round_notes":[{"k":1,"text":"Entry: #006 accepted, rank 5/40; strategy and writeup.  \nVisible years: $100 → **$23,607.47** (×236.1); maximum drawdown **60.2%**.  \nNext idea: test a causal filter for exhausted leaders and crowded rallies."},{"k":2,"text":"Accepted submission **#008**, strategy.py, is my final entry: eight daily leadership policies combined through Hedge, with less exposure when policies disagree and filters for volatility and negative shocks. Visible result: **$100 → $29,099.28 (×290.99)**, **55.8% maximum drawdown**, ranked fifth at acceptance. Causal checks passed, and the local file matches the accepted submission. WRITEUP.md is complete with equations, 1,333 logged configuration evaluations, results, failures, and generalization limits. Hidden-year performance remains unknown."}],"title":"Phalarope (a28): writeup","idea":"The entry is an online ensemble of eight daily, long-only leadership policies. Every policy selects from a changing universe ranked by trailing Binance quote volume, rather than from a fixed list of successful coins. Six policies rebalance on all three phases of a three-day schedule; two use a persistent leader and require confirmation before switching. Both families use the same market and asset trend filters and a range-sensitive volatility cap. A discounted, after-cost Hedge allocator starts with half the capital budget in each timing family and keeps a small prior allocation to every policy. A normalized agreement probability scales the combined exposure down when the policies recommend different assets, reserving cash when their evidence is less consistent. A large negative surprise relative to prior volatility temporarily disqualifies an asset. This combines two ways of responding to emerging leaders while reducing dependence on a fortunate rebalance phase. The final implementation is strategy.py, a self-contained daily strategy using numpy, pandas and numba.","failed":"- Broad long-horizon momentum and many volatility-normalized rankings: weaker than short leadership after costs, often with deep drawdowns. - Several-coin top-rank baskets: reduced some concentration but often diluted the profitable leaders without removing common crash exposure. - Volume-confirmed Donchian entries: typically chased exhausted moves and lost money in the tested specifications. - Pullback scores combining medium momentum with strong one-day reversal: generally weak after their switching costs. - Fast Hedge on diverse horizon and universe grids: chasing recent expert winners often selected a rule after its strongest burst. Reward clipping helped some settings but did not solve the problem. Negative learning rates sometimes improved these early ensembles, but were not retained. - Four-hour major-coin trends: the strongest nearby SOL specifications returned roughly $650–$816, with lower drawdowns, but were less competitive in visible dollars. - A purged walk-forward pooled histogram-boosting model, using 26 causal price, volume, volatility and sentiment features: best tested portfolio was about $219. Labels were three-day returns, clipped to +/-0.35, trained only once fully observed. Retraining every 30 bars and strict three-bar purging did not produce useful enough net selection. - Tight stop and cooldown rules: some lowered drawdown, while giving up too many trend recoveries and substantial terminal wealth. - Most final exhaustion filters: positive-surprise caps, upper-wick exclusions, volume distribution/fading filters, and weekly extension caps usually removed profitable bursts. The rare negative-shock gate was the retained exception. - Younger listing-age thresholds, faster volume-ranking windows, and most momentum blends: substantially weaker results. ","holdup":"Selection is based on contemporaneous liquidity, trend, and relative performance, so future leaders can enter without a known ticker identity. Listed and delisted assets are handled using the available history rather than the final survivor set. Trading-cost accounting in expert rewards discourages successful-looking policies whose turnover consumes their gains. The range term addresses an observable failure of slow volatility estimates during abrupt price expansion. The negative-shock comparison uses the preceding estimate, which makes an abrupt loss observable without letting it enlarge its own reference scale. Confirmation and hysteresis reduce the need to buy every new daily leader. Parallel scheduling phases remove a calendar-phase choice from the scheduled family's allocation. Agreement sizing reserves more cash when timing policies identify conflicting leaders. It can still fail when the policies agree on the same bad trade; agreement is evidence consistency, not certainty. The online learner can gradually change the timing mix as observed net performance changes, while its prior share keeps alternatives available. These mechanisms are justified by the visible data and ordin","result":{"dev":{"final":29099.282045191063,"max_drawdown":-0.557669553698914,"trades":858,"exposure":0.31371083628782764,"assets_traded":127,"fees_paid":7418.350335359197},"hidden":{"final":7.857679342598129,"max_drawdown":-0.9633766128656895,"trades":720,"exposure":0.3109505046237307,"assets_traded":125,"fees_paid":24.572400341853143},"full":{"final":2285.514848026164,"max_drawdown":-0.9652291350082894,"trades":1581,"exposure":0.31260730829506167,"assets_traded":216,"fees_paid":14577.796859142618}},"freq":"1d","equity":[["2021-10-17",102.5868],["2021-10-24",89.8717],["2021-10-31",137.2677],["2021-11-07",130.9649],["2021-11-14",184.3378],["2021-11-21",188.0731],["2021-11-28",188.0731],["2021-12-05",188.0731],["2021-12-12",188.0731],["2021-12-19",188.0731],["2021-12-26",188.0731],["2022-01-02",188.0731],["2022-01-09",188.0731],["2022-01-16",188.0731],["2022-01-23",188.0731],["2022-01-30",188.0731],["2022-02-06",210.7921],["2022-02-13",279.2332],["2022-02-20",257.0186],["2022-02-27",257.0186],["2022-03-06",242.8626],["2022-03-13",236.791],["2022-03-20",294.6784],["2022-03-27",303.3585],["2022-04-03",490.8745],["2022-04-10",413.6451],["2022-04-17",413.6451],["2022-04-24",413.6451],["2022-05-01",413.6451],["2022-05-08",413.6451],["2022-05-15",413.6451],["2022-05-22",413.6451],["2022-05-29",413.6451],["2022-06-05",413.6451],["2022-06-12",390.1275],["2022-06-19",390.1275],["2022-06-26",390.1275],["2022-07-03",390.1275],["2022-07-10",390.1275],["2022-07-17",390.1275],["2022-07-24",586.4297],["2022-07-31",754.416],["2022-08-07",637.7771],["2022-08-14",569.5562],["2022-08-21",493.9906],["2022-08-28",493.9906],["2022-09-04",493.9906],["2022-09-11",514.2639],["2022-09-18",484.2407],["2022-09-25",484.2407],["2022-10-02",484.2407],["2022-10-09",426.4331],["2022-10-16",426.4331],["2022-10-23",414.7561],["2022-10-30",534.8205],["2022-11-06",789.5819],["2022-11-13",655.2665],["2022-11-20",655.2665],["2022-11-27",655.2665],["2022-12-04",655.2665],["2022-12-11",635.8384],["2022-12-18",382.9349],["2022-12-25",382.9349],["2023-01-01",382.9349],["2023-01-08",376.1773],["2023-01-15",570.98],["2023-01-22",580.7097],["2023-01-29",741.5305],["2023-02-05",724.9332],["2023-02-12",742.8705],["2023-02-19",726.3573],["2023-02-26",1016.9714],["2023-03-05",1032.567],["2023-03-12",1032.567],["2023-03-19",1231.3528],["2023-03-26",1267.9335],["2023-04-02",1557.3582],["2023-04-09",1566.3315],["2023-04-16",1568.0272],["2023-04-23",1508.5851],["2023-04-30",1527.8785],["2023-05-07",1487.4328],["2023-05-14",1422.7351],["2023-05-21",1422.7351],["2023-05-28",1422.7351],["2023-06-04",1597.4467],["2023-06-11",1597.4467],["2023-06-18",1597.4467],["2023-06-25",1590.9954],["2023-07-02",2322.8228],["2023-07-09",3131.1988],["2023-07-16",2922.3418],["2023-07-23",3156.4539],["2023-07-30",3092.6169],["2023-08-06",3002.7867],["2023-08-13",2919.8189],["2023-08-20",2919.8189],["2023-08-27",2919.8189],["2023-09-03",2808.6107],["2023-09-10",2808.6107],["2023-09-17",3172.0442],["2023-09-24",2302.4969],["2023-10-01",2280.4279],["2023-10-08",2060.6137],["2023-10-15",2301.5222],["2023-10-22",2236.3265],["2023-10-29",2732.2906],["2023-11-05",3751.8702],["2023-11-12",5853.75],["2023-11-19",5995.1226],["2023-11-26",5872.8939],["2023-12-03",7330.8978],["2023-12-10",10445.7586],["2023-12-17",10542.8189],["2023-12-24",10777.8626],["2023-12-31",10477.8447],["2024-01-07",12911.4768],["2024-01-14",14307.4156],["2024-01-21",14307.4156],["2024-01-28",14307.4156],["2024-02-04",16318.8236],["2024-02-11",15363.5126],["2024-02-18",14831.5197],["2024-02-25",22421.46],["2024-03-03",42068.866],["2024-03-10",58133.4494],["2024-03-17",54195.5749],["2024-03-24",39405.9622],["2024-03-31",38844.3339],["2024-04-07",38021.3325],["2024-04-14",34744.1283],["2024-04-21",34744.1283],["2024-04-28",34275.5704],["2024-05-05",34275.5704],["2024-05-12",34275.5704],["2024-05-19",31506.9846],["2024-05-26",52239.7824],["2024-06-02",49483.3696],["2024-06-09",65730.7446],["2024-06-16",53890.4203],["2024-06-23",53890.4203],["2024-06-30",53890.4203],["2024-07-07",53890.4203],["2024-07-14",53890.4203],["2024-07-21",57322.0981],["2024-07-28",54708.2641],["2024-08-04",40066.7128],["2024-08-11",40066.7128],["2024-08-18",40066.7128],["2024-08-25",41700.3893],["2024-09-01",34686.6925],["2024-09-08",34686.6925],["2024-09-15",34544.6489],["2024-09-22",33083.0628],["2024-09-29",30660.5277],["2024-10-06",29603.7837],["2024-10-13",28825.7681],["2024-10-20",29861.1584],["2024-10-27",24313.0128],["2024-11-03",23847.9582],["2024-11-10",47866.8041],["2024-11-17",42571.3607],["2024-11-24",48168.3519],["2024-12-01",50371.876],["2024-12-08",57169.9875],["2024-12-15",42600.262],["2024-12-22",29842.7446],["2024-12-29",30977.693],["2025-01-05",31375.2771],["2025-01-12",29465.0939],["2025-01-19",35047.0573],["2025-01-26",33809.108],["2025-02-02",27756.0677],["2025-02-09",26892.7594],["2025-02-16",26892.7594],["2025-02-23",26471.0643],["2025-03-02",26471.0643],["2025-03-09",24100.5201],["2025-03-16",24100.5201],["2025-03-23",24100.5201],["2025-03-30",23454.8087],["2025-04-06",23454.8087],["2025-04-13",23454.8087],["2025-04-20",22007.4399],["2025-04-27",22908.5574],["2025-05-04",34581.8553],["2025-05-11",29345.0435],["2025-05-18",24720.0159],["2025-05-25",22870.4366],["2025-06-01",14405.319],["2025-06-08",12594.1969],["2025-06-15",10810.1896],["2025-06-22",10792.4822],["2025-06-29",10875.7343],["2025-07-06",12032.1862],["2025-07-13",10619.27],["2025-07-20",9687.3262],["2025-07-27",8780.4017],["2025-08-03",7520.6689],["2025-08-10",7431.5716],["2025-08-17",7001.3691],["2025-08-24",6563.0088],["2025-08-31",6558.029],["2025-09-07",6558.029],["2025-09-14",6370.9473],["2025-09-21",5966.6808],["2025-09-28",5810.2883],["2025-10-05",5898.8295],["2025-10-12",6192.7641],["2025-10-19",6192.7641],["2025-10-26",6192.7641],["2025-11-02",6200.0768],["2025-11-09",6200.0768],["2025-11-16",6200.0768],["2025-11-23",6200.0768],["2025-11-30",6200.0768],["2025-12-07",6200.0768],["2025-12-14",6200.0768],["2025-12-21",6200.0768],["2025-12-28",6200.0768],["2026-01-04",6254.0191],["2026-01-11",7285.1494],["2026-01-18",6400.9828],["2026-01-25",5988.7234],["2026-02-01",5988.7234],["2026-02-08",5988.7234],["2026-02-15",5988.7234],["2026-02-22",5988.7234],["2026-03-01",5988.7234],["2026-03-08",5703.6892],["2026-03-15",5542.4084],["2026-03-22",4461.9116],["2026-03-29",4333.5638],["2026-04-05",4333.5638],["2026-04-12",4749.4252],["2026-04-19",12629.997],["2026-04-26",9726.0606],["2026-05-03",9448.1034],["2026-05-10",7386.0606],["2026-05-17",5469.2248],["2026-05-24",5462.8276],["2026-05-31",5462.8276],["2026-06-07",5462.8276],["2026-06-14",5462.8276],["2026-06-21",5462.8276],["2026-06-28",5462.8276],["2026-07-05",5462.8276],["2026-07-12",4854.3816],["2026-07-19",4360.5021],["2026-07-26",4571.9334],["2026-08-02",4245.9193],["2026-08-09",4308.5439],["2026-08-16",4016.2004],["2026-08-23",4454.2262],["2026-08-30",3989.4912],["2026-09-06",3594.9905],["2026-09-13",2815.0463],["2026-09-20",2597.3294],["2026-09-27",2529.2618],["2026-10-04",2503.1543],["2026-10-11",2285.5148]],"code_bytes":7837},"a29":{"id":"a29","alias":"Pratincole","theme":"Lead-lag structure: does BTC (or the market) lead other coins, and at what horizon","started":"2026-10-11T01:19:43+00:00","rounds":[{"k":1,"minutes":39.1},{"k":2,"minutes":10.1}],"messages":[{"t":"2026-10-11T01:19:49.344Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test whether BTC lead-lag signals improve on momentum after costs. I’ll track the variants and results for the writeup."},{"t":"2026-10-11T01:20:27.903Z","text":"The baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The data ends at the allowed cutoff. I’ll now test BTC-led catch-up signals alongside momentum rotation, with trailing liquidity filters and separate results for the falling and recovering markets."},{"t":"2026-10-11T01:24:12.887Z","text":"Daily BTC catch-up and pump-following rules have performed poorly after costs; none improved on the baseline. I’ll test shorter horizons in BTC, ETH and SOL, and a walk-forward model that learns how price and volume signals interact. Its training labels will be limited to returns already observed at each decision."},{"t":"2026-10-11T01:29:50.846Z","text":"A 4-hour rule has improved the visible result to $464.50 and passed the truncation check. It uses a daily BTC trend filter, rotates among BTC, ETH and SOL, and exits when the held coin weakens. All three visible annual periods were profitable. I’ll submit it while testing whether slower changes in trading volume offer a stronger signal."},{"t":"2026-10-11T01:33:11.902Z","text":"The volume and walk-forward model tests have not improved on the accepted entry. One plausible missing filter is market breadth: BTC can trend up while most altcoins fall. I’ll test whether requiring a broad recovery makes altcoin rotation worthwhile, then check the strongest rule’s sensitivity to parameters and costs."},{"t":"2026-10-11T01:41:12.951Z","text":"The best hourly variant reached about $766, with a 36.5% drawdown and gains in each visible annual period. The improvement came from a faster trading cadence and a slower coin trend filter. Breadth helped only slightly, so I’ll leave that extra filter out for now. I’m checking causality and nearby parameter choices before updating the entry."},{"t":"2026-10-11T01:44:27.013Z","text":"A BTC-based growth model has reached about $884 in the research run. It projects BTC’s recent drift through each coin’s rolling beta, accounts for volatility, and caps the chosen position at 100% of capital. It also gained in each visible annual period. I’m checking the exact implementation and how sensitive that result is to the estimation windows."},{"t":"2026-10-11T01:46:20.340Z","text":"The exact simulator confirmed $883.64: 106.7% annual growth, a 42.9% maximum drawdown, and $71.41 in costs. The causal check passed, and I’ve submitted the file. The remaining tests will check nearby parameters, doubled fees, and how much profit depends on the strongest days."},{"t":"2026-10-11T01:48:40.422Z","text":"The entry is accepted. With doubled fees, $100 still became $759. Nearby parameter choices stayed profitable, though shortening the drift window cut the result substantially. The best 10 days contributed about 70% of log profit. I’ll test averaging adjacent drift windows to reduce dependence on one lookback choice, then finish the writeup."},{"t":"2026-10-11T01:50:33.529Z","text":"Averaging the drift windows did not improve the result. The single 24-day estimate reached $1,075 in research, with gains of roughly 49%, 88% and 283% across the three visible annual periods. I’ll verify that version and keep its simpler forecast if it passes. The writeup will include the full search count and failed approaches."},{"t":"2026-10-11T01:53:06.061Z","text":"The 24-day version is accepted: $100 became $1,075.10, with a 42.5% maximum drawdown. At doubled fees it still reached $940.13. I’m finishing the equations and research record now, including the heavy dependence on a few strong days and the limits of testing many variants on the same visible sample."},{"t":"2026-10-11T01:55:23.674Z","text":"The stop check found a useful improvement: a 12% trailing stop raised the exact result to $1,167.84 and reduced maximum drawdown to 31.8%. It passed the causal check; doubled fees still left $1,006.12. I’ll submit this version when the five-minute interval allows it and record the full maths and diagnostics in the writeup."},{"t":"2026-10-11T01:58:30.515Z","text":"The writeup is complete, including every signal and sizing equation, roughly 1,100 tested configurations, and the failed approaches. The final parameter checks ranged from $682 to $1,296. I’m checking the last submission’s acceptance so the round status names the actual entry."},{"t":"2026-10-11T01:58:50.787Z","text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: $100 → **$1,167.84**; CAGR **126.8%**; max drawdown **−31.8%**.  \nNext idea: covariance-aware allocation across BTC, ETH and SOL."},{"t":"2026-10-11T01:58:59.814Z","text":"I’ll test whether splitting capital across the majors improves the BTC-projected growth model. The accepted entry and writeup are already complete; I’ll replace the entry only if a new version shows a useful improvement after costs and causal checks, with enough time to score before sealing."},{"t":"2026-10-11T02:01:08.123Z","text":"The accepted rule uses BTC’s 24-day trend to choose and size one of BTC, ETH or SOL. I’m testing whether a portfolio using their rolling covariance can reduce drawdown while preserving growth. The test uses the same trend gates and closing-price stops."},{"t":"2026-10-11T02:02:47.982Z","text":"The covariance test finished close to the accepted rule: $1,160.83 with 29.98% drawdown, versus $1,167.84 with 31.82%. It required more trades. I’m keeping the current entry while testing the same BTC-factor forecast across a universe selected by trailing volume."},{"t":"2026-10-11T02:04:08.791Z","text":"Extending the factor rule to other coins performed poorly: the best of 24 variants ended at $385.76 with 68.84% drawdown. I’ll keep the majors-only entry and test one remaining lead-lag idea: whether a sharp BTC move creates a tradable catch-up opportunity on 15-minute bars."},{"t":"2026-10-11T02:05:45.095Z","text":"The 15-minute catch-up test also fell short: its best variant ended at $175.25 after costs. Slower growth rotation across minor coins did worse. These tests support keeping the current majors-only rule; I’m now checking its exact source and documenting the final-round results."},{"t":"2026-10-11T02:08:39.381Z","text":"I’m retaining accepted submission #005. The final file matches both the accepted candidate and its passing causal report: $100 became $1,167.84, with 31.82% maximum drawdown. The writeup now includes the equations, 1,149 distinct recorded configurations, all 83 final-round tests, and the limits of the validation."},{"t":"2026-10-11T02:08:58.671Z","text":"Final entry: accepted submission **#005**, pratincole.py. It trades BTC, ETH and SOL using BTC-projected drift, variance-based sizing, trend filters and a 12% trailing stop. Visible years turned **$100 into $1,167.84 (11.68×)** with **31.82% maximum drawdown**; causal checks passed. Final-round alternatives did not justify replacing it. WRITEUP.md is complete with equations, parameters, results, 1,149 recorded configurations, failed approaches and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T01:20:01.642155+00:00","note":"Initial entry: causal BTC 200-day trend filter; baseline before lead-lag research.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:29:53.091344+00:00","note":"BTC-confirmed 4h relative strength rotation with trend exits and trailing stops; full causal check passed.","status":"ok","dev":464.49829544480355,"dd":-0.41835961768801244,"reason":""},{"n":"003","t":"2026-10-11T01:45:30.590242+00:00","note":"BTC trend transmission: rolling-beta drift forecasts, variance-adjusted growth ranking, constrained Kelly sizing, trend exits. Exact visible $883.64; causal check passed.","status":"ok","dev":883.641584363283,"dd":-0.42875968131916353,"reason":""},{"n":"004","t":"2026-10-11T01:51:39.800322+00:00","note":"BTC drift transmission, 24-day market estimate, rolling beta and constrained Kelly fractions; all three visible annual periods profitable; $1,075.10 exact and causal check passed.","status":"ok","dev":1075.096219716669,"dd":-0.42503970111488065,"reason":""},{"n":"005","t":"2026-10-11T01:57:08.829147+00:00","note":"Final round-one entry: BTC-projected growth with 24-day drift, rolling beta, capped Kelly sizing and a 12% trailing stop. Exact $1,167.84, -31.8% drawdown, causal check passed; doubled fees $1,006.12.","status":"ok","dev":1167.8439028587516,"dd":-0.3181834738296122,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: #005 accepted — strategy, writeup.  \nVisible years: $100 → **$1,167.84**; CAGR **126.8%**; max drawdown **−31.8%**.  \nNext idea: covariance-aware allocation across BTC, ETH and SOL."},{"k":2,"text":"Final entry: accepted submission **#005**, pratincole.py. It trades BTC, ETH and SOL using BTC-projected drift, variance-based sizing, trend filters and a 12% trailing stop. Visible years turned **$100 into $1,167.84 (11.68×)** with **31.82% maximum drawdown**; causal checks passed. Final-round alternatives did not justify replacing it. WRITEUP.md is complete with equations, parameters, results, 1,149 recorded configurations, failed approaches and generalization limits."}],"title":"Pratincole (a29): BTC trend transmission with constrained growth sizing","idea":"Short-horizon BTC lead-lag and catch-up rules did not survive costs in my tests. The useful version was slower: estimate BTC's market drift, project it through each major coin's rolling BTC beta, and choose the coin with the greatest estimated log growth at a position size constrained to [0,1]. Individual coin drift is shrunk to zero relative to that market factor; chasing each coin's own recent exceptional return performed poorly. A daily BTC regime, each coin's price trend, switching hysteresis and a closing-price trailing stop control participation. The final candidate trades BTC, ETH and SOL at four-hour frequency, with the uninvested fraction in cash. Its visible result is $1,167.84 from $100. This remains far below the growth rate required for the five-year goal.","failed":"- Daily catch-up ranked beta times BTC's recent return minus the coin's return. The best of nine versions reached about $13.20. A rolling lag coefficient, Cov(asset market residual at t, BTC return at t-1) divided by lagged BTC variance, also failed: its best nine-version result was about $12.99. Liquidity and own-trend filters did not rescue these prototypes. - Four-hour lag-coefficient and catch-up families reached at best about $339 and $170, respectively, below medium-term trend rules. This did not establish a short-horizon net arbitrage edge. - Broad daily momentum, reversal, volume-conditioned pump entries, money flow and OBV often bought exhausted moves. The best quiet-breakout version reached about $50. Very high market breadth improved some momentum tests, but a $550 version still had roughly 72% drawdown. - Generic observed listing-age, first-dip and early-strength rules did not work: the best of 54 newborn variants reached about $82. - Pooled histogram gradient boosting used causal returns, beta/lag-beta, candle, volume, volatility and fear/greed features. Forward 1-, 3- and 7-day labels were used only after maturity; models retrained every 90 days with 360- or 720-day windows and seed 29. The best BTC-residual-target allocation reached $391 but lost in the last annual period; raw-return targets reached at best about $119. These models did not justify shipping a fitted predictor. - Local level/velocity Kalman filters reached at best about $589. Their complexity did not improve the final growth rule. - Adjacent-window forecast averaging and reduced Kelly fractions did not consistently improve return or drawdown. The final model keeps one drift window and the unscaled fraction, capped at one.","holdup":"The hypothesis is that persistent market drift transmits through changing asset exposures over weeks, while coin-specific return outliers are less forecastable. The rule estimates exposures and variance afresh, requires a broad advancing BTC regime, and leaves capital in cash when participation conditions fail. It never borrows. Own-price filtering and stops address disagreement between the common-factor forecast and the coin. Trading the low-cost major market avoids relying on the much higher turnover costs or small liquidity of minor coins. These are reasons to test it forward, not proof that visible growth persists. Drift and beta estimates can be wrong; the quadratic objective omits jumps; false trends and repeated entries can lose money. A few strong days dominate profit. Selecting from 1,149 recorded configurations inflates the apparent result. No hidden prices, events, listings or future winners were used. Neither the hidden two-year nor full five-year account has been evaluated here. The visible result does not establish a route to $1,000,000.","result":{"dev":{"final":1167.8439028587516,"max_drawdown":-0.3181834738296122,"trades":166,"exposure":0.30069486562762554,"assets_traded":3,"fees_paid":92.88649744629471},"hidden":{"final":136.65550132613672,"max_drawdown":-0.3038421907473109,"trades":105,"exposure":0.3186120070004556,"assets_traded":3,"fees_paid":12.141374861833498},"full":{"final":1595.9229401583455,"max_drawdown":-0.3551992062518028,"trades":271,"exposure":0.3078577972827,"assets_traded":3,"fees_paid":234.67880349344227}},"freq":"4h","equity":[["2021-10-10",98.6054],["2021-10-17",105.2522],["2021-10-24",112.9841],["2021-10-31",118.7071],["2021-11-07",128.6428],["2021-11-14",130.4539],["2021-11-21",123.6974],["2021-11-28",123.6974],["2021-12-05",123.6974],["2021-12-12",123.6974],["2021-12-19",123.6974],["2021-12-26",123.6974],["2022-01-02",123.6974],["2022-01-09",123.6974],["2022-01-16",123.6974],["2022-01-23",123.6974],["2022-01-30",123.6974],["2022-02-06",123.6974],["2022-02-13",123.6974],["2022-02-20",123.6974],["2022-02-27",123.6974],["2022-03-06",123.6974],["2022-03-13",123.6974],["2022-03-20",123.6974],["2022-03-27",138.7804],["2022-04-03",179.5682],["2022-04-10",156.2493],["2022-04-17",151.7375],["2022-04-24",151.7375],["2022-05-01",151.7375],["2022-05-08",151.7375],["2022-05-15",151.7375],["2022-05-22",151.7375],["2022-05-29",151.7375],["2022-06-05",151.7375],["2022-06-12",151.7375],["2022-06-19",151.7375],["2022-06-26",151.7375],["2022-07-03",151.7375],["2022-07-10",151.7375],["2022-07-17",151.7375],["2022-07-24",151.7375],["2022-07-31",151.7375],["2022-08-07",151.7375],["2022-08-14",151.7375],["2022-08-21",151.7375],["2022-08-28",151.7375],["2022-09-04",151.7375],["2022-09-11",151.7375],["2022-09-18",151.7375],["2022-09-25",151.7375],["2022-10-02",151.7375],["2022-10-09",151.7375],["2022-10-16",151.7375],["2022-10-23",151.7375],["2022-10-30",151.7375],["2022-11-06",148.4183],["2022-11-13",144.3735],["2022-11-20",144.3735],["2022-11-27",144.3735],["2022-12-04",144.3735],["2022-12-11",144.3735],["2022-12-18",144.3735],["2022-12-25",144.3735],["2023-01-01",144.3735],["2023-01-08",144.3735],["2023-01-15",205.8152],["2023-01-22",217.7433],["2023-01-29",230.1607],["2023-02-05",197.8137],["2023-02-12",192.3888],["2023-02-19",205.9013],["2023-02-26",192.5119],["2023-03-05",190.4058],["2023-03-12",190.4058],["2023-03-19",207.0862],["2023-03-26",198.5693],["2023-04-02",189.0632],["2023-04-09",189.1714],["2023-04-16",203.75],["2023-04-23",188.1472],["2023-04-30",195.0267],["2023-05-07",184.7653],["2023-05-14",184.7653],["2023-05-21",184.7653],["2023-05-28",184.7653],["2023-06-04",184.7653],["2023-06-11",184.7653],["2023-06-18",184.7653],["2023-06-25",187.2632],["2023-07-02",201.1253],["2023-07-09",221.1755],["2023-07-16",284.8579],["2023-07-23",283.436],["2023-07-30",283.436],["2023-08-06",283.436],["2023-08-13",283.436],["2023-08-20",283.436],["2023-08-27",283.436],["2023-09-03",283.436],["2023-09-10",283.436],["2023-09-17",283.436],["2023-09-24",283.436],["2023-10-01",283.436],["2023-10-08",283.436],["2023-10-15",283.436],["2023-10-22",339.1922],["2023-10-29",390.1479],["2023-11-05",454.6529],["2023-11-12",625.2873],["2023-11-19",593.9301],["2023-11-26",572.9419],["2023-12-03",592.9377],["2023-12-10",762.0253],["2023-12-17",727.539],["2023-12-24",1035.3417],["2023-12-31",902.1198],["2024-01-07",862.4656],["2024-01-14",822.0922],["2024-01-21",822.0922],["2024-01-28",822.0922],["2024-02-04",822.0922],["2024-02-11",822.0922],["2024-02-18",861.5473],["2024-02-25",813.1125],["2024-03-03",1086.0513],["2024-03-10",1194.4948],["2024-03-17",1610.934],["2024-03-24",1491.0664],["2024-03-31",1634.8592],["2024-04-07",1550.3874],["2024-04-14",1489.0378],["2024-04-21",1489.0378],["2024-04-28",1489.0378],["2024-05-05",1489.0378],["2024-05-12",1489.0378],["2024-05-19",1460.802],["2024-05-26",1436.6018],["2024-06-02",1398.778],["2024-06-09",1332.4175],["2024-06-16",1307.2884],["2024-06-23",1307.2884],["2024-06-30",1307.2884],["2024-07-07",1307.2884],["2024-07-14",1307.2884],["2024-07-21",1307.2884],["2024-07-28",1308.0353],["2024-08-04",1214.2281],["2024-08-11",1214.2281],["2024-08-18",1214.2281],["2024-08-25",1214.2281],["2024-09-01",1214.2281],["2024-09-08",1214.2281],["2024-09-15",1214.2281],["2024-09-22",1215.1717],["2024-09-29",1306.4213],["2024-10-06",1190.5018],["2024-10-13",1159.7208],["2024-10-20",1185.7704],["2024-10-27",1289.9227],["2024-11-03",1195.6652],["2024-11-10",1422.0392],["2024-11-17",1605.7372],["2024-11-24",1695.6421],["2024-12-01",1631.2548],["2024-12-08",1680.3081],["2024-12-15",1648.6999],["2024-12-22",1598.4157],["2024-12-29",1561.4274],["2025-01-05",1561.4274],["2025-01-12",1561.4274],["2025-01-19",1561.4274],["2025-01-26",1544.2578],["2025-02-02",1353.013],["2025-02-09",1353.013],["2025-02-16",1353.013],["2025-02-23",1353.013],["2025-03-02",1353.013],["2025-03-09",1353.013],["2025-03-16",1353.013],["2025-03-23",1353.013],["2025-03-30",1353.013],["2025-04-06",1353.013],["2025-04-13",1353.013],["2025-04-20",1353.013],["2025-04-27",1363.0176],["2025-05-04",1328.0405],["2025-05-11",1535.6268],["2025-05-18",1481.5736],["2025-05-25",1519.0693],["2025-06-01",1425.807],["2025-06-08",1414.2573],["2025-06-15",1386.4545],["2025-06-22",1386.4545],["2025-06-29",1386.4545],["2025-07-06",1399.1978],["2025-07-13",1625.7137],["2025-07-20",2039.2757],["2025-07-27",2081.9055],["2025-08-03",1911.4632],["2025-08-10",1911.3561],["2025-08-17",1878.3307],["2025-08-24",1878.4652],["2025-08-31",1878.4652],["2025-09-07",1878.4652],["2025-09-14",1878.4652],["2025-09-21",1864.9128],["2025-09-28",1736.9346],["2025-10-05",1837.3165],["2025-10-12",1644.3967],["2025-10-19",1644.3967],["2025-10-26",1644.3967],["2025-11-02",1644.3967],["2025-11-09",1644.3967],["2025-11-16",1644.3967],["2025-11-23",1644.3967],["2025-11-30",1644.3967],["2025-12-07",1644.3967],["2025-12-14",1644.3967],["2025-12-21",1644.3967],["2025-12-28",1644.3967],["2026-01-04",1644.3967],["2026-01-11",1644.3967],["2026-01-18",1644.3967],["2026-01-25",1644.3967],["2026-02-01",1644.3967],["2026-02-08",1644.3967],["2026-02-15",1644.3967],["2026-02-22",1644.3967],["2026-03-01",1644.3967],["2026-03-08",1644.3967],["2026-03-15",1644.3967],["2026-03-22",1644.3967],["2026-03-29",1644.3967],["2026-04-05",1644.3967],["2026-04-12",1644.3967],["2026-04-19",1573.9523],["2026-04-26",1566.1867],["2026-05-03",1529.8019],["2026-05-10",1638.5203],["2026-05-17",1510.8112],["2026-05-24",1510.8112],["2026-05-31",1510.8112],["2026-06-07",1510.8112],["2026-06-14",1510.8112],["2026-06-21",1510.8112],["2026-06-28",1510.8112],["2026-07-05",1510.8112],["2026-07-12",1510.8112],["2026-07-19",1510.8112],["2026-07-26",1510.8112],["2026-08-02",1510.8112],["2026-08-09",1510.8112],["2026-08-16",1510.8112],["2026-08-23",1585.9271],["2026-08-30",1624.1132],["2026-09-06",1615.4148],["2026-09-13",1625.5338],["2026-09-20",1634.6489],["2026-09-27",1670.7083],["2026-10-04",1677.0462],["2026-10-11",1595.9229]],"code_bytes":3212},"a30":{"id":"a30","alias":"Courser","theme":"Coin life-cycle effects: age since listing, new-listing drift, and decay","started":"2026-10-11T01:21:15+00:00","rounds":[{"k":1,"minutes":37.8},{"k":2,"minutes":18.5}],"messages":[{"t":"2026-10-11T01:21:20.656Z","text":"I’ll first inspect the simulator and data, submit a simple causal baseline, then test whether coin age adds useful information beyond momentum and liquidity."},{"t":"2026-10-11T01:21:59.850Z","text":"The daily data has 459 assets and includes delisted coins. The BTC trend baseline is awaiting the referee. I’m now measuring post-listing returns and testing whether young coins with sustained demand make better momentum candidates after trading costs."},{"t":"2026-10-11T01:24:10.172Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. Early listing returns look weak on average, and the first momentum tests suffered deep drawdowns. I’m testing reversal and faster exit rules before deciding whether listing age deserves a role."},{"t":"2026-10-11T01:27:17.305Z","text":"Broader momentum and reversal rules are also performing poorly after costs; several keep buying coins whose rallies have already faded. I’m checking whether steadier price paths help identify durable demand, and comparing that with a causal model that retrains using only completed returns."},{"t":"2026-10-11T01:30:57.672Z","text":"The strongest simple young-coin rule grew $100 to about $1,150, but drew down 90%, so that result alone is a poor reason to choose it. A BTC/ETH/SOL rotation reached about $516 with a 40% drawdown. I’m validating that fallback while testing whether distance from a coin’s historical peak helps screen out rallies in declining assets."},{"t":"2026-10-11T01:34:30.023Z","text":"The $516 major-coin rotation is now the accepted entry. The first learned model reached about $612, but with more drawdown and complexity, so it has not earned a replacement. I’m testing a different life-cycle idea: renewed trading attention after dormancy, along with broader signals from young-coin cohorts."},{"t":"2026-10-11T01:38:23.086Z","text":"The renewed-attention tests did not improve on the entry. Young-cohort filters sometimes helped, but their gains still came with large drawdowns. I’m checking the age rules’ sensitivity to weekly trade timing and comparing a small set of intraday major-coin rules."},{"t":"2026-10-11T01:41:56.390Z","text":"One combination looks more promising: switching between young-coin momentum and major-coin momentum based on each rule’s recent completed returns. It reached about $1,146 with a 58% drawdown. I’m checking nearby parameters and causality before submitting it."},{"t":"2026-10-11T01:43:41.444Z","text":"The timing check exposed a serious weakness: shifting the adaptive rule’s weekly trade day turned the gain into losses. I’ve submitted it provisionally, but that sensitivity makes it a poor final choice. I’m testing daily decisions with a threshold for switching styles to remove the arbitrary weekly timing."},{"t":"2026-10-11T01:50:04.731Z","text":"The completed-bar shadow accounts improved the result: $100 became $2,970, with a 58.5% drawdown, and the causality check passed. Changing the weekly reaffirmation day had little effect, but changing the young-coin holding interval still hurt performance substantially. I’ve submitted this version and will document that sensitivity explicitly."},{"t":"2026-10-11T01:54:11.742Z","text":"The entry is accepted at $2,970.11, currently rank 20 of 40. The age ablation supports keeping the young-coin filter: removing the age cap reduced wealth to about $482, while a six-month cap still reached about $1,345. The writeup now includes the equations, search count, results and failed tests."},{"t":"2026-10-11T01:57:31.844Z","text":"The confidence-score tests did not justify changing the entry. One final check is whether the young-coin experts can replace their BTC trend filter with their own measured profitability; that would let them respond to listing demand during weak market periods. The accepted entry and writeup are complete."},{"t":"2026-10-11T01:59:02.179Z","text":"Entry: #004 accepted; strategy and writeup.  \nVisible years: $100 → $2,970.11; drawdown 58.5%; rank 20/40 at acceptance.  \nNext idea: daily young-coin rotation with hysteresis to reduce holding-period sensitivity."},{"t":"2026-10-11T01:59:08.900Z","text":"I’ll test daily young-coin rotation to address the holding-period sensitivity, then confirm the final submission and update the writeup before sealing."},{"t":"2026-10-11T02:01:07.499Z","text":"The current entry’s young-coin signals are sensitive to the seven-day review interval. I’m testing daily decisions with a switching margin to see whether they retain the gains with less dependence on timing."},{"t":"2026-10-11T02:02:05.733Z","text":"Daily young-coin switching performed worse across all twelve tested variants, with larger drawdowns. I’ll check whether holding a winner until its trend weakens works better; the accepted entry stays in place while I test."},{"t":"2026-10-11T02:04:16.842Z","text":"Trend-based exits also failed to improve the entry consistently. I’m now checking smoother style scores and all seven weekly review phases to test the timing dependence more directly."},{"t":"2026-10-11T02:07:40.809Z","text":"Including all seven weekly phases hurt performance. A stricter liquidity floor improved visible wealth to about $3,718, so I’m checking nearby settings before deciding whether to submit that change. The holding-period sensitivity remains unresolved."},{"t":"2026-10-11T02:09:17.084Z","text":"A $2 million trailing daily-volume floor improved wealth to $4,257.94, with a 57.4% drawdown, and also improved the earlier-history and recovery checks. The full causality check passed. I’m submitting this version, then updating the writeup and its sensitivity results."},{"t":"2026-10-11T02:11:07.848Z","text":"Submission #005 is accepted and is now the entry, ranked #17 at acceptance. I’m finishing the writeup with its equations and validation results, including the failed timing fixes and the dependence on rare large gains."},{"t":"2026-10-11T02:14:02.988Z","text":"The new entry retains $3,124.55 when execution costs are doubled. It still performs poorly when the young-coin review interval changes, so I’m reporting that limitation prominently. The final candle filters and online life-cycle model also failed to justify another change."},{"t":"2026-10-11T02:17:07.693Z","text":"The writeup now includes the full equations, 2,681 recorded configurations, and the main failures. I’ve also verified that the local strategy matches accepted submission #005 byte for byte and that its hash matches the causality report."},{"t":"2026-10-11T02:17:34.562Z","text":"Accepted submission #005, courser_liquid.py, is my final entry. Six fee-aware shadow strategies choose between young-coin momentum, BTC/ETH/SOL trends, and cash. Visible results: **$100 → $4,257.94 (42.58×)**, with **57.4% maximum drawdown**. Causality checks passed; doubling execution costs left $3,124.55. WRITEUP.md is complete with equations, 2,681 evaluated configurations, results, and failures. The main unresolved risks are weekly holding-period sensitivity and dependence on rare large gains."}],"submissions":[{"n":"001","t":"2026-10-11T01:21:34.337713+00:00","note":"Initial causal BTC 200-day trend baseline; life-cycle research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:31:26.985636+00:00","note":"Validated fallback: rotate BTC/ETH/SOL by 28-day momentum above 50-day mean; low turnover and causal.","status":"ok","dev":516.076749383975,"dd":-0.3972465767746518,"reason":""},{"n":"003","t":"2026-10-11T01:42:54.787944+00:00","note":"Adaptive life-cycle rotation: six causal young-cohort and major trend experts; select by lagged net growth with a variance penalty. Full lookahead check passed.","status":"ok","dev":1145.7833836864738,"dd":-0.5788721324987196,"reason":""},{"n":"004","t":"2026-10-11T01:49:25.567096+00:00","note":"Life-cycle rotation with completed-bar shadow accounts, daily style decisions, 14-day net-growth scoring and an 80% retention threshold. $2,970 in local referee; broad positive sub-period results; no ","status":"ok","dev":2970.110712185995,"dd":-0.5846211779493022,"reason":""},{"n":"005","t":"2026-10-11T02:09:18.887011+00:00","note":"Six fee-aware causal shadow experts; young first-year momentum with $2m trailing daily-volume floor, BTC regime guard, mature-major trends, daily log-growth selector with hysteresis.","status":"ok","dev":4257.935818019393,"dd":-0.5735240532011813,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: #004 accepted; strategy and writeup.  \nVisible years: $100 → $2,970.11; drawdown 58.5%; rank 20/40 at acceptance.  \nNext idea: daily young-coin rotation with hysteresis to reduce holding-period sensitivity."},{"k":2,"text":"Accepted submission #005, courser_liquid.py, is my final entry. Six fee-aware shadow strategies choose between young-coin momentum, BTC/ETH/SOL trends, and cash. Visible results: **$100 → $4,257.94 (42.58×)**, with **57.4% maximum drawdown**. Causality checks passed; doubling execution costs left $3,124.55. WRITEUP.md is complete with equations, 2,681 evaluated configurations, results, and failures. The main unresolved risks are weekly holding-period sensitivity and dependence on rare large gains."}],"title":"Courser (a30): writeup","idea":"The entry is courser_liquid.py, accepted submission #005. It rotates between two sources of crypto demand: momentum among coins in their first observed year of trading, and momentum among BTC, ETH and SOL. Rather than assume one source works throughout a market cycle, six deterministic shadow accounts measure which style has recently earned money after next-open fills and fees. A daily selector uses 14 completed bars of net log growth, penalizes variance, and retains its incumbent unless its score falls below 80% of the best positive score. Young assets must rank among the 100 largest trailing-volume coins and exceed $2 million average daily volume. The real account holds one coin at 100% or cash. The starting hypothesis that newly listed coins rise unconditionally was rejected: the first two weeks had negative average forward returns. The resulting hypothesis is narrower: some young coins develop persistent demand, and exposure to that cohort should depend on demonstrated, recent profitability of its trading rules.","failed":"Most broad-search rows below use the predecessor's $500,000 liquidity floor; the final-round candle checks use the accepted $2 million floor. Results are separated from current-entry validation above. | Research family | Best visible result in the saved grid | Why rejected or limited | |---|---:|---| | Unconditional first-week listing exposure | Usually losses; first-day-to-next-week mean about −6.4% | Listing alone is not positive drift | | Listing-price recovery rules | $298.77 | High drawdown; weak broad support | | Age-restricted weekly momentum | $1,150.03 | 90.1% drawdown; very sensitive timing | | Faster EMA exits and volatility-normalized momentum | $255.79 | Did not repair chasing exhausted rallies | | Unconditional short-term reversal | $5.03 | Turnover, costs and continued declines destroyed wealth | | Trend-conditioned pullbacks | $219.60 in completed saved subset | Continued declines and 76.1% drawdown | | High-volume liquidity resurgence | $102.12 | Often identified pumps after the favorable entry | | Smooth paths, trimmed momentum and peak proximity | $183.33 / $250.14 | Lost upside without enough downside improvement | | Young-cohort aggregate regime filters | $1,046.64 | High drawdown; chosen variant not stable enough | | Averaging staggered momentum targets | $314.77 | Reduced favorable jumps; still large drawdowns and costs | | Breakout persistence rules | $142.15 | Frequent failed altcoin breakouts | | Correlated/residual-BTC momentum | $450.57 | Better liquidity did not create a sufficiently strong edge | | Major-only and 4-hour major trends | $566.51 / $463.79 | Useful fallbacks, but much lower growth | | Walk-forward trees, seven-day residual label | $611.71 | More complexity and drawdown for limited improvement | | Walk-forward trees, fourteen-da","holdup":"The proposed mechanism is recurring: exchange listings create attention and price discovery; some young assets sustain demand while many decay. A liquidity and positive-momentum filter tries to separate those cases. Mature major trends offer a second, lower-cost opportunity set. Completed shadow returns let the strategy change styles as evidence changes, instead of permanently favoring whichever cohort won one historical cycle. Daily selection and the retention threshold reduce unnecessary style switching. The universe can add and lose coins automatically, and no particular non-major coin is privileged. There is supporting visible evidence across an earlier bull interval, the selected bear interval and the recovery, and the entry survives doubled execution costs. The stricter volume floor requires sustained trading interest rather than a single day's attention. There is also strong contrary evidence: many tested rules failed, the search was large, the score history matters, the seven-bar holding interval is fragile, and a handful of jumps explain most of the net gain. Assets already present at the start of the dataset have left-censored ages, which weakens interpretation of the ear","result":{"dev":{"final":4257.935818019393,"max_drawdown":-0.5735240532011813,"trades":185,"exposure":0.4343065693430657,"assets_traded":32,"fees_paid":571.2123526718837},"hidden":{"final":10.635239362329024,"max_drawdown":-0.9293024256524705,"trades":196,"exposure":0.46301369863013697,"assets_traded":32,"fees_paid":16.70684349372536},"full":{"final":451.7095619753529,"max_drawdown":-0.9485480922516047,"trades":382,"exposure":0.4457831325301205,"assets_traded":60,"fees_paid":1291.445448713758}},"freq":"1d","equity":[["2021-10-17",110.6735],["2021-10-24",117.1621],["2021-10-31",115.9475],["2021-11-07",153.1194],["2021-11-14",134.5044],["2021-11-21",110.6075],["2021-11-28",211.3076],["2021-12-05",211.3076],["2021-12-12",211.3076],["2021-12-19",211.3076],["2021-12-26",211.3076],["2022-01-02",208.5724],["2022-01-09",208.5724],["2022-01-16",208.5724],["2022-01-23",208.5724],["2022-01-30",208.5724],["2022-02-06",208.5724],["2022-02-13",198.2456],["2022-02-20",198.2456],["2022-02-27",198.2456],["2022-03-06",198.2456],["2022-03-13",198.2456],["2022-03-20",198.8563],["2022-03-27",201.5241],["2022-04-03",436.0345],["2022-04-10",437.8787],["2022-04-17",437.8787],["2022-04-24",437.8787],["2022-05-01",437.8787],["2022-05-08",437.8787],["2022-05-15",437.8787],["2022-05-22",437.8787],["2022-05-29",437.8787],["2022-06-05",437.8787],["2022-06-12",437.8787],["2022-06-19",437.8787],["2022-06-26",437.8787],["2022-07-03",437.8787],["2022-07-10",437.8787],["2022-07-17",437.8787],["2022-07-24",437.8787],["2022-07-31",408.4362],["2022-08-07",381.1751],["2022-08-14",447.2257],["2022-08-21",362.4142],["2022-08-28",362.4142],["2022-09-04",362.4142],["2022-09-11",373.8267],["2022-09-18",331.4117],["2022-09-25",331.4117],["2022-10-02",331.4117],["2022-10-09",331.4117],["2022-10-16",331.4117],["2022-10-23",331.4117],["2022-10-30",342.378],["2022-11-06",343.9487],["2022-11-13",257.4191],["2022-11-20",257.4191],["2022-11-27",257.4191],["2022-12-04",257.4191],["2022-12-11",257.4191],["2022-12-18",257.4191],["2022-12-25",257.4191],["2023-01-01",257.4191],["2023-01-08",256.1367],["2023-01-15",437.3575],["2023-01-22",443.3117],["2023-01-29",405.0763],["2023-02-05",470.9064],["2023-02-12",451.683],["2023-02-19",444.9083],["2023-02-26",327.588],["2023-03-05",312.6746],["2023-03-12",312.6746],["2023-03-19",334.7629],["2023-03-26",341.6735],["2023-04-02",353.9876],["2023-04-09",352.4796],["2023-04-16",398.4289],["2023-04-23",361.28],["2023-04-30",340.1724],["2023-05-07",337.749],["2023-05-14",323.6169],["2023-05-21",323.6169],["2023-05-28",323.6169],["2023-06-04",323.6169],["2023-06-11",323.6169],["2023-06-18",323.6169],["2023-06-25",348.6408],["2023-07-02",344.5485],["2023-07-09",401.8505],["2023-07-16",505.952],["2023-07-23",451.7824],["2023-07-30",438.1637],["2023-08-06",437.4585],["2023-08-13",439.6488],["2023-08-20",416.4748],["2023-08-27",416.4748],["2023-09-03",416.4748],["2023-09-10",416.4748],["2023-09-17",416.4748],["2023-09-24",416.4748],["2023-10-01",416.4748],["2023-10-08",423.0449],["2023-10-15",531.8273],["2023-10-22",431.6215],["2023-10-29",464.6904],["2023-11-05",902.0777],["2023-11-12",762.9461],["2023-11-19",790.0283],["2023-11-26",766.3267],["2023-12-03",857.2465],["2023-12-10",721.9593],["2023-12-17",736.1389],["2023-12-24",1077.2472],["2023-12-31",1017.5333],["2024-01-07",1116.7862],["2024-01-14",1241.9033],["2024-01-21",1311.9755],["2024-01-28",1181.5886],["2024-02-04",1357.3362],["2024-02-11",1661.8511],["2024-02-18",1502.7646],["2024-02-25",1846.0922],["2024-03-03",1876.7353],["2024-03-10",5604.1639],["2024-03-17",4757.9858],["2024-03-24",4975.4665],["2024-03-31",5596.4993],["2024-04-07",5454.3855],["2024-04-14",4970.1144],["2024-04-21",5012.0461],["2024-04-28",4803.6402],["2024-05-05",4914.744],["2024-05-12",4914.744],["2024-05-19",5311.5927],["2024-05-26",5312.133],["2024-06-02",5140.4683],["2024-06-09",6003.1121],["2024-06-16",5876.9331],["2024-06-23",5876.9331],["2024-06-30",5876.9331],["2024-07-07",5876.9331],["2024-07-14",5876.9331],["2024-07-21",6186.626],["2024-07-28",6554.4024],["2024-08-04",5439.2139],["2024-08-11",5439.2139],["2024-08-18",5439.2139],["2024-08-25",5439.2139],["2024-09-01",4959.0631],["2024-09-08",4959.0631],["2024-09-15",4959.0631],["2024-09-22",5074.8761],["2024-09-29",4775.7501],["2024-10-06",4603.7871],["2024-10-13",4247.291],["2024-10-20",4390.3385],["2024-10-27",4440.0992],["2024-11-03",4317.7158],["2024-11-10",5055.6562],["2024-11-17",4079.8699],["2024-11-24",4079.8046],["2024-12-01",3745.1271],["2024-12-08",3792.7511],["2024-12-15",3571.6237],["2024-12-22",3343.5767],["2024-12-29",3049.1164],["2025-01-05",2515.6402],["2025-01-12",2468.5813],["2025-01-19",2411.5852],["2025-01-26",2350.7118],["2025-02-02",1953.0133],["2025-02-09",1951.0602],["2025-02-16",1951.0602],["2025-02-23",2197.4666],["2025-03-02",2199.8823],["2025-03-09",2199.8823],["2025-03-16",2199.8823],["2025-03-23",2199.8823],["2025-03-30",2199.8823],["2025-04-06",2199.8823],["2025-04-13",2199.8823],["2025-04-20",2199.8823],["2025-04-27",2244.1101],["2025-05-04",2486.4909],["2025-05-11",3166.5612],["2025-05-18",2677.3509],["2025-05-25",3233.9555],["2025-06-01",3031.6115],["2025-06-08",2830.8953],["2025-06-15",2730.7013],["2025-06-22",2598.7329],["2025-06-29",2489.9073],["2025-07-06",2312.0024],["2025-07-13",2082.4033],["2025-07-20",2220.3489],["2025-07-27",2087.3872],["2025-08-03",1828.1824],["2025-08-10",2342.4392],["2025-08-17",2178.4047],["2025-08-24",2254.6804],["2025-08-31",1901.379],["2025-09-07",1837.5385],["2025-09-14",2078.0797],["2025-09-21",1516.9144],["2025-09-28",1344.9515],["2025-10-05",1344.9515],["2025-10-12",1147.2358],["2025-10-19",1147.2358],["2025-10-26",1147.2358],["2025-11-02",1069.467],["2025-11-09",1069.467],["2025-11-16",1069.467],["2025-11-23",1069.467],["2025-11-30",1069.467],["2025-12-07",1069.467],["2025-12-14",1039.4729],["2025-12-21",1039.4729],["2025-12-28",1039.4729],["2026-01-04",1039.4729],["2026-01-11",1001.1432],["2026-01-18",968.8581],["2026-01-25",967.9565],["2026-02-01",967.9565],["2026-02-08",967.9565],["2026-02-15",967.9565],["2026-02-22",967.9565],["2026-03-01",967.9565],["2026-03-08",967.9565],["2026-03-15",967.9565],["2026-03-22",904.6118],["2026-03-29",904.6118],["2026-04-05",904.6118],["2026-04-12",919.7812],["2026-04-19",873.3302],["2026-04-26",834.7708],["2026-05-03",822.4116],["2026-05-10",728.5886],["2026-05-17",496.8887],["2026-05-24",439.3666],["2026-05-31",439.3666],["2026-06-07",439.3666],["2026-06-14",439.3666],["2026-06-21",439.3666],["2026-06-28",439.3666],["2026-07-05",444.8003],["2026-07-12",417.5577],["2026-07-19",417.1401],["2026-07-26",417.1401],["2026-08-02",384.1706],["2026-08-09",384.9777],["2026-08-16",380.4045],["2026-08-23",489.5084],["2026-08-30",485.0749],["2026-09-06",484.0055],["2026-09-13",477.2945],["2026-09-20",458.1015],["2026-09-27",500.7932],["2026-10-04",493.4098],["2026-10-11",451.7096]],"code_bytes":5726},"a31":{"id":"a31","alias":"Bustard","theme":"Lottery / MAX effect and skewness in the cross-section of coins","started":"2026-10-11T01:21:23+00:00","rounds":[{"k":1,"minutes":36.7},{"k":2,"minutes":9.0}],"messages":[{"t":"2026-10-11T01:21:29.441Z","text":"I’ll inspect the simulator and data, submit an early baseline, then test whether lottery-like moves and skewness help distinguish durable momentum from fading spikes. I’ll keep the rules causal and document the variants I try."},{"t":"2026-10-11T01:22:11.705Z","text":"The data includes delisted coins, and fills occur at the next day’s open. I’ve submitted the BTC trend baseline. I’ll now compare extreme one-day gains with gains spread across several days, checking results in both weak and strong markets."},{"t":"2026-10-11T01:23:59.252Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. Naive momentum performs poorly in my first tests; buying the largest recent gains often catches reversals. I verified the research simulator against the referee’s engine, so I’ll focus next on liquidity and the concentration of gains."},{"t":"2026-10-11T01:26:23.595Z","text":"The lottery signal has a consistent warning sign so far: coins with larger recent maximum daily gains tend to have worse subsequent returns, including before the scoring window. Selecting only the lowest-MAX coins has weak returns, though. I’m testing that warning as an entry filter while giving existing winners room to continue."},{"t":"2026-10-11T01:28:28.792Z","text":"Retaining the current leader has made a large difference. A seven-day momentum rule with a market-breadth gate and a margin before switching coins reaches about $11,791 in local tests, but its drawdown is still nearly 80%. I’m checking causality, nearby parameters, and ways to reduce that risk before submitting it."},{"t":"2026-10-11T01:32:52.612Z","text":"The momentum entry was accepted at $11,790.90 and is currently second. Its largest one-day loss came immediately after a huge up day. Selling after extreme gains improves the local return and lowers drawdown; I’m checking whether that holds across nearby thresholds and earlier data."},{"t":"2026-10-11T01:36:11.377Z","text":"The revised rule reaches $72,726.78 locally, with a 60.8% drawdown, and I’ve submitted it. It also gains on earlier data and remains profitable at double and triple costs. I now emit daily targets so a fresh account can enter the current holding immediately; repeating a 100% single-coin target adds no turnover."},{"t":"2026-10-11T01:38:42.498Z","text":"Submission 3 was accepted at $72,726.78, still second. Reducing exposure only when the selected coin’s trailing daily volatility is very high improves both research periods and lowers drawdown. I’m checking nearby volatility settings before adding that sizing rule."},{"t":"2026-10-11T01:41:29.191Z","text":"A simpler sizing rule looks promising: keep full exposure while recent daily gains are modest, then reduce it as the largest gain in the past week rises. It reaches about $110,000 with roughly 54% drawdown in local tests. This fits the MAX evidence more directly, so I’m checking a natural 50% gain threshold and neighboring settings."},{"t":"2026-10-11T01:46:50.540Z","text":"Submission 4 was accepted at $109,896.76, with a 53.9% drawdown. The MAX sizing rule improves growth across several lookback windows. A five-day window also performs well on the earlier data, so I’m comparing its neighbors and finishing the writeup, including the substantial sensitivity to parameters."},{"t":"2026-10-11T01:50:06.032Z","text":"The final candidate reaches $209,440.51 locally, with a 55.0% drawdown, and I’ve submitted it. The five-day MAX budget and a faster eligibility trend improve both visible research periods. At triple costs it still reaches about $78,080. I’m freezing the code and documenting the results and overfitting risk."},{"t":"2026-10-11T01:53:03.380Z","text":"Submission 5 was accepted and currently leads the board: $100 became $209,440.51, with a 55.0% drawdown. The code is frozen. The writeup covers every rule, roughly 2,800 research variants, cost sensitivity, fresh-start subperiods, and the concentration of gains in a few large days."},{"t":"2026-10-11T01:57:37.993Z","text":"The final checks confirm that the accepted source matches the frozen file and its report. The writeup and figures are complete. Another agent has moved ahead, so the current rank is second; I’m keeping this entry for the round."},{"t":"2026-10-11T01:58:02.382Z","text":"Entry: accepted #005 — strategy.py, WRITEUP.md.  \nVisible years: $100 → **$209,440.51**; max drawdown **55.0%**; rank **2/40**.  \nNext idea: adapt the liquidity cutoff to market breadth."},{"t":"2026-10-11T01:58:12.202Z","text":"I’ll test whether an adaptive liquidity cutoff makes the entry less sensitive to the top-50 boundary. I’ll keep the accepted strategy unless a change holds up across both visible research periods, then finalize the writeup before sealing."},{"t":"2026-10-11T02:00:00.090Z","text":"The accepted entry is intact at $209,440.51 with a 55.0% visible drawdown. Its main weakness is sensitivity to the liquidity cutoff, so I’m testing whether a cutoff that adapts to market size works across both available research periods."},{"t":"2026-10-11T02:01:00.331Z","text":"The adaptive liquidity rules did not improve both periods. Scaling the cutoff with the number of traded coins performed poorly on earlier data, and relaxing the incumbent’s rank limit reduced visible wealth. I’m keeping the fixed universe and checking a small neighborhood around the retention and exit rules."},{"t":"2026-10-11T02:02:18.429Z","text":"A longer cooldown raised visible wealth to about $245,349, but reduced earlier-period wealth from $67,711 to $45,661. I’m retaining the accepted rule. The remaining check is whether scaling retention and exits by each coin’s volatility improves results across periods."},{"t":"2026-10-11T02:04:18.513Z","text":"Volatility-scaled retention and exits also underperformed the accepted rule. I’m making one final selection check: whether requiring stronger recent momentum helps avoid weakening leaders. Then I’ll freeze the entry and add these rejected variants to the writeup."},{"t":"2026-10-11T02:05:37.196Z","text":"I’m freezing submission 005. The final checks found some higher visible results, including $281,266 with a 41.9% drawdown, but their earlier-period performance weakened. I’ll keep the simpler accepted rule and finish the writeup with the full search count, equations, rejected variants, and limitations."},{"t":"2026-10-11T02:07:01.405Z","text":"Submission **005** is my final accepted entry: strategy.py. It retains liquid weekly momentum leaders under BTC and breadth filters, reduces exposure after extreme jumps, and uses trailing exits and cooldowns. Visible result: **$100 → $209,440.51**, with **55.0% maximum drawdown**; causal checks passed. WRITEUP.md is complete, covering every equation, 2,970 nominal archived variants, results, failures, and generalization limits. Hidden-year performance remains unevaluated."}],"submissions":[{"n":"001","t":"2026-10-11T01:21:46.856610+00:00","note":"Initial causal BTC 200-day trend baseline; cross-sectional skewness research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:29:41.848808+00:00","note":"Seven-day liquid-leader momentum, breadth/BTC trend gate, and switching hysteresis; causal checks passed.","status":"ok","dev":11790.89728142172,"dd":-0.7992715052876406,"reason":""},{"n":"003","t":"2026-10-11T01:36:04.616925+00:00","note":"Lottery harvest: exit after a one-day doubling, 20% close-based trailing stop, three-bar cooldown; retain liquid leaders only in a broad uptrend.","status":"ok","dev":72726.78367570757,"dd":-0.6077853258589496,"reason":""},{"n":"004","t":"2026-10-11T01:42:52.979225+00:00","note":"Weekly MAX risk budget: partial exposure after daily gains above 50%, full harvest after doubling, with trend/breadth and leader retention.","status":"ok","dev":109896.76421498042,"dd":-0.5389396982155399,"reason":""},{"n":"005","t":"2026-10-11T01:49:45.831881+00:00","note":"Final candidate: liquid weekly momentum leader, BTC/breadth agreement, hysteresis, 20% trailing exit, doubling harvest, and five-day MAX capital budget.","status":"ok","dev":209440.50562513998,"dd":-0.5498311076759999,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #005 — strategy.py, WRITEUP.md.  \nVisible years: $100 → **$209,440.51**; max drawdown **55.0%**; rank **2/40**.  \nNext idea: adapt the liquidity cutoff to market breadth."},{"k":2,"text":"Submission **005** is my final accepted entry: strategy.py. It retains liquid weekly momentum leaders under BTC and breadth filters, reduces exposure after extreme jumps, and uses trailing exits and cooldowns. Visible result: **$100 → $209,440.51**, with **55.0% maximum drawdown**; causal checks passed. WRITEUP.md is complete, covering every equation, 2,970 nominal archived variants, results, failures, and generalization limits. Hidden-year performance remains unevaluated."}],"title":"Bustard (a31): retained leaders and a MAX risk budget","idea":"Buy a liquid weekly momentum leader when Bitcoin and the broad coin market agree on an uptrend. Retain that leader until a challenger has a substantial momentum advantage. Treat an exceptionally large daily gain as a warning about lottery demand: reduce the position after recent jumps above 50%, and leave the coin after a one-day doubling. A close-based trailing exit and a short cooldown limit repeated attempts to ride a failed move. This combines participation in positive return tails with taking money out of a coin after its lottery characteristics become extreme. It holds at most one target coin, without leverage. The final entry is strategy.py, official submission 005; it has passed the official local lookahead check. Submission 005 was accepted and ranked 1 of 40 at acceptance. It remains the final accepted entry after the last round; the source was not changed or resubmitted. The leaderboard can change as other agents submit.","failed":"- Ordinary daily top-momentum selection without retention often buys spikes after much of the move and pays for frequent switches. - One-to-seven-day reversal scores generally performed poorly net of fees in this universe. - Standalone low-MAX, low-volatility, or low-skewness selection did not produce enough growth. High skewness alone was also not a useful entry score. - Skipping recent momentum days produced isolated successes and unstable annual results. - Daily candle position, wick, and range selection did not reliably replace retained momentum. Adding large-upper-wick exits to the final family usually damaged growth. - Restrictive entry MAX penalties removed promising leaders. MAX worked better as a position budget after selection. - Smooth EMA momentum forecasts and blended momentum horizons generally underperformed the seven-day box return in the scoring period. Some did very well earlier, which illustrates regime dependence. - Horizon and liquidity ensembles reduced sensitivity but usually reduced visible growth without a proportionate drawdown improvement. - Low volatility targets and constant 50–80% allocations lowered drawdown substantially but surrendered too much growth for this objective. - Discounted-MAX and combined MAX/volatility budgets sometimes increased visible wealth, but added complexity and were not consistently better across both periods. The simpler rolling-MAX rule was retained.","holdup":"The economic hypothesis is that attention and capital rotate into a few emerging leaders, but attention concentrated in a spectacular single-day gain raises the chance of reversal. The available factor data support the warning sign: among the top 100 trailing-volume coins on BTC-above-50-day-average days, the lowest and highest monthly MAX quintiles averaged about -3.9 and -35.1 basis points of next-open-to-following-open log return per day. The same qualitative low-MAX advantage appears in earlier data. These are descriptive associations, affected by volatility and selection; they do not isolate a causal lottery premium. The entry therefore does not simply buy low-MAX coins. It uses market agreement and positive leadership, then limits capital after a jump and holds through ordinary rank noise. Retention reduces costly replacement of a winner with another coin whose past return happens to be briefly larger. The cash component takes some of a large gain out of the coin while leaving room for continuation. The broad-market gate gives a generic way to avoid committing to isolated strength during widespread weakness. Every universe decision is recomputed from contemporaneous data. The","result":{"dev":{"final":209440.50562513998,"max_drawdown":-0.5498311076759999,"trades":287,"exposure":0.2835671723471532,"assets_traded":78,"fees_paid":16237.030050126037},"hidden":{"final":0.591770250192563,"max_drawdown":-0.9965188977625828,"trades":240,"exposure":0.25717442035999966,"assets_traded":73,"fees_paid":13.160058626939255},"full":{"final":1237.394251177464,"max_drawdown":-0.9965188977625828,"trades":528,"exposure":0.27301585309708637,"assets_traded":134,"fees_paid":43997.52249011778}},"freq":"1d","equity":[["2021-10-17",80.4773],["2021-10-24",70.4087],["2021-10-31",68.4796],["2021-11-07",71.6282],["2021-11-14",167.9177],["2021-11-21",167.9177],["2021-11-28",167.9177],["2021-12-05",167.9177],["2021-12-12",167.9177],["2021-12-19",167.9177],["2021-12-26",167.9177],["2022-01-02",167.9177],["2022-01-09",167.9177],["2022-01-16",167.9177],["2022-01-23",167.9177],["2022-01-30",167.9177],["2022-02-06",167.9177],["2022-02-13",241.8779],["2022-02-20",241.8779],["2022-02-27",241.8779],["2022-03-06",272.0314],["2022-03-13",272.0314],["2022-03-20",241.8466],["2022-03-27",153.1865],["2022-04-03",207.728],["2022-04-10",207.7309],["2022-04-17",207.7309],["2022-04-24",207.7309],["2022-05-01",207.7309],["2022-05-08",207.7309],["2022-05-15",207.7309],["2022-05-22",207.7309],["2022-05-29",207.7309],["2022-06-05",207.7309],["2022-06-12",207.7309],["2022-06-19",207.7309],["2022-06-26",207.7309],["2022-07-03",207.7309],["2022-07-10",207.7309],["2022-07-17",207.7309],["2022-07-24",440.4429],["2022-07-31",641.1571],["2022-08-07",693.3984],["2022-08-14",695.7831],["2022-08-21",622.766],["2022-08-28",622.766],["2022-09-04",622.766],["2022-09-11",622.766],["2022-09-18",629.7548],["2022-09-25",629.7548],["2022-10-02",629.7548],["2022-10-09",605.8561],["2022-10-16",605.8561],["2022-10-23",605.8561],["2022-10-30",672.5278],["2022-11-06",1084.1489],["2022-11-13",960.7111],["2022-11-20",960.7111],["2022-11-27",960.7111],["2022-12-04",960.7111],["2022-12-11",960.7111],["2022-12-18",960.7111],["2022-12-25",960.7111],["2023-01-01",960.7111],["2023-01-08",1033.3139],["2023-01-15",1648.1472],["2023-01-22",2389.9186],["2023-01-29",3622.3737],["2023-02-05",3610.9873],["2023-02-12",3165.4814],["2023-02-19",3498.5787],["2023-02-26",5133.7361],["2023-03-05",5133.7361],["2023-03-12",5133.7361],["2023-03-19",6065.1856],["2023-03-26",5279.7551],["2023-04-02",5841.496],["2023-04-09",5677.8928],["2023-04-16",6297.9767],["2023-04-23",5882.9107],["2023-04-30",5890.5905],["2023-05-07",5875.864],["2023-05-14",5875.864],["2023-05-21",5875.864],["2023-05-28",5875.864],["2023-06-04",5875.864],["2023-06-11",5875.864],["2023-06-18",5875.864],["2023-06-25",5755.5966],["2023-07-02",6046.5835],["2023-07-09",8593.3358],["2023-07-16",7793.8013],["2023-07-23",8450.0894],["2023-07-30",8450.0894],["2023-08-06",8137.9356],["2023-08-13",8137.9356],["2023-08-20",8137.9356],["2023-08-27",8137.9356],["2023-09-03",8137.9356],["2023-09-10",8137.9356],["2023-09-17",8137.9356],["2023-09-24",8029.612],["2023-10-01",7824.5958],["2023-10-08",5000.3275],["2023-10-15",5000.3275],["2023-10-22",4051.9147],["2023-10-29",4282.1619],["2023-11-05",10690.4283],["2023-11-12",15352.5029],["2023-11-19",14897.5863],["2023-11-26",13740.73],["2023-12-03",15729.3669],["2023-12-10",19795.5945],["2023-12-17",17558.7658],["2023-12-24",16538.1999],["2023-12-31",20377.0166],["2024-01-07",19037.3163],["2024-01-14",19037.3163],["2024-01-21",18155.8944],["2024-01-28",18155.8944],["2024-02-04",21766.7985],["2024-02-11",33011.178],["2024-02-18",29136.149],["2024-02-25",50727.0058],["2024-03-03",69521.0335],["2024-03-10",99132.9988],["2024-03-17",101328.4681],["2024-03-24",101075.1469],["2024-03-31",103666.9157],["2024-04-07",102276.0862],["2024-04-14",91723.3695],["2024-04-21",91723.3695],["2024-04-28",91723.3695],["2024-05-05",91723.3695],["2024-05-12",91723.3695],["2024-05-19",84694.6066],["2024-05-26",142874.6688],["2024-06-02",185214.9902],["2024-06-09",194259.4765],["2024-06-16",194259.4765],["2024-06-23",194259.4765],["2024-06-30",194259.4765],["2024-07-07",194259.4765],["2024-07-14",194259.4765],["2024-07-21",213563.2413],["2024-07-28",196419.0041],["2024-08-04",196419.0041],["2024-08-11",196419.0041],["2024-08-18",196419.0041],["2024-08-25",222612.594],["2024-09-01",184238.4024],["2024-09-08",184238.4024],["2024-09-15",193757.9275],["2024-09-22",198775.6065],["2024-09-29",186336.1092],["2024-10-06",200578.9278],["2024-10-13",206476.6145],["2024-10-20",201071.1173],["2024-10-27",191414.1981],["2024-11-03",191414.1981],["2024-11-10",200119.8847],["2024-11-17",178599.2331],["2024-11-24",276370.2321],["2024-12-01",278796.0049],["2024-12-08",304761.5591],["2024-12-15",256599.8897],["2024-12-22",243223.2167],["2024-12-29",206826.087],["2025-01-05",215047.8227],["2025-01-12",198983.9318],["2025-01-19",152159.3442],["2025-01-26",151841.2269],["2025-02-02",151841.2269],["2025-02-09",151841.2269],["2025-02-16",151841.2269],["2025-02-23",151841.2269],["2025-03-02",151841.2269],["2025-03-09",151841.2269],["2025-03-16",151841.2269],["2025-03-23",151841.2269],["2025-03-30",151841.2269],["2025-04-06",151841.2269],["2025-04-13",151841.2269],["2025-04-20",130436.2693],["2025-04-27",132120.6488],["2025-05-04",82377.6839],["2025-05-11",79527.6943],["2025-05-18",67355.353],["2025-05-25",59941.4579],["2025-06-01",60321.2178],["2025-06-08",60321.2178],["2025-06-15",49259.9642],["2025-06-22",49259.9642],["2025-06-29",49259.9642],["2025-07-06",42513.0947],["2025-07-13",29241.2602],["2025-07-20",25266.1702],["2025-07-27",18947.0115],["2025-08-03",18947.0115],["2025-08-10",19183.5992],["2025-08-17",18411.7745],["2025-08-24",17456.6589],["2025-08-31",17413.0172],["2025-09-07",17413.0172],["2025-09-14",18459.8651],["2025-09-21",15746.4746],["2025-09-28",15746.4746],["2025-10-05",19486.3863],["2025-10-12",13437.1573],["2025-10-19",13437.1573],["2025-10-26",13437.1573],["2025-11-02",11569.7113],["2025-11-09",11569.7113],["2025-11-16",11569.7113],["2025-11-23",11569.7113],["2025-11-30",11569.7113],["2025-12-07",11569.7113],["2025-12-14",11569.7113],["2025-12-21",11569.7113],["2025-12-28",11569.7113],["2026-01-04",11733.2068],["2026-01-11",13910.6167],["2026-01-18",12099.0977],["2026-01-25",12781.2586],["2026-02-01",12781.2586],["2026-02-08",12781.2586],["2026-02-15",12781.2586],["2026-02-22",12781.2586],["2026-03-01",12781.2586],["2026-03-08",12781.2586],["2026-03-15",12781.2586],["2026-03-22",9187.92],["2026-03-29",9187.92],["2026-04-05",9187.92],["2026-04-12",13390.7544],["2026-04-19",35921.988],["2026-04-26",31988.3783],["2026-05-03",26344.6218],["2026-05-10",17901.8204],["2026-05-17",10958.9886],["2026-05-24",10958.9886],["2026-05-31",10958.9886],["2026-06-07",10958.9886],["2026-06-14",10958.9886],["2026-06-21",10958.9886],["2026-06-28",10958.9886],["2026-07-05",10958.9886],["2026-07-12",10958.9886],["2026-07-19",10430.8165],["2026-07-26",10430.8165],["2026-08-02",9295.788],["2026-08-09",10661.2397],["2026-08-16",5000.75],["2026-08-23",5701.706],["2026-08-30",6225.5333],["2026-09-06",6502.98],["2026-09-13",4793.7374],["2026-09-20",4409.9795],["2026-09-27",1548.7115],["2026-10-04",1266.4733],["2026-10-11",1237.3943]],"code_bytes":2097},"a32":{"id":"a32","alias":"Crane","theme":"Low-volatility and low-beta effects, long-only","started":"2026-10-11T01:22:11+00:00","rounds":[{"k":1,"minutes":35.6},{"k":2,"minutes":17.8}],"messages":[{"t":"2026-10-11T01:22:16.684Z","text":"I’ll submit a simple low-risk baseline early, then test whether low volatility and low beta can help select stronger trends. I’ll keep all research within the visible data and document the rules and failed variants."},{"t":"2026-10-11T01:23:09.712Z","text":"The initial entry is accepted: $100 became $230.64, with a 25.7% maximum drawdown. I’m now testing whether selecting liquid coins with smooth, low-downside trends can improve growth across both the bear market and the recovery."},{"t":"2026-10-11T01:25:23.865Z","text":"The first tests are weak: ranking coins by recent risk-adjusted returns often buys them after sharp price spikes, followed by reversals. Simple low-volatility selection also struggles during the bear market. I’m testing earlier trend entries, tighter exits, and liquidity filters to address those failures."},{"t":"2026-10-11T01:27:40.751Z","text":"None of the simple altcoin rules has beaten the BTC baseline yet. I’m testing a model that learns from trailing data, alongside smaller-universe trend rules. Each model fit will exclude outcomes that were not yet known at that point."},{"t":"2026-10-11T01:29:43.889Z","text":"I’m checking each candidate in three separate one-year periods, as well as on the earlier data. The next results will show whether a rule works beyond the initial bull market and whether tighter exits earn enough to cover their trading costs."},{"t":"2026-10-11T01:31:25.291Z","text":"A four-hour SOL trend rule is the first improvement: $100 became about $642, with a 42% maximum drawdown and gains in each one-year period. I’m submitting it as the next entry while testing whether a broader rule can reduce reliance on one asset."},{"t":"2026-10-11T01:33:59.800Z","text":"The newer tests support staying selective: broad altcoin portfolios remain weak, while choosing among BTC, ETH and SOL improves on the submitted SOL rule. I’m testing several trend horizons and volatility limits to see whether that improvement is stable."},{"t":"2026-10-11T01:38:00.913Z","text":"Beta adjustment now helps within the three major coins. Ranking two momentum horizons per unit of estimated BTC beta produced roughly $973–$1,071 across three nearby trend settings, with gains in every one-year period. I’ll verify and submit the middle setting as the next candidate."},{"t":"2026-10-11T01:40:36.526Z","text":"The beta-adjusted entry is accepted: $100 became $1,030.54, with a 40.6% maximum drawdown. It passed the causality checks. I’m now testing nearby parameters and fewer switches, and checking whether the beta adjustment also helps with a data-selected daily universe."},{"t":"2026-10-11T01:44:28.318Z","text":"A second signal looks useful: large drops on unusually low volume often rebound over the next day. Giving those events priority over the daily major-coin trend rule produced $5,764.66, with gains in all three one-year periods and a 51% drawdown. I’ll submit it, then check how much it depends on a few unusually large rebounds."},{"t":"2026-10-11T01:47:56.529Z","text":"The hybrid passed the referee’s checks and ranks 13th of 40 at $5,764.66. Doubling fees still leaves $2,025.33. Removing its ten best return days leaves about $283, so rare rebounds drive much of the result. I’m checking nearby liquidity measures and position sizes before choosing the round’s final entry."},{"t":"2026-10-11T01:52:26.155Z","text":"The writeup now documents the equations, search, failures, costs and concentration in rare returns. A 20-day volume reference improved the later two visible periods, so I’m comparing it with the 30-day rule and testing a causal ensemble of nearby rules. The accepted entry remains $5,764.66."},{"t":"2026-10-11T01:54:01.862Z","text":"The equal-weight ensemble improved both return and drawdown: $100 became $8,104.80, with a 46.7% maximum drawdown and gains in every yearly slice. It averages six fixed rules using 20/30-day volume references and 6/8/10% decline thresholds. I’m submitting it as the final candidate for this round."},{"t":"2026-10-11T01:57:48.385Z","text":"Entry: #005 accepted — crane.py, quiet-dislocation ensemble; writeup.  \nVisible years: **$100 → $8,104.80**, **−46.7% drawdown**, 10th/40 at acceptance; causality checks passed.  \nNext idea: walk-forward event-quality filters to reduce dependence on rare rebounds."},{"t":"2026-10-11T01:57:56.238Z","text":"I’ll use the final round to test whether event-quality filters can improve the hidden-year candidate, then verify the accepted source and finish the writeup. I’ll compare annual results, drawdown and cost sensitivity before making any replacement."},{"t":"2026-10-11T02:00:18.566Z","text":"The accepted ensemble is still submission #005: $8,104.80 after costs, with a 46.7% drawdown. The final tests will focus on event quality and holding time; a replacement needs stronger results across the annual slices to justify changing the entry."},{"t":"2026-10-11T02:02:04.825Z","text":"Filtering out coins after large recent run-ups improved all three annual slices in the first tests. I’m checking nearby settings and longer holds to see how broadly that result holds; it may still reflect a few avoided losses."},{"t":"2026-10-11T02:05:04.566Z","text":"Allowing several days for recovery made a much larger difference: the three-day version reached about $81,600, while averaging two-, three- and four-day holds reached about $42,700 with a lower drawdown. Growth remains uneven across years. I’m comparing cost sensitivity and dependence on the best days before choosing."},{"t":"2026-10-11T02:07:53.099Z","text":"The recovery rule looks more balanced when it holds a rebound position only while it remains below its entry price, then exits within three to five days. Averaging nearby settings reached about $32,200 with a 38% drawdown; using a milder beta penalty in the trend core raised that to about $45,200. I’m finishing the checks for that candidate."},{"t":"2026-10-11T02:11:50.435Z","text":"Submission #006 is with the referee. It passed the local causality checks and turned $100 into $45,164.45, with a 38.0% drawdown; doubled costs left $15,418.80. I’m updating the writeup with the final state rules and all 1,962 scored configurations, including the limits of the historical evidence."},{"t":"2026-10-11T02:15:25.438Z","text":"The referee accepted #006, placing it fifth at acceptance. The writeup is complete, and I verified that `crane.py` matches the accepted source byte for byte. I’m keeping this entry for the hidden years: it has more balanced annual results and less reliance on its best days than the higher-scoring ordinary-hold variants."},{"t":"2026-10-11T02:15:34.694Z","text":"Accepted entry **#006**, crane.py, averages 54 quiet-decline recovery rules with a beta-adjusted BTC/ETH/SOL trend fallback. Visible results: **$100 → $45,164.45 (451.64×)**, **38.0% maximum drawdown**, fifth at acceptance. Causality checks passed; doubled costs still left $15,418.80. WRITEUP.md is complete with equations, all 1,962 scored configurations, failures and the rationale for unseen years."}],"submissions":[{"n":"001","t":"2026-10-11T01:22:30.935842+00:00","note":"Initial BTC 200-day trend baseline; low-volatility research follows.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:31:35.728383+00:00","note":"Replace baseline: SOL 4h 6/60 EMA trend, positive 14-day momentum, causal BTC 100-day gate.","status":"ok","dev":642.4061247585988,"dd":-0.42119630643712047,"reason":""},{"n":"003","t":"2026-10-11T01:38:17.457527+00:00","note":"Low-beta direction recovered: rank 7/14-day trend momentum per BTC beta; 4h 6/60 EMA and causal daily BTC 50-day gate.","status":"ok","dev":1030.5387267178141,"dd":-0.40590993271495046,"reason":""},{"n":"004","t":"2026-10-11T01:44:47.611861+00:00","note":"Hybrid: buy severe low-participation daily dislocations for one day; otherwise hold the strongest major trend per BTC beta.","status":"ok","dev":5764.658446754497,"dd":-0.514414990246549,"reason":""},{"n":"005","t":"2026-10-11T01:54:03.836970+00:00","note":"Final round candidate: equal-weight six quiet-dislocation rules, 20/30-day volume and 6/8/10% drops, with a common low-beta major trend core.","status":"ok","dev":8104.79680550818,"dd":-0.46687274018633695,"reason":""},{"n":"006","t":"2026-10-11T02:08:45.931979+00:00","note":"Quiet-dislocation recovery ensemble: 54 fixed rules, three-to-five-day underwater holds, data-selected extension caps, square-root beta major trend core; annual consistency and cost stress prioritized","status":"ok","dev":45164.45441877173,"dd":-0.3800877123124301,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: #005 accepted — crane.py, quiet-dislocation ensemble; writeup.  \nVisible years: **$100 → $8,104.80**, **−46.7% drawdown**, 10th/40 at acceptance; causality checks passed.  \nNext idea: walk-forward event-quality filters to reduce dependence on rare rebounds."},{"k":2,"text":"Accepted entry **#006**, crane.py, averages 54 quiet-decline recovery rules with a beta-adjusted BTC/ETH/SOL trend fallback. Visible results: **$100 → $45,164.45 (451.64×)**, **38.0% maximum drawdown**, fifth at acceptance. Causality checks passed; doubled costs still left $15,418.80. WRITEUP.md is complete with equations, all 1,962 scored configurations, failures and the rationale for unseen years."}],"title":"Crane (a32): quiet dislocations, recovery holds and a beta trend core","idea":"The entry buys a liquid, established coin after a substantial daily decline only when its trading volume is unusually small relative to its own recent volume. Low participation distinguishes temporarily weak bids from an actively traded selloff, although it cannot establish the cause of a decline. Coins far above their own price average are excluded to reduce purchases after speculative run-ups. Each virtual rule chooses the strongest eligible price movement per unit of participation. If the next day brings no new event, it waits for recovery only while the position remains below its reference entry price, within a three-, four- or five-day age limit. Otherwise it returns to a major-coin trend or cash. The core selects eligible BTC, ETH or SOL momentum divided by the square root of estimated BTC beta, when BTC's broader trend is positive. The actual account equally averages 54 fixed rules across participation horizons, decline thresholds, extension caps and recovery horizons. Low beta supplies the starting direction in the core; low participation supplies the opportunistic component. Every other asset is selected from contemporaneous data. Entry: crane.py, daily bars, accepted submission #006, ranked 5th of 40 at acceptance. Source SHA256: 64777320e1b95fdf0fa5a24d0af04fdacd7e84ea7e8f1096884b423eba4fc97e. The six-rule, one-day ensemble in submission #005 is retained as cra","failed":"- Unconditional low-volatility selection among positive-trend liquid altcoins: a weekly three-coin portfolio reduced $100 to about $16.68. Relative quietness did not imply positive absolute drift. - Daily cross-sectional momentum: 30-day return divided by 30-day volatility, top three and weekly rebalancing, ended near $6.84. Recent winners often represented mature spikes; turnover and later declines compounded losses. - Broad low-beta and residual-momentum screens did not repair that failure. Small market covariance can coexist with large idiosyncratic losses. - Ordinary pullbacks, generic daily breakouts and trailing paper trend-expert ranking generally failed to beat the BTC baseline. Tighter stops often added churn without enough avoided losses. - Walk-forward boosted-tree prediction used matured three-/seven-day labels, fixed seeds, generic features and purged training endpoints. The strongest evaluated absolute-return threshold variant was only about $116.87. These predictive models were not promoted. - Four-hour panic-rebound rules on the majors were weak; hourly trend decisions did not improve the better four-hour candidates. - Buying ordinary/high-volume declines failed badly. Event-only 8% log declines with participation below 0.75 ended near $20.28; below 1.0 ended near $8.36. Low participation was essential in those tests. - An early event-only three-day 12%/0.30 rule reached about $1,990.58, but nearly all growth came in the first annual slice. Longer holds became useful in later hybrid/ensemble tests; the entry further conditions continuation on being below its reference entry price. - Volatility-standardized and residual shocks, or selecting events directly by low beta, generally weakened the event component. Beta adjustment worked better in the trend core","holdup":"The proposed mechanism is temporary price pressure when participation has dried up. In the available data, actively traded declines were poor reversal opportunities, whereas quiet declines had a more favorable response. Ranking movement per unit of participation targets that distinction. Avoiding highly extended prices reduces purchases after a speculative expansion, when disappearing participation may instead signal exhausted demand. Neither condition proves the decline is temporary. A weak-bid disturbance can need several days to resolve. Carrying only below the reference entry price keeps a modest recovery window; reaching that price ends the carry condition, and the rule returns to a trend or cash. Several fixed age and extension choices acknowledge uncertainty about duration and overextension. The risk is that underwater positions reflect permanent loss of demand rather than delayed recovery. Fresh-event priority can prolong exposure during repeated declines, so these are not guaranteed maximum-duration or tightly stopped trades. The core follows persistence in major-coin selection, with coin and BTC trend gates. Its square-root beta penalty favors stronger momentum for compar","result":{"dev":{"final":45164.45441877173,"max_drawdown":-0.3800877123124301,"trades":1733,"exposure":0.5816320195034601,"assets_traded":208,"fees_paid":14878.234708396558},"hidden":{"final":264.72313006763284,"max_drawdown":-0.7622816792745888,"trades":1671,"exposure":0.6876450999565984,"assets_traded":242,"fees_paid":263.2636085132481},"full":{"final":119560.75741534177,"max_drawdown":-0.7622816792745888,"trades":3404,"exposure":0.6240140286659963,"assets_traded":365,"fees_paid":133779.80717657608}},"freq":"1d","equity":[["2021-10-17",108.7326],["2021-10-24",133.3591],["2021-10-31",129.1932],["2021-11-07",157.4521],["2021-11-14",144.7281],["2021-11-21",159.8573],["2021-11-28",196.8624],["2021-12-05",176.0632],["2021-12-12",174.0283],["2021-12-19",184.7747],["2021-12-26",182.921],["2022-01-02",187.9206],["2022-01-09",188.4993],["2022-01-16",204.5756],["2022-01-23",197.2225],["2022-01-30",203.7346],["2022-02-06",210.5353],["2022-02-13",191.4266],["2022-02-20",188.9382],["2022-02-27",191.3821],["2022-03-06",185.8056],["2022-03-13",197.0584],["2022-03-20",209.5437],["2022-03-27",220.2582],["2022-04-03",286.8575],["2022-04-10",261.5983],["2022-04-17",249.7496],["2022-04-24",255.2972],["2022-05-01",297.9812],["2022-05-08",300.9796],["2022-05-15",257.435],["2022-05-22",293.5884],["2022-05-29",611.0473],["2022-06-05",635.1756],["2022-06-12",550.3852],["2022-06-19",501.9618],["2022-06-26",623.9139],["2022-07-03",662.8863],["2022-07-10",664.5684],["2022-07-17",1116.3965],["2022-07-24",1247.2194],["2022-07-31",1305.1389],["2022-08-07",1218.9344],["2022-08-14",1396.8328],["2022-08-21",1323.044],["2022-08-28",1262.6972],["2022-09-04",1306.9566],["2022-09-11",1360.8236],["2022-09-18",1200.9512],["2022-09-25",1301.3659],["2022-10-02",1411.3179],["2022-10-09",1389.7836],["2022-10-16",1306.6842],["2022-10-23",1287.8012],["2022-10-30",1348.6775],["2022-11-06",1391.9339],["2022-11-13",1185.3587],["2022-11-20",1367.8142],["2022-11-27",1546.282],["2022-12-04",1629.3267],["2022-12-11",1656.2579],["2022-12-18",1671.9397],["2022-12-25",1679.7888],["2023-01-01",1720.4064],["2023-01-08",1679.243],["2023-01-15",3107.2405],["2023-01-22",3149.542],["2023-01-29",3131.9159],["2023-02-05",3126.1869],["2023-02-12",3165.3024],["2023-02-19",3202.3227],["2023-02-26",2873.8417],["2023-03-05",2683.6267],["2023-03-12",3696.5083],["2023-03-19",4376.8597],["2023-03-26",4434.4852],["2023-04-02",4987.4338],["2023-04-09",4979.1992],["2023-04-16",5313.7611],["2023-04-23",5630.0713],["2023-04-30",5472.6008],["2023-05-07",4957.6929],["2023-05-14",4670.5187],["2023-05-21",4683.8141],["2023-05-28",4688.2816],["2023-06-04",4737.1435],["2023-06-11",4910.5496],["2023-06-18",5062.9832],["2023-06-25",5489.2336],["2023-07-02",5590.0569],["2023-07-09",6433.8276],["2023-07-16",8100.5459],["2023-07-23",8368.0701],["2023-07-30",8838.6419],["2023-08-06",8563.4312],["2023-08-13",8563.4312],["2023-08-20",8047.8654],["2023-08-27",8162.9041],["2023-09-03",8102.2811],["2023-09-10",7638.5507],["2023-09-17",7618.8021],["2023-09-24",8347.0391],["2023-10-01",8941.7929],["2023-10-08",9503.2834],["2023-10-15",9297.7982],["2023-10-22",10869.4522],["2023-10-29",11725.8108],["2023-11-05",15683.067],["2023-11-12",20692.6005],["2023-11-19",21466.6897],["2023-11-26",20471.81],["2023-12-03",20409.8198],["2023-12-10",20999.2243],["2023-12-17",21796.2848],["2023-12-24",30213.369],["2023-12-31",27665.9001],["2024-01-07",25645.9172],["2024-01-14",28744.1072],["2024-01-21",26638.3186],["2024-01-28",26691.8029],["2024-02-04",27317.2405],["2024-02-11",28784.508],["2024-02-18",31061.1712],["2024-02-25",33828.7218],["2024-03-03",38654.985],["2024-03-10",43547.3939],["2024-03-17",52807.0891],["2024-03-24",49591.4514],["2024-03-31",51619.3789],["2024-04-07",52712.7108],["2024-04-14",34994.2909],["2024-04-21",40320.5157],["2024-04-28",40415.9971],["2024-05-05",41289.4506],["2024-05-12",41596.1723],["2024-05-19",45381.4752],["2024-05-26",46115.0328],["2024-06-02",46014.2349],["2024-06-09",43860.817],["2024-06-16",43044.5882],["2024-06-23",43715.0357],["2024-06-30",43586.3579],["2024-07-07",42694.6454],["2024-07-14",43771.8662],["2024-07-21",47051.5293],["2024-07-28",55876.6998],["2024-08-04",59073.5592],["2024-08-11",61304.7823],["2024-08-18",62370.6812],["2024-08-25",59006.5426],["2024-09-01",48133.9231],["2024-09-08",46688.4042],["2024-09-15",46940.3493],["2024-09-22",49899.4736],["2024-09-29",49106.7719],["2024-10-06",45071.611],["2024-10-13",46410.8201],["2024-10-20",48384.9075],["2024-10-27",54004.9366],["2024-11-03",55858.2669],["2024-11-10",43162.2084],["2024-11-17",43930.8656],["2024-11-24",46990.5166],["2024-12-01",47615.4291],["2024-12-08",51377.5057],["2024-12-15",48752.09],["2024-12-22",45778.7848],["2024-12-29",51557.6056],["2025-01-05",52383.7465],["2025-01-12",47000.2612],["2025-01-19",52935.035],["2025-01-26",48734.3681],["2025-02-02",53005.936],["2025-02-09",45358.1549],["2025-02-16",46679.7701],["2025-02-23",48299.3493],["2025-03-02",47403.0324],["2025-03-09",51259.2241],["2025-03-16",50117.3215],["2025-03-23",51583.6494],["2025-03-30",48515.1924],["2025-04-06",49985.7797],["2025-04-13",53242.0082],["2025-04-20",69996.3686],["2025-04-27",83799.1372],["2025-05-04",93343.8686],["2025-05-11",106590.0684],["2025-05-18",93046.9294],["2025-05-25",102260.1145],["2025-06-01",100521.1524],["2025-06-08",98619.1426],["2025-06-15",95592.823],["2025-06-22",92147.2639],["2025-06-29",96293.3252],["2025-07-06",94044.0123],["2025-07-13",109886.7991],["2025-07-20",133533.291],["2025-07-27",138831.2329],["2025-08-03",124934.7335],["2025-08-10",148196.4134],["2025-08-17",214543.6819],["2025-08-24",217942.1847],["2025-08-31",217976.8255],["2025-09-07",224818.1402],["2025-09-14",238772.727],["2025-09-21",234997.3419],["2025-09-28",267819.7016],["2025-10-05",296026.7517],["2025-10-12",250818.3032],["2025-10-19",254837.9639],["2025-10-26",271887.4826],["2025-11-02",234406.3746],["2025-11-09",233921.6769],["2025-11-16",148673.221],["2025-11-23",127843.7333],["2025-11-30",118165.2158],["2025-12-07",125275.669],["2025-12-14",126351.6773],["2025-12-21",117484.3029],["2025-12-28",116487.9437],["2026-01-04",118114.0438],["2026-01-11",178164.7945],["2026-01-18",182223.3701],["2026-01-25",173684.1308],["2026-02-01",159655.4647],["2026-02-08",158423.0919],["2026-02-15",153909.0066],["2026-02-22",151121.6075],["2026-03-01",139406.4459],["2026-03-08",135177.3898],["2026-03-15",149189.3841],["2026-03-22",142707.0494],["2026-03-29",136292.7113],["2026-04-05",151673.2903],["2026-04-12",163158.9355],["2026-04-19",156718.5914],["2026-04-26",133866.6053],["2026-05-03",89117.9566],["2026-05-10",106707.3052],["2026-05-17",93855.7151],["2026-05-24",96822.4399],["2026-05-31",91585.2991],["2026-06-07",78024.6068],["2026-06-14",93124.8984],["2026-06-21",75495.8517],["2026-06-28",120235.7163],["2026-07-05",143785.7839],["2026-07-12",142533.9487],["2026-07-19",150401.8393],["2026-07-26",132132.3988],["2026-08-02",106468.677],["2026-08-09",114079.6935],["2026-08-16",97728.1178],["2026-08-23",134116.9479],["2026-08-30",152198.8297],["2026-09-06",142655.3807],["2026-09-13",125747.6651],["2026-09-20",123946.0033],["2026-09-27",134529.9681],["2026-10-04",127885.3293],["2026-10-11",119560.7574]],"code_bytes":4087},"a33":{"id":"a33","alias":"Stork","theme":"Drawdown control: CPPI, constant-proportion insurance, and drawdown-based de-risking","started":"2026-10-11T01:22:20+00:00","rounds":[{"k":1,"minutes":39.4},{"k":2,"minutes":13.0}],"messages":[{"t":"2026-10-11T01:22:26.643Z","text":"I’ll inspect the simulator and data, submit an initial drawdown-controlled strategy, then test improvements across separate parts of the visible history. I’ll record every variant and its equations in the writeup."},{"t":"2026-10-11T01:23:24.956Z","text":"The first entry is awaiting the referee: it rotates among three liquid momentum leaders and moves to cash when Bitcoin loses its broad trend. Next I’ll test whether faster rotation and drawdown-based sizing improve returns after costs, including in the bear-market portion."},{"t":"2026-10-11T01:24:36.689Z","text":"The initial strategy failed badly: $100 fell to $13.62 with a 93% drawdown. I verified that my faster research simulator matches the referee’s result. Ranking by raw returns appears to concentrate the portfolio in unstable price spikes, so I’m testing volatility-adjusted trends and short-term reversal before adding insurance."},{"t":"2026-10-11T01:27:55.336Z","text":"The screens so far still lose money after costs. I’m now testing two ways to avoid buying late in a spike: entering fresh breakouts and holding through the trend, and a walk-forward model that learns return relationships from completed observations. Neither uses hand-picked altcoins."},{"t":"2026-10-11T01:31:18.598Z","text":"The replacement entry was accepted: $100 became $515.90, with a 54.4% drawdown. The altcoin breakout and walk-forward tests have not beaten it. I’m shifting focus to Bitcoin, Ether and Solana, where trading costs are lower, and testing whether shorter-bar exits and CPPI-style sizing preserve more gains."},{"t":"2026-10-11T01:34:09.318Z","text":"A useful pattern emerged among recently listed coins: two-week momentum in their first two months outperformed the same rule on older coins. On one version, the insurance layer raised the result from $1,029 to $1,751 and reduced drawdown from 70% to 45%. I’m checking nearby parameters and averaging entry schedules to assess how much depends on timing."},{"t":"2026-10-11T01:36:43.098Z","text":"The insured recent-listing strategy is awaiting the referee. Its research estimate is $2,008.57 with a 48.7% drawdown, and all three fresh annual windows finish above $100. Averaging entry schedules keeps much of the gain. I’m now testing whether reserving capital for major-coin trends helps when fewer young coins qualify."},{"t":"2026-10-11T01:38:01.504Z","text":"Submission #003 passed the causal checks and is now the entry: $100 became $2,008.57, with a 48.7% drawdown. The writeup now contains its full equations. Remaining checks will cover doubled costs, dependence on the best days, nearby parameters, and coin-level risk limits."},{"t":"2026-10-11T01:41:12.315Z","text":"Allocating unused capital to a major-coin trend sleeve raises the research result to about $5,077. Drawdown also rises to 64%, so I’m testing insurance on that sleeve separately. Nearby major trend settings show similar gains, which supports checking this combination further."},{"t":"2026-10-11T01:43:08.775Z","text":"Submission #004 was accepted at $5,077.49. Adding insurance to the combined reference portfolio looks more promising than insuring the major sleeve alone: the tested version returns about $7,430 while reducing drawdown from 64% to 41%. I’ll verify it with the official simulator before submitting it."},{"t":"2026-10-11T01:48:13.331Z","text":"The simulator confirms $7,430.11, and submission #005 is awaiting its causal check. The checks also expose a weakness: shifting the two-day schedule changes the early results substantially. Both schedules perform similarly in the last visible year. I’m testing immediate exits when a held coin fails its filter to see whether that reduces timing dependence."},{"t":"2026-10-11T01:51:01.600Z","text":"Submission #005 is accepted and ranks tenth: $100 became $7,430.11 with a 41.3% drawdown. Immediate exits and partial profit-taking lowered returns in the tested variants, so I’m keeping the current rule. I’ll finish its cost stress test and writeup, including the timing sensitivity and dependence on a few large gains."},{"t":"2026-10-11T01:56:23.507Z","text":"The cost stress test passed: doubling execution costs still leaves $5,045.54, with a 43.4% drawdown. The writeup is complete and the working file matches the accepted entry byte for byte. I’m using the remaining round time for one test of whether low nominal prices add useful information within the young-coin cohort."},{"t":"2026-10-11T01:59:07.049Z","text":"The nominal-price filter weakened returns, so I’ve left it out. The accepted entry and writeup are ready. I’m finishing one small test of whether avoiding the very strongest mature-coin spikes produces a useful second momentum sleeve."},{"t":"2026-10-11T02:00:47.914Z","text":"The mature-coin test also underperformed. I’m retaining submission #005 for this round, with the stress results and failed approaches documented. Next I’ll test rebalancing by each coin’s listing age, which could reduce reliance on the portfolio’s two-day calendar phase."},{"t":"2026-10-11T02:01:41.411Z","text":"Entry: accepted #005 — stork.py; writeup complete.  \nVisible result: $100 → **$7,430.11**; max drawdown **41.3%**; causal checks passed.  \nNext idea: rebalance by each coin’s listing age to reduce calendar sensitivity."},{"t":"2026-10-11T02:01:47.602Z","text":"I’ll test listing-age rebalancing first, then compare any promising variant with the accepted $7,430 strategy. I’ll leave time for referee scoring and update the writeup to match the final accepted entry."},{"t":"2026-10-11T02:04:09.272Z","text":"The accepted $7,430 entry is intact. Its largest weakness is sensitivity to the two-day rebalance schedule, so I’m testing whether each coin’s own trading age provides a more stable clock."},{"t":"2026-10-11T02:05:09.801Z","text":"The listing-age clocks returned $1,171–$4,317, below the accepted entry. Some did better in the last visible year, but with weaker earlier results. I’m now testing whether retaining young winners while their trends remain intact helps more than regularly replacing them by rank."},{"t":"2026-10-11T02:06:07.061Z","text":"Retaining winners also fell short: the best variant reached $4,127. I’m testing one final insurance change—letting the reference high-water mark decay gradually, which could restore exposure more smoothly after losses than a rolling maximum."},{"t":"2026-10-11T02:07:41.721Z","text":"Gradual floor decay also underperformed; its best variant reached $5,407. I’ll keep the accepted insurance rule. My last test will combine the calendar and listing-age schedules using only their past reference returns to see whether diversification improves timing stability."},{"t":"2026-10-11T02:10:45.121Z","text":"The schedule mixtures improved the last visible year but had lower overall growth and higher drawdowns. I’m checking whether insurance on the combined mixture offsets that weakness. I’ll then freeze the entry and finish the writeup."},{"t":"2026-10-11T02:12:34.187Z","text":"I’m keeping submission #005 as the final entry. The newer mixtures improved recent-year returns, but none showed a clear enough advantage across the visible sample to justify replacing it. I’ll record those tradeoffs and verify that `stork.py` exactly matches the accepted file."},{"t":"2026-10-11T02:14:38.113Z","text":"My final entry remains accepted submission #005, stork.py: recent-listing momentum, BTC/ETH/SOL trend allocation, and two renewable CPPI insurance layers. It turned **$100 into $7,430.11** over the visible years after costs, with **41.3% maximum drawdown**. WRITEUP.md is complete, covering every equation and parameter, 1,153 logged evaluations, results, failed ideas, and the case for—and limits of—generalization."}],"submissions":[{"n":"001","t":"2026-10-11T01:23:06.813536+00:00","note":"Initial liquid top-three momentum with broad trend cash filter","status":"ok","dev":13.622760861369475,"dd":-0.9273060216033145,"reason":""},{"n":"002","t":"2026-10-11T01:28:59.443470+00:00","note":"Replace failed raw altcoin rotation with slow major-asset leadership and own-price trend gate","status":"ok","dev":515.9040233968162,"dd":-0.544108627838015,"reason":""},{"n":"003","t":"2026-10-11T01:34:59.455834+00:00","note":"Recent-listing momentum with renewable shadow-wealth CPPI: two-thirds floor, multiplier six, ninety-day renewal","status":"ok","dev":2008.566423242205,"dd":-0.4868833601950425,"reason":""},{"n":"004","t":"2026-10-11T01:41:18.556785+00:00","note":"Renewable young-coin insurance with unused capital allocated to gated BTC/ETH/SOL leadership","status":"ok","dev":5077.490902643872,"dd":-0.6399011771291332,"reason":""},{"n":"005","t":"2026-10-11T01:46:47.853572+00:00","note":"Hierarchical renewable CPPI: young-listing leadership, major trend cash allocation, combined reference floor","status":"ok","dev":7430.112339570055,"dd":-0.413489681394952,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #005 — stork.py; writeup complete.  \nVisible result: $100 → **$7,430.11**; max drawdown **41.3%**; causal checks passed.  \nNext idea: rebalance by each coin’s listing age to reduce calendar sensitivity."},{"k":2,"text":"My final entry remains accepted submission #005, stork.py: recent-listing momentum, BTC/ETH/SOL trend allocation, and two renewable CPPI insurance layers. It turned **$100 into $7,430.11** over the visible years after costs, with **41.3% maximum drawdown**. WRITEUP.md is complete, covering every equation and parameter, 1,153 logged evaluations, results, failed ideas, and the case for—and limits of—generalization."}],"title":"Stork (a33): hierarchical renewable capital insurance","idea":"The entry combines recent-listing momentum with major-coin trend following and two layers of reference-capital insurance. It buys up to three liquid momentum leaders in their first two months of observed trading. A renewable CPPI rule reduces that sleeve's allocation when a continuously simulated uninsured version loses capital. Unused target capital goes to the strongest eligible trend among BTC, ETH and SOL. A second renewable CPPI rule monitors a reference version of this combined allocation and can move the whole portfolio toward cash. Rolling rather than permanent high-water marks permit recovery after a bad episode. This is risk management using reference portfolios: neither floor is a guarantee on the actual account. All altcoin selection comes from the supplied data.","failed":"- Raw momentum over 7–56 bars among the top 80 liquid coins repeatedly selected mature spikes. Every screened combination lost capital over the visible interval. - Standard-deviation adjustment, near-high conditions, jump restrictions and pullback ranking did not create a reliable broad altcoin sleeve. - Buying the strongest daily or weekly losers incurred costs and persistent-distress risk. - Holding fresh breakouts and buying sudden volume expansion after quiet trading did not beat major trend following. - Walk-forward shallow boosting on generic price, volatility, volume, market and sentiment features underperformed simpler entries. - Extending the young cohort to 180 or 365 observations weakened it. Buying even earlier with adaptive momentum and only two to ten observations also weakened performance. - Limiting young assets to one-third slots and filling the rest with majors lowered concentration but did not clearly improve the insured rule. - Separate insurance on the major sleeve reduced growth; insurance on the combined reference portfolio worked better. - Immediate exits when a held coin failed eligibility, partial profit-taking after a large daily jump, and an additional per-asset volatility cap reduced final growth. Corrected volatility-cap tests returned $2,914.43 and $1,124.13 under the two entry schedules. - Averaging calendar phases and adaptively weighting them reduced first-year gains without materially improving the final visible year. - Own-age update clocks, incumbent-winner retention and continuous high-water-mark decay did not improve the full visible-period result. Mixtures of own-age schedules improved the most recent periods but added trading and model complexity; their advantage on a selected recent window was insufficient to replace the accepte","holdup":"The proposed edge concerns a recurring market process rather than named winners. Recent venue listings can experience adoption, price discovery and changing demand. The entry requires observed positive momentum and liquidity instead of assuming every launch succeeds. The universe renews itself from the data, and a short own-price condition plus a broad market gate remove many weak candidates. A separate major trend sleeve offers opportunities when the young cohort is sparse or its reference performance deteriorates. The insurance mechanism responds to a strategy's recent realized reference losses rather than estimating a delicate expected-return model. It retains full exposure near a reference high, reduces exposure during deterioration, and eventually renews after a poor episode. The global layer improved both visible growth and drawdown relative to allocating every unused dollar to majors. It also survived higher execution costs and many nearby parameter settings. This is a hypothesis supported by a short, selected historical sample. Listing policy, momentum persistence, market regimes, correlated token failures and entry timing can change. Approximately 1,000 configurations and ","result":{"dev":{"final":7430.112339570055,"max_drawdown":-0.413489681394952,"trades":592,"exposure":0.345413840794182,"assets_traded":82,"fees_paid":1010.4600147701595},"hidden":{"final":102.10178171052401,"max_drawdown":-0.7283196187735861,"trades":445,"exposure":0.30784123572214267,"assets_traded":55,"fees_paid":36.51833652418674},"full":{"final":7565.731781910026,"max_drawdown":-0.7283196187735862,"trades":1040,"exposure":0.33039302934698117,"assets_traded":131,"fees_paid":3735.0364911253887}},"freq":"1d","equity":[["2021-10-17",96.9233],["2021-10-24",110.0072],["2021-10-31",108.8668],["2021-11-07",143.7687],["2021-11-14",134.3951],["2021-11-21",108.4745],["2021-11-28",108.4745],["2021-12-05",108.4745],["2021-12-12",108.4745],["2021-12-19",108.4745],["2021-12-26",108.4745],["2022-01-02",108.4745],["2022-01-09",108.4745],["2022-01-16",108.4745],["2022-01-23",108.4745],["2022-01-30",108.4745],["2022-02-06",108.4745],["2022-02-13",92.1742],["2022-02-20",89.8411],["2022-02-27",89.8411],["2022-03-06",112.493],["2022-03-13",118.7974],["2022-03-20",113.8475],["2022-03-27",114.3239],["2022-04-03",278.7247],["2022-04-10",227.2304],["2022-04-17",228.6307],["2022-04-24",228.6307],["2022-05-01",228.6307],["2022-05-08",228.6307],["2022-05-15",228.6307],["2022-05-22",228.6307],["2022-05-29",228.6307],["2022-06-05",228.6307],["2022-06-12",228.6307],["2022-06-19",228.6307],["2022-06-26",228.6307],["2022-07-03",228.6307],["2022-07-10",228.6307],["2022-07-17",228.6307],["2022-07-24",297.3327],["2022-07-31",434.0148],["2022-08-07",408.1588],["2022-08-14",402.4144],["2022-08-21",409.6592],["2022-08-28",409.6592],["2022-09-04",409.6592],["2022-09-11",409.6592],["2022-09-18",409.6592],["2022-09-25",409.6592],["2022-10-02",409.6592],["2022-10-09",405.1544],["2022-10-16",405.1544],["2022-10-23",405.1544],["2022-10-30",449.5431],["2022-11-06",396.0572],["2022-11-13",370.3045],["2022-11-20",370.3045],["2022-11-27",370.3045],["2022-12-04",370.3045],["2022-12-11",370.3045],["2022-12-18",370.3045],["2022-12-25",370.3045],["2023-01-01",370.3045],["2023-01-08",358.8014],["2023-01-15",522.976],["2023-01-22",740.1354],["2023-01-29",1134.8469],["2023-02-05",1082.5174],["2023-02-12",994.7376],["2023-02-19",1105.6951],["2023-02-26",953.8923],["2023-03-05",919.9053],["2023-03-12",919.9053],["2023-03-19",908.5767],["2023-03-26",796.4357],["2023-04-02",796.0978],["2023-04-09",796.0978],["2023-04-16",794.5592],["2023-04-23",769.2731],["2023-04-30",769.2731],["2023-05-07",769.2731],["2023-05-14",769.2731],["2023-05-21",769.2731],["2023-05-28",769.2731],["2023-06-04",769.2731],["2023-06-11",769.2731],["2023-06-18",769.2731],["2023-06-25",804.6115],["2023-07-02",794.7859],["2023-07-09",772.5318],["2023-07-16",842.1092],["2023-07-23",751.9489],["2023-07-30",771.8823],["2023-08-06",713.0898],["2023-08-13",713.0898],["2023-08-20",713.0898],["2023-08-27",713.0898],["2023-09-03",713.0898],["2023-09-10",713.0898],["2023-09-17",713.0898],["2023-09-24",803.2452],["2023-10-01",806.7286],["2023-10-08",876.5117],["2023-10-15",827.6076],["2023-10-22",988.9217],["2023-10-29",1061.3132],["2023-11-05",2020.7093],["2023-11-12",2647.5763],["2023-11-19",2891.3136],["2023-11-26",2597.3525],["2023-12-03",3104.1035],["2023-12-10",4011.1377],["2023-12-17",4913.4804],["2023-12-24",6185.3831],["2023-12-31",6259.5049],["2024-01-07",5155.7342],["2024-01-14",5199.9122],["2024-01-21",5307.9381],["2024-01-28",5307.9381],["2024-02-04",5054.6197],["2024-02-11",5522.9616],["2024-02-18",5518.4018],["2024-02-25",5615.2391],["2024-03-03",6523.19],["2024-03-10",7239.1254],["2024-03-17",8216.449],["2024-03-24",7916.2885],["2024-03-31",9254.8182],["2024-04-07",7447.3385],["2024-04-14",6737.6542],["2024-04-21",6737.6542],["2024-04-28",6737.6542],["2024-05-05",6737.6542],["2024-05-12",6737.6542],["2024-05-19",6737.6542],["2024-05-26",6737.6542],["2024-06-02",6737.6542],["2024-06-09",6050.9222],["2024-06-16",6050.9222],["2024-06-23",6050.9222],["2024-06-30",6050.9222],["2024-07-07",6050.9222],["2024-07-14",6050.9222],["2024-07-21",6050.9222],["2024-07-28",6050.9222],["2024-08-04",6050.9222],["2024-08-11",6050.9222],["2024-08-18",6050.9222],["2024-08-25",6050.9222],["2024-09-01",6050.9222],["2024-09-08",6050.9222],["2024-09-15",6050.9222],["2024-09-22",6050.9222],["2024-09-29",6050.9222],["2024-10-06",6050.9222],["2024-10-13",8704.9221],["2024-10-20",8705.9389],["2024-10-27",6018.2553],["2024-11-03",6018.2553],["2024-11-10",6028.4195],["2024-11-17",6065.5815],["2024-11-24",6357.0998],["2024-12-01",6507.2402],["2024-12-08",8645.9586],["2024-12-15",10363.0653],["2024-12-22",15467.3046],["2024-12-29",15015.2172],["2025-01-05",15385.0797],["2025-01-12",13822.9366],["2025-01-19",13907.1334],["2025-01-26",13546.176],["2025-02-02",13288.9961],["2025-02-09",13288.9961],["2025-02-16",13288.9961],["2025-02-23",13288.9961],["2025-03-02",13288.9961],["2025-03-09",13288.9961],["2025-03-16",13288.9961],["2025-03-23",13288.9961],["2025-03-30",13288.9961],["2025-04-06",13288.9961],["2025-04-13",13288.9961],["2025-04-20",12170.7547],["2025-04-27",13702.5143],["2025-05-04",15641.785],["2025-05-11",13501.8296],["2025-05-18",12833.4585],["2025-05-25",12576.8359],["2025-06-01",10475.3241],["2025-06-08",10469.932],["2025-06-15",10469.932],["2025-06-22",10469.932],["2025-06-29",10469.932],["2025-07-06",10469.932],["2025-07-13",10469.932],["2025-07-20",10546.9488],["2025-07-27",9434.031],["2025-08-03",9434.031],["2025-08-10",9472.2996],["2025-08-17",9863.2216],["2025-08-24",8971.6258],["2025-08-31",8962.6542],["2025-09-07",8962.6542],["2025-09-14",9370.7793],["2025-09-21",8997.055],["2025-09-28",8423.6889],["2025-10-05",8958.474],["2025-10-12",7638.2868],["2025-10-19",7571.1333],["2025-10-26",7571.1333],["2025-11-02",7525.4016],["2025-11-09",7525.4016],["2025-11-16",7525.4016],["2025-11-23",7525.4016],["2025-11-30",7525.4016],["2025-12-07",7525.4016],["2025-12-14",7525.4016],["2025-12-21",7525.4016],["2025-12-28",7525.4016],["2026-01-04",7863.7595],["2026-01-11",6934.0063],["2026-01-18",6934.0063],["2026-01-25",6239.6529],["2026-02-01",6239.6529],["2026-02-08",6239.6529],["2026-02-15",6239.6529],["2026-02-22",6239.6529],["2026-03-01",6239.6529],["2026-03-08",6239.6529],["2026-03-15",6239.6529],["2026-03-22",5709.2601],["2026-03-29",5665.2026],["2026-04-05",5665.2026],["2026-04-12",7448.4289],["2026-04-19",7856.2306],["2026-04-26",11177.6361],["2026-05-03",10530.7707],["2026-05-10",10572.2409],["2026-05-17",10375.5007],["2026-05-24",10375.5007],["2026-05-31",10375.5007],["2026-06-07",10375.5007],["2026-06-14",10375.5007],["2026-06-21",10375.5007],["2026-06-28",10375.5007],["2026-07-05",10375.5007],["2026-07-12",10375.5007],["2026-07-19",10375.5007],["2026-07-26",9699.6673],["2026-08-02",8538.1756],["2026-08-09",8488.3964],["2026-08-16",8297.2923],["2026-08-23",8959.9213],["2026-08-30",8939.4723],["2026-09-06",9382.5985],["2026-09-13",9516.6921],["2026-09-20",10071.6642],["2026-09-27",9197.0353],["2026-10-04",8158.6293],["2026-10-11",7565.7318]],"code_bytes":4708},"a34":{"id":"a34","alias":"Spoonbill","theme":"Optimal stopping and trailing-stop design for holding winners","started":"2026-10-11T01:22:58+00:00","rounds":[{"k":1,"minutes":39.5},{"k":2,"minutes":17.0}],"messages":[{"t":"2026-10-11T01:23:04.219Z","text":"I’ll inspect the simulator and data, submit an early baseline, then test ways to hold strong winners while controlling losses. I’ll track every variant and its maths in the writeup."},{"t":"2026-10-11T01:24:01.648Z","text":"The BTC baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. I’m now testing liquid coins with strong momentum and comparing rotation against trailing stops that let winners run."},{"t":"2026-10-11T01:26:10.147Z","text":"Buying the largest recent gainers has performed poorly, even with trailing stops. Large spikes often reverse before the next daily fill. I’m testing signals for steadier trends and measuring whether short-term pullbacks offer better entries."},{"t":"2026-10-11T01:28:14.807Z","text":"The simple reversal rules are also losing money after costs. I’m testing volume surges with price confirmation next. I also have a walk-forward prediction experiment, though the installed LightGBM package currently fails to load because an offline library is missing."},{"t":"2026-10-11T01:30:16.893Z","text":"So far, the simpler BTC/ETH/SOL rotation has the best balance: roughly $685 from $100, with a 46% drawdown. The learned model has not improved on it. I’m testing a small basket next to see whether diversification reduces the damage from failed momentum trades."},{"t":"2026-10-11T01:32:49.485Z","text":"The improved entry is accepted: $100 became $673.45, and the causality check passed. The basket tests have mostly underperformed. I’m now testing entries at the onset of a rally, before recent returns become extreme, with tighter exits after gains accumulate."},{"t":"2026-10-11T01:35:40.066Z","text":"The compressed-base breakout rules found a stronger candidate, around $1,220, but its 76% drawdown is high. I’m testing an adaptive stopping rule next: compare the recent net performance of simulated trailing-stop policies, then select among policies that still advise holding."},{"t":"2026-10-11T01:40:08.266Z","text":"Four-hour stops on BTC, ETH and SOL are looking stronger. A volatility-based barrier that only moves upward produced about $1,214 from $100, with a 29.5% drawdown in the research simulator. I’m running the referee check now, then will test nearby settings and higher costs."},{"t":"2026-10-11T01:43:19.370Z","text":"The referee accepted the strategy at $1,214.35. Doubling costs still leaves about $1,153. All 20 parameter checks at ±20% remain profitable, ranging from roughly $632 to $1,217. Most gains came in the final visible year, so I’m checking whether the market filter exits winners too early."},{"t":"2026-10-11T01:49:04.096Z","text":"The writeup now documents the signals, search and failed ideas. The remaining market-filter variants have not improved the entry. I’m making one final comparison using the same holding rule on finer bars, to see whether quicker exits justify the extra trades and costs."},{"t":"2026-10-11T01:51:05.919Z","text":"Finer bars lowered returns and increased trades in the tested settings. I’m keeping the accepted four-hour entry. The last experiment separates entry from holding: use faster momentum to enter a confirmed bull market, then retain the position using the slower leadership signal."},{"t":"2026-10-11T01:55:09.089Z","text":"The faster-entry rules and adaptive selection among stop policies also underperformed. I’m keeping the accepted entry while testing one final holding rule: let an established winner stay invested through a weak BTC market signal if its own gain still shows strong evidence of a positive trend."},{"t":"2026-10-11T02:01:06.372Z","text":"The final checks left the accepted entry unchanged. The writeup includes the equations, roughly 4,200 trials, robustness results and failed ideas. The local strategy copy matches the accepted submission’s hash, and no submission is pending."},{"t":"2026-10-11T02:02:30.112Z","text":"Entry: winner_stop.py, accepted #003; writeup complete.  \nVisible years: $100 → **$1,214.35**; 29.5% max drawdown; causality check passed.  \nNext idea: learn stop distances from completed trade paths."},{"t":"2026-10-11T02:02:37.687Z","text":"I’ll test whether completed trade paths can guide stop distances, while keeping the accepted entry in place. I’ll leave time for the referee to score any improvement and for the writeup to match the final entry."},{"t":"2026-10-11T02:04:49.957Z","text":"The accepted strategy remains at $1,214.35 from $100, with a 29.5% drawdown. I’m testing a different learning target now: the return after a trailing-stop exit. The earlier models predicted fixed-horizon returns and performed poorly; this target may better match the trades the strategy actually makes."},{"t":"2026-10-11T02:06:52.889Z","text":"The stop-payoff model failed to improve the entry: its best tested policy reached $384 with a 74% drawdown. Rules for newly listed coins also proved fragile; a selected variant reached $1,368 but drew down 71%. Neither gives enough evidence to replace the accepted strategy."},{"t":"2026-10-11T02:08:19.880Z","text":"A stop based on cumulative declines in volatility units reached about $1,217 with a 24.8% drawdown, roughly matching the accepted entry’s growth. I’m checking whether that result holds nearby and comparing range-based volatility estimates, which could make stop distances less sensitive to a single large close-to-close move."},{"t":"2026-10-11T02:11:20.094Z","text":"Range-based stops gave a modest improvement—about $1,262 with a 24.8% drawdown—but the parameter groups performed similarly overall. Giving profitable trades more room also failed to improve growth. I’m keeping the accepted rule unless the remaining tests show a clear reason to change it."},{"t":"2026-10-11T02:12:50.575Z","text":"The final trend-quality test found a stronger candidate: fitting a line to log prices, penalizing poor fit, and scaling by volatility reached $2,796 with a 47.5% drawdown. It ranks coins by sustained trend quality rather than raw gains. I’m checking nearby parameters and costs before deciding whether to replace the entry."},{"t":"2026-10-11T02:13:27.554Z","text":"The trend-quality candidate stayed profitable under all 26 parameter perturbations and reached $2,121 at double costs. Its weak points are clear: changing the lookback or volatility scaling sharply reduced returns. I’ll submit it as the final entry for its broader opportunity set and stronger measured growth, and document both that sensitivity and the higher drawdown."},{"t":"2026-10-11T02:15:24.222Z","text":"Submission #004 is accepted and is now the entry: $100 became $2,796.48, with a 47.5% drawdown and 119 trades. It passed the referee’s causality checks. I’m finishing the writeup to match this code, including the 6,075 logged evaluations and the result’s dependence on a few large winning days."},{"t":"2026-10-11T02:18:48.818Z","text":"The writeup now matches accepted submission #004 and includes the signal equations, search ledger, failed ideas, cost checks and transfer risks. One terminology correction: the regression uses 126 daily crypto bars, about four months. I’m checking the final file hash and documented results before sealing the work."},{"t":"2026-10-11T02:19:27.919Z","text":"Final entry: quality_winner.py, accepted submission **#004**. It ranks liquid coins by persistent trend quality and holds winners behind a volatility-scaled ratcheting stop. Visible results: **$100 → $2,796.48 (27.96×)**, with **47.54% maximum drawdown**; referee causality checks passed. WRITEUP.md is complete, including equations, 6,075 logged evaluations, failed ideas, cost checks, parameter sensitivity and concentration risks."}],"submissions":[{"n":"001","t":"2026-10-11T01:23:15.204673+00:00","note":"Early baseline: BTC above its trailing 200-day average, trade only signal changes.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:30:49.743507+00:00","note":"Winner retention: 126-day momentum among BTC/ETH/SOL, 200-day entry trend, BTC 50-day gate and 30% close-based trailing exit.","status":"ok","dev":673.4472962216441,"dd":-0.45949269664603476,"reason":""},{"n":"003","t":"2026-10-11T01:40:09.616339+00:00","note":"Four-hour winner retention: volatility-scaled ratcheting barrier, slow relative momentum, BTC daily trend gate, and hysteresis against rotation costs.","status":"ok","dev":1214.3545852912318,"dd":-0.29520554540182353,"reason":""},{"n":"004","t":"2026-10-11T02:14:12.561719+00:00","note":"Persistent log-price trend quality, volatility-scaled ratcheting exits; no fitted model or selected tickers","status":"ok","dev":2796.477369224968,"dd":-0.4753994037814232,"reason":""}],"entry":"004","round_notes":[{"k":1,"text":"Entry: winner_stop.py, accepted #003; writeup complete.  \nVisible years: $100 → **$1,214.35**; 29.5% max drawdown; causality check passed.  \nNext idea: learn stop distances from completed trade paths."},{"k":2,"text":"Final entry: quality_winner.py, accepted submission **#004**. It ranks liquid coins by persistent trend quality and holds winners behind a volatility-scaled ratcheting stop. Visible results: **$100 → $2,796.48 (27.96×)**, with **47.54% maximum drawdown**; referee causality checks passed. WRITEUP.md is complete, including equations, 6,075 logged evaluations, failed ideas, cost checks, parameter sensitivity and concentration risks."}],"title":"Spoonbill (a34): persistent trend quality and winner retention","idea":"The final entry is quality_winner.py, accepted submission #004. It trades one Binance spot asset at a time, selected from the top 100 by trailing quote volume. A 126-day regression of log price rewards persistent growth, penalizes a poor linear fit, and scales the resulting score by volatility. A BTC trend gate controls participation. Once invested, a volatility-scaled price barrier ratchets upward without retreating; a challenger needs a score advantage and a minimum holding time to replace the winner. The accepted visible result is $100 → $2,796.48, with 47.54% maximum drawdown. This is a heuristic stopping and selection policy, not a claim to have solved optimal stopping or established the million-dollar objective. The final file's SHA-256 is 19ddbff2b42ac47df9a7b8f31e6c6403dde2ef5b9ca27d8a97772089d835df27. strategy.py is an identical convenience copy. The earlier four-hour entry remains in winner_stop.py, but is superseded by #004.","failed":"- Largest recent gainers: log(C_t/C_{t-L}) ranking, from one week to several months, with percentage stops. Late spike entries, repeated failed rallies and bear exposure overwhelmed winners. The best initial ungated trial ended near $105 with approximately 91% drawdown. - Risk-normalized/skipped returns: M_L / sigma^p, p in {0.5, 1}, and M_L - M_3. These did not reliably cure late entry or broad cross-sectional instability. - Reversal and trend pullbacks: -M_1, -M_3 or -M_7 ranking, with/without volatility normalization, positive longer momentum and moving-average filters. Declines often continued, and altcoin daily costs hurt short holds. - Volume/onset entries: relative volume V_t / mean(V over prior 20 bars), recent volume acceleration, new 7/20-day highs and compressed prior ranges. A selected onset trial reached about $1,220 with 76% drawdown, while nearby settings deteriorated. Compression used max(prior 14 highs)/min(prior 14 lows) - 1 < 0.5. - Periodic baskets: top k = 1, 3 or 7, every three/seven days, equal weights and momentum/trend/market filters. The best selected one-asset result was about $1,386 with 86% drawdown. - Virtual stop-policy selection: each coin had a hypothetical after-cost trend/stop wealth V_i; select on log(V_{i,t}/V_{i,t-L}), sometimes divided by sigma. This produced unstable switching. - Walk-forward forecasts: anonymous pooled coin features included multi-horizon returns, volatility, price/EMA ratios, drawdown, range position, relative volume, market returns and Fear & Greed. Labels were log(O_{t+1+h}/O_{t+1}), clipped to [-0.4, 0.4], for h = 1/7/21/42. Training every 180 bars used only matured labels, a rolling 900-bar sample, fixed seed 17, small histogram-boosted trees and large minimum leaves. The one-day model's best selected policy","holdup":"The hypothesis is that some crypto trends build gradually and subsequently continue or accelerate. A slow fitted slope with a quality penalty seeks established, distributed growth rather than ranking solely on one large recent jump. Volatility normalization and contemporaneous liquidity controls reduce the influence of extreme noisy assets. None requires knowledge of a future listing, coin identity or event. The holding policy gives a successful asset time to realize that asymmetry. Rising prices lift the barrier, while volatility shocks cannot move existing protection downward. The market gate and barrier terminate adverse paths; rotation hysteresis avoids paying for a small change in rank. These mechanisms depend on observed prices and cost arithmetic rather than specific historical dates. This remains a hypothesis. Regression on log levels is descriptive, not a statistical proof of predictable returns. The search was large and the winning ranking is sensitive. A slow fit can enter late; single-asset allocation is concentrated; stops can gap; sideways markets can whipsaw; and future leaders may have little price history. All three measured annual slices were profitable, but the f","result":{"dev":{"final":2796.477369224968,"max_drawdown":-0.4753994037814232,"trades":119,"exposure":0.4854014598540146,"assets_traded":35,"fees_paid":238.47462277144962},"hidden":{"final":12.031179420046632,"max_drawdown":-0.9169696716261443,"trades":85,"exposure":0.5178082191780822,"assets_traded":25,"fees_paid":11.602545909872404},"full":{"final":335.6080867081244,"max_drawdown":-0.9196220457542454,"trades":205,"exposure":0.49835706462212487,"assets_traded":51,"fees_paid":569.117230391472}},"freq":"1d","equity":[["2021-10-17",95.3012],["2021-10-24",118.039],["2021-10-31",112.5685],["2021-11-07",148.0975],["2021-11-14",139.4983],["2021-11-21",118.5547],["2021-11-28",118.5547],["2021-12-05",118.5547],["2021-12-12",118.5547],["2021-12-19",118.5547],["2021-12-26",118.5547],["2022-01-02",118.5547],["2022-01-09",118.5547],["2022-01-16",118.5547],["2022-01-23",118.5547],["2022-01-30",118.5547],["2022-02-06",118.5547],["2022-02-13",107.1194],["2022-02-20",109.061],["2022-02-27",109.061],["2022-03-06",99.1205],["2022-03-13",90.153],["2022-03-20",97.7672],["2022-03-27",92.277],["2022-04-03",97.2094],["2022-04-10",108.9532],["2022-04-17",118.2866],["2022-04-24",118.2866],["2022-05-01",118.2866],["2022-05-08",118.2866],["2022-05-15",118.2866],["2022-05-22",118.2866],["2022-05-29",118.2866],["2022-06-05",118.2866],["2022-06-12",118.2866],["2022-06-19",118.2866],["2022-06-26",118.2866],["2022-07-03",118.2866],["2022-07-10",118.2866],["2022-07-17",118.2866],["2022-07-24",114.3716],["2022-07-31",148.817],["2022-08-07",161.159],["2022-08-14",213.8921],["2022-08-21",156.4236],["2022-08-28",156.4236],["2022-09-04",156.4236],["2022-09-11",156.4236],["2022-09-18",145.9106],["2022-09-25",145.9106],["2022-10-02",145.9106],["2022-10-09",152.0999],["2022-10-16",152.0999],["2022-10-23",152.0999],["2022-10-30",150.879],["2022-11-06",208.7277],["2022-11-13",159.7233],["2022-11-20",159.7233],["2022-11-27",159.7233],["2022-12-04",159.7233],["2022-12-11",159.7233],["2022-12-18",159.7233],["2022-12-25",159.7233],["2023-01-01",159.7233],["2023-01-08",160.2155],["2023-01-15",193.116],["2023-01-22",193.2699],["2023-01-29",197.1393],["2023-02-05",216.574],["2023-02-12",207.2743],["2023-02-19",219.2122],["2023-02-26",219.1488],["2023-03-05",225.8346],["2023-03-12",225.8346],["2023-03-19",233.9636],["2023-03-26",212.0958],["2023-04-02",267.0393],["2023-04-09",248.6111],["2023-04-16",253.1338],["2023-04-23",211.371],["2023-04-30",233.5206],["2023-05-07",209.9474],["2023-05-14",206.0744],["2023-05-21",206.0744],["2023-05-28",206.0744],["2023-06-04",206.0744],["2023-06-11",206.0744],["2023-06-18",206.0744],["2023-06-25",210.3141],["2023-07-02",179.5491],["2023-07-09",183.255],["2023-07-16",182.7728],["2023-07-23",206.7043],["2023-07-30",249.935],["2023-08-06",225.8699],["2023-08-13",225.8699],["2023-08-20",225.8699],["2023-08-27",225.8699],["2023-09-03",225.8699],["2023-09-10",225.8699],["2023-09-17",225.8699],["2023-09-24",235.2209],["2023-10-01",237.8931],["2023-10-08",218.1851],["2023-10-15",219.8921],["2023-10-22",298.5088],["2023-10-29",416.9429],["2023-11-05",467.4297],["2023-11-12",412.0854],["2023-11-19",401.5064],["2023-11-26",414.7302],["2023-12-03",436.88],["2023-12-10",475.1766],["2023-12-17",549.2042],["2023-12-24",1016.2056],["2023-12-31",975.3264],["2024-01-07",941.9459],["2024-01-14",962.8401],["2024-01-21",899.9173],["2024-01-28",899.9173],["2024-02-04",906.8451],["2024-02-11",986.0921],["2024-02-18",993.0461],["2024-02-25",980.0481],["2024-03-03",1139.5048],["2024-03-10",1209.5237],["2024-03-17",1830.9686],["2024-03-24",2691.2145],["2024-03-31",2600.3881],["2024-04-07",2583.7681],["2024-04-14",2779.9451],["2024-04-21",2771.2919],["2024-04-28",2771.2919],["2024-05-05",2771.2919],["2024-05-12",2771.2919],["2024-05-19",2809.7487],["2024-05-26",2550.8858],["2024-06-02",3054.1244],["2024-06-09",3778.0842],["2024-06-16",3181.6782],["2024-06-23",2730.2931],["2024-06-30",2730.2931],["2024-07-07",2730.2931],["2024-07-14",2730.2931],["2024-07-21",3437.5987],["2024-07-28",3762.8858],["2024-08-04",2910.0054],["2024-08-11",2910.0054],["2024-08-18",2910.0054],["2024-08-25",2945.1844],["2024-09-01",2438.7585],["2024-09-08",2438.7585],["2024-09-15",2360.7061],["2024-09-22",2689.9222],["2024-09-29",2702.3903],["2024-10-06",2674.5126],["2024-10-13",2837.7926],["2024-10-20",2923.8957],["2024-10-27",2638.0694],["2024-11-03",2559.1251],["2024-11-10",2399.6693],["2024-11-17",2995.4747],["2024-11-24",2737.4021],["2024-12-01",2768.546],["2024-12-08",3392.3013],["2024-12-15",3551.8443],["2024-12-22",3475.1394],["2024-12-29",3245.493],["2025-01-05",3771.2351],["2025-01-12",3315.6266],["2025-01-19",3211.9079],["2025-01-26",3018.7959],["2025-02-02",2589.746],["2025-02-09",2182.8349],["2025-02-16",2182.8349],["2025-02-23",2182.8349],["2025-03-02",2182.8349],["2025-03-09",2182.8349],["2025-03-16",2182.8349],["2025-03-23",2182.8349],["2025-03-30",2182.8349],["2025-04-06",2182.8349],["2025-04-13",2182.8349],["2025-04-20",2182.8349],["2025-04-27",2182.8349],["2025-05-04",2182.8349],["2025-05-11",2182.8349],["2025-05-18",2166.9426],["2025-05-25",2172.5689],["2025-06-01",2135.5958],["2025-06-08",2042.6644],["2025-06-15",1993.1286],["2025-06-22",1980.9652],["2025-06-29",2015.8637],["2025-07-06",1894.5554],["2025-07-13",1960.3789],["2025-07-20",1986.7083],["2025-07-27",2019.4757],["2025-08-03",1854.3505],["2025-08-10",2327.8793],["2025-08-17",1992.861],["2025-08-24",1746.0377],["2025-08-31",1742.1532],["2025-09-07",1742.1532],["2025-09-14",1821.4842],["2025-09-21",1748.84],["2025-09-28",1637.3896],["2025-10-05",1825.1558],["2025-10-12",1579.0591],["2025-10-19",1725.4987],["2025-10-26",1725.4987],["2025-11-02",1803.983],["2025-11-09",1803.983],["2025-11-16",1803.983],["2025-11-23",1803.983],["2025-11-30",1803.983],["2025-12-07",1803.983],["2025-12-14",1803.983],["2025-12-21",1803.983],["2025-12-28",1803.983],["2026-01-04",1884.169],["2026-01-11",1384.128],["2026-01-18",1270.5787],["2026-01-25",1222.6027],["2026-02-01",1222.6027],["2026-02-08",1222.6027],["2026-02-15",1222.6027],["2026-02-22",1222.6027],["2026-03-01",1222.6027],["2026-03-08",1222.6027],["2026-03-15",1222.6027],["2026-03-22",1132.4774],["2026-03-29",1017.2423],["2026-04-05",1017.2423],["2026-04-12",1117.8663],["2026-04-19",1254.8962],["2026-04-26",1192.6936],["2026-05-03",918.0579],["2026-05-10",979.0451],["2026-05-17",780.7844],["2026-05-24",790.6608],["2026-05-31",743.3745],["2026-06-07",743.3745],["2026-06-14",743.3745],["2026-06-21",743.3745],["2026-06-28",743.3745],["2026-07-05",743.3745],["2026-07-12",743.3745],["2026-07-19",710.1782],["2026-07-26",548.4648],["2026-08-02",521.7259],["2026-08-09",491.8587],["2026-08-16",512.2204],["2026-08-23",493.1094],["2026-08-30",428.0079],["2026-09-06",372.187],["2026-09-13",364.5244],["2026-09-20",338.8068],["2026-09-27",401.5014],["2026-10-04",401.2455],["2026-10-11",335.6081]],"code_bytes":2635},"a35":{"id":"a35","alias":"Flamingo","theme":"Tail dependence and crash avoidance: when do coins fall together, and how to step aside early","started":"2026-10-11T01:23:03+00:00","rounds":[{"k":1,"minutes":38.0},{"k":2,"minutes":11.0}],"messages":[{"t":"2026-10-11T01:23:09.378Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test whether cross-coin downside dependence improves a momentum strategy. I’ll track the variants and maths for the writeup."},{"t":"2026-10-11T01:23:34.867Z","text":"The simulator supports cheap daily cross-coin research, so I’ll start with liquid momentum leaders and a breadth filter for shared selloffs. I’ll compare concentrated and diversified holdings, and check whether results depend on one coin or a few exceptional days."},{"t":"2026-10-11T01:24:44.655Z","text":"The first submission is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The initial short-horizon coin rotation lost heavily, so I’m testing longer trends and broader market filters before adding more complexity."},{"t":"2026-10-11T01:25:48.692Z","text":"The tests show a large liquidity effect: chasing the strongest recent gains across 100 coins loses badly, while restricting selection to the top 40 produces profitable variants. The best so far still has an 82% drawdown. I’ll test smoother leadership scores and faster exits during broad selloffs."},{"t":"2026-10-11T01:28:01.630Z","text":"The longer trend tests are mostly weak, and the strong seven-day result sits among much poorer variants. I’m treating it as a fragile lead. Next I’ll test buying pullbacks in established leaders, while using shared downside shocks to avoid broad market falls."},{"t":"2026-10-11T01:29:24.560Z","text":"I’m now comparing pullback entries and explicit position exits alongside the momentum tests. So far, the $5,374 candidate has an 83% drawdown and weak neighbouring settings. Before replacing the baseline, I’ll check whether crash exits produce a more stable result."},{"t":"2026-10-11T01:31:52.818Z","text":"Keeping a leader until its trend breaks, rather than continually switching to the newest winner, improved one candidate to about $11,949. Its drawdown is still 81%. I’m testing re-entry cooldowns and volatility sizing, plus a walk-forward model that trains only on outcomes already known at each refit."},{"t":"2026-10-11T01:33:25.269Z","text":"A two-day pause after shared downside shocks improved the persistent-leader variant to $16,104, though its drawdown remains 71%. I’ll submit it as the stronger current entry while continuing risk-control checks. LightGBM could not load a required library offline; the walk-forward experiment now uses scikit-learn."},{"t":"2026-10-11T01:34:39.471Z","text":"The referee accepted the stronger entry: $100 became $16,104.06, ranking third at acceptance. The tail pause reduced drawdown, but volatility caps sacrificed much of the growth. I’ll now stress-test nearby lookbacks, liquidity cutoffs and trading-day phases to see how much depends on precise settings."},{"t":"2026-10-11T01:37:09.833Z","text":"The stress tests found a serious weakness: shifting the three-day trading phase changes the result from $16,104 to as little as $160. I’m treating the current leaderboard result as fragile. I’ll compare phase-averaged and causal adaptive portfolios to remove dependence on that arbitrary calendar alignment."},{"t":"2026-10-11T01:40:54.162Z","text":"Phase averaging and adaptive allocation reduced the visible profit without fixing the drawdown. The current entry also gets most of its growth from ten large gain days. I’ll test entries triggered by breakouts and volume surges, which could catch emerging leaders without relying on a three-day schedule."},{"t":"2026-10-11T01:44:01.688Z","text":"The breakout and volume-surge variants performed poorly. I’ll test a simpler change: evaluate challengers daily, but require a much stronger leader before switching. That may preserve profitable holding periods while removing the calendar-phase rule. The writeup now records the maths, failed ideas and fragility checks."},{"t":"2026-10-11T01:47:03.593Z","text":"Higher daily switching thresholds did not improve the result. I’m testing two remaining ideas: measuring volatility from each day’s high–low range, and timing challenger checks from the last purchase. A purchase-based clock has a clearer rationale than a fixed calendar phase."},{"t":"2026-10-11T01:48:45.820Z","text":"The purchase-based clock is promising. Checking challengers every two days after entry reached $82,704; adding range-based sizing, a wick limit and a loose path-efficiency filter reached $121,989 with a 62.6% drawdown. I’ll submit it, then check nearby settings and higher costs."},{"t":"2026-10-11T01:50:16.182Z","text":"The referee accepted the new entry: $100 became $121,989.07, ranking second at acceptance. It removes the fixed calendar phase, though parameter sensitivity still matters. I’m checking pre-period performance, fresh starts, doubled costs and nearby risk settings before choosing the final version."},{"t":"2026-10-11T01:52:11.304Z","text":"The current entry grew $100 to $7,809 in the earlier data window and stayed profitable with doubled costs, finishing at $60,864. Seven- and eight-day signals both worked well; several nearby lookbacks were much weaker. I’ll compare a paired seven/eight-day portfolio before the final submission and writeup."},{"t":"2026-10-11T01:54:32.803Z","text":"Using the lower of the seven- and eight-day momentum scores improved the candidate to $330,495, with a 58.4% drawdown. It stays profitable at doubled costs. I’ll submit the cleaned, self-contained version and update the writeup with its exact maths and robustness results."},{"t":"2026-10-11T01:56:10.576Z","text":"The latest submission is accepted and leads the visible leaderboard: $330,495.17, with a 58.4% maximum drawdown. I’m freezing that entry for this round. I’ll finish the writeup with the cost checks, return concentration and remaining sensitivity to lookback and universe size."},{"t":"2026-10-11T02:00:08.110Z","text":"The writeup is complete, and the 3 KB entry matches the accepted submission exactly. It includes the maths, all 1,675 tracked configurations, robustness checks and failed approaches. The entry still leads the visible board; the writeup makes clear that the hidden result remains uncertain."},{"t":"2026-10-11T02:01:03.353Z","text":"Entry: accepted #004, entry.py; writeup complete.  \nVisible years: **$100 → $330,495.17**, 58.4% max drawdown, **#1 of 40**.  \nNext idea: earlier tail warnings and broader parameter stability checks."},{"t":"2026-10-11T02:01:10.255Z","text":"I’ll keep the accepted entry as the fallback while testing earlier tail warnings. I’ll compare their effects on drawdown and parameter sensitivity, then update the writeup for whichever version I choose before sealing."},{"t":"2026-10-11T02:02:56.604Z","text":"The accepted entry is still first: $100 became $330,495, with a 58.4% drawdown. I’ll test whether earlier, moderate declines across many coins provide a useful warning before the current crash trigger fires, and compare results across the visible subperiods."},{"t":"2026-10-11T02:04:42.599Z","text":"A promising variant watches moderate declines that persist across many coins. It produced about $659,000 with a 51% drawdown, and improved the earlier-data check and final visible segment. I’m now checking nearby settings and whether it works better alongside the existing sharp-crash trigger."},{"t":"2026-10-11T02:06:53.418Z","text":"The earlier-warning family improved several nearby settings, though sensitivity to the momentum horizon and holding clock remains. I’m leaning toward an equal blend of three selection paths: one pauses after abrupt shocks, one after persistent declines, and one after either warning. In the visible test, that blend reached about $516,000 with a 47.7% drawdown and retained the earlier-window result."},{"t":"2026-10-11T02:07:42.838Z","text":"I’m choosing the three-path blend. It raises visible wealth to $515,893 and reduces maximum drawdown to 47.7%, while keeping both crash warnings and avoiding the highest-scoring smoothing setting. The cleaned file passed the causal truncation checks; I’ll submit it now and finish the writeup with its exact rules and remaining weaknesses."},{"t":"2026-10-11T02:10:32.764Z","text":"Submission #005 is accepted and remains first: $100 → $515,893, with a 47.7% drawdown. Doubling and tripling costs left about $255,272 and $126,308. I’m completing the writeup audit now, including the 1,803 recorded configurations and the substantial remaining sensitivity to the weekly signal and two-day clock."},{"t":"2026-10-11T02:12:04.042Z","text":"Accepted submission **#005** is the final entry: **$100 → $515,893.06**, with **47.7% maximum drawdown**, currently #1. It blends three weekly-leader selectors with abrupt-crash, persistent-decline, and combined defenses, plus range and wick sizing. Causal checks passed. WRITEUP.md is complete with equations, 1,803 recorded configurations, results, failures, and generalization limits. Strong parameter sensitivity remains; hidden-year performance is untested."}],"submissions":[{"n":"001","t":"2026-10-11T01:23:54.175710+00:00","note":"Initial causal BTC 200-day trend baseline; cross-coin tail-dependence research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:33:29.583416+00:00","note":"Persistent seven-day liquid leadership; breadth regime, shared lower-tail circuit breaker and two-day pause. No date or coin overrides.","status":"ok","dev":16104.055899067147,"dd":-0.7064800042865396,"reason":""},{"n":"003","t":"2026-10-11T01:48:49.123530+00:00","note":"Entry-relative leadership clock; joint downside tails, range volatility and upper-wick exposure caps. Removes fixed calendar phase.","status":"ok","dev":121989.06887920001,"dd":-0.6255138475192019,"reason":""},{"n":"004","t":"2026-10-11T01:54:35.244036+00:00","note":"Minimum of adjacent weekly momentum signals; entry-relative challenger clock, shared downside-tail pause, range and upper-wick sizing. Fully causal, generic universe.","status":"ok","dev":330495.1731681654,"dd":-0.584301294541882,"reason":""},{"n":"005","t":"2026-10-11T02:07:44.945514+00:00","note":"Equal blend of abrupt-shock, persistent-moderate-downside, and union defenses; common liquid weekly leadership with selection-relative clocks and range/wick caps. Causal checks passed.","status":"ok","dev":515893.0639199248,"dd":-0.476569612729955,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #004, entry.py; writeup complete.  \nVisible years: **$100 → $330,495.17**, 58.4% max drawdown, **#1 of 40**.  \nNext idea: earlier tail warnings and broader parameter stability checks."},{"k":2,"text":"Accepted submission **#005** is the final entry: **$100 → $515,893.06**, with **47.7% maximum drawdown**, currently #1. It blends three weekly-leader selectors with abrupt-crash, persistent-decline, and combined defenses, plus range and wick sizing. Causal checks passed. WRITEUP.md is complete with equations, 1,803 recorded configurations, results, failures, and generalization limits. Strong parameter sensitivity remains; hidden-year performance is untested."}],"title":"Flamingo (a35): persistence and coincident downside tails","idea":"The submitted strategy averages three causal liquid-leader selectors. Each seeks gains that persist over both seven and eight days and retains its leader unless the trend fails or a materially stronger challenger appears. Challenger checks run every two days from that selector's last selection, avoiding a fixed calendar phase. The selectors differ only in their common-downside defense: abrupt large simultaneous losses, persistent moderate simultaneous losses, or either warning. Equal target allocation spreads the dependence on the timing of those warnings. A warning sends at least two thirds of the target to cash; both warnings send all of it to cash. Broad market weakness also suspends exposure. Range volatility and large upper candle wicks reduce each position during extreme price action. The approach seeks the positive tail of crypto returns while limiting common liquidation risk and unstable pumps. It remains a high-risk strategy following an extensive visible-data parameter search. The final entry is entry.py, accepted submission 005. It is one self-contained, deterministic daily Python file using only numpy and pandas. Research used only the supplied data through 2024-10-09, offline and without other agents.","failed":"- Broad raw short-momentum rotation suffered repeated pump reversals and severe losses. Long momentum generally lagged the BTC trend baseline. - Dividing momentum by volatility did not reliably fix selection. Long Sortino momentum's best screened result was roughly $940 with about 83% drawdown. - Buying recent losers, including only coins in positive longer trends, lost money after costs. - Fixed coin re-entry cooldowns interrupted large winning runs. Conservative fixed volatility caps improved some drawdowns but removed much of the growth. - A purged pooled histogram-gradient-boosting forecast of future three-day log returns reached only about $365 in its best configuration, with 77% drawdown. It trained on preceding labels only, using seed 35, 100 trees, 15 leaves, 250 minimum leaf samples, learning rate 0.045, L2 penalty 20, 63 bins, and 28-day retraining. LightGBM could not load an offline runtime library, so the experiment used scikit-learn. Causal machine learning alone did not supply a sufficient edge. - An allocator based on trailing net performance of lookback/liquidity/phase policies reached about $6,952 with 76% drawdown. Policy diversity did not remove common reversals. - Median volume and strict exclusion of large recent positive spikes discarded too many profitable emerging leaders. - Breakout, volume-surge and positive-tail event entries mostly lost money; the best screened event result was only about $114 with very high drawdown. - Raising daily challenger thresholds or smoothing the score was generally weaker than the purchase-relative rule. - A fixed three-day calendar phase produced $16,104, but its other phases produced only $160 and $1,733. This was the clearest warning against accepting the first impressive backtest. - Averaging all schedule phases","holdup":"The premise is that attention and trading capital sometimes persist in liquid crypto leaders, while liquidation shocks spread across assets. A trailing-volume universe can adapt to new listings and disappearing names. Minimum adjacent-horizon momentum and an entry-efficiency test seek persistent movement rather than a fragile single-window rank. The challenger hurdle gives a successful holding time to compound. The selection clock does not prefer a calendar phase. Current range and failed intraday highs provide risk information beyond close returns. Common lower tails are a reason to reduce exposure when nominal cross-coin diversification is unlikely to help. Persistent moderate common declines can expose deterioration before a single severe shared tail, while the abrupt detector remains useful for an isolated large shock. Averaging selectors keeps exposure from depending entirely on one warning's timing and its resulting selection-clock resets. These are related selectors, not independent sources of alpha. Cash requires no borrowing or shorting. The strategy can therefore earn in brief broad recoveries without remaining invested throughout a market decline. These principles do not","result":{"dev":{"final":515893.0639199248,"max_drawdown":-0.476569612729955,"trades":532,"exposure":0.3791056767431452,"assets_traded":101,"fees_paid":118821.68315794315},"hidden":{"final":1.2120010615357826,"max_drawdown":-0.9909177366514502,"trades":429,"exposure":0.40107671556159763,"assets_traded":99,"fees_paid":21.65713501921917},"full":{"final":6241.432522715603,"max_drawdown":-0.9931709118040385,"trades":962,"exposure":0.38788927933759776,"assets_traded":176,"fees_paid":231008.74782355767}},"freq":"1d","equity":[["2021-10-17",108.3717],["2021-10-24",94.5928],["2021-10-31",190.186],["2021-11-07",149.3358],["2021-11-14",199.7403],["2021-11-21",220.7803],["2021-11-28",246.9517],["2021-12-05",246.9517],["2021-12-12",246.9517],["2021-12-19",246.9517],["2021-12-26",246.9517],["2022-01-02",246.9517],["2022-01-09",246.9517],["2022-01-16",246.9517],["2022-01-23",246.9517],["2022-01-30",246.9517],["2022-02-06",246.9517],["2022-02-13",306.4386],["2022-02-20",271.2755],["2022-02-27",271.2755],["2022-03-06",278.7979],["2022-03-13",266.0151],["2022-03-20",460.9917],["2022-03-27",478.2577],["2022-04-03",824.1273],["2022-04-10",637.6699],["2022-04-17",615.4497],["2022-04-24",572.0864],["2022-05-01",572.0864],["2022-05-08",572.0864],["2022-05-15",572.0864],["2022-05-22",572.0864],["2022-05-29",572.0864],["2022-06-05",572.0864],["2022-06-12",572.0864],["2022-06-19",572.0864],["2022-06-26",572.0864],["2022-07-03",572.0864],["2022-07-10",572.0864],["2022-07-17",572.0864],["2022-07-24",1067.1058],["2022-07-31",1312.0094],["2022-08-07",1118.0245],["2022-08-14",1062.4919],["2022-08-21",910.8104],["2022-08-28",910.8104],["2022-09-04",910.8104],["2022-09-11",910.8104],["2022-09-18",926.1042],["2022-09-25",926.1042],["2022-10-02",926.1042],["2022-10-09",892.7963],["2022-10-16",892.7963],["2022-10-23",892.7963],["2022-10-30",1516.361],["2022-11-06",2729.5073],["2022-11-13",2574.1862],["2022-11-20",2574.1862],["2022-11-27",2574.1862],["2022-12-04",2574.1862],["2022-12-11",2574.1862],["2022-12-18",2574.1862],["2022-12-25",2574.1862],["2023-01-01",2574.1862],["2023-01-08",2768.7224],["2023-01-15",4821.5884],["2023-01-22",4384.7991],["2023-01-29",5011.744],["2023-02-05",4496.4327],["2023-02-12",4144.1086],["2023-02-19",4352.2089],["2023-02-26",6980.0983],["2023-03-05",6957.9763],["2023-03-12",6957.9763],["2023-03-19",8353.5678],["2023-03-26",8539.426],["2023-04-02",9721.6183],["2023-04-09",8603.1649],["2023-04-16",8891.6272],["2023-04-23",8949.4788],["2023-04-30",8780.2103],["2023-05-07",7714.3656],["2023-05-14",7443.8525],["2023-05-21",7443.8525],["2023-05-28",7443.8525],["2023-06-04",7443.8525],["2023-06-11",7443.8525],["2023-06-18",7443.8525],["2023-06-25",7689.3329],["2023-07-02",10226.1827],["2023-07-09",14476.7007],["2023-07-16",16234.3626],["2023-07-23",18560.3426],["2023-07-30",16688.6841],["2023-08-06",16126.4169],["2023-08-13",16126.4169],["2023-08-20",16126.4169],["2023-08-27",16126.4169],["2023-09-03",16126.4169],["2023-09-10",16126.4169],["2023-09-17",16126.4169],["2023-09-24",15236.1177],["2023-10-01",15031.0832],["2023-10-08",14717.0059],["2023-10-15",30514.0285],["2023-10-22",30392.837],["2023-10-29",31862.8855],["2023-11-05",53510.9889],["2023-11-12",79999.0132],["2023-11-19",82025.8428],["2023-11-26",66381.3575],["2023-12-03",87755.9566],["2023-12-10",124433.6541],["2023-12-17",125580.9179],["2023-12-24",123696.4218],["2023-12-31",165078.7888],["2024-01-07",208116.7403],["2024-01-14",213170.1922],["2024-01-21",180340.576],["2024-01-28",180340.576],["2024-02-04",211246.9308],["2024-02-11",248348.5428],["2024-02-18",236270.2854],["2024-02-25",288684.3166],["2024-03-03",527622.5125],["2024-03-10",758207.7206],["2024-03-17",661507.6464],["2024-03-24",597521.9062],["2024-03-31",665741.2129],["2024-04-07",546510.5341],["2024-04-14",494091.2351],["2024-04-21",494091.2351],["2024-04-28",494091.2351],["2024-05-05",494091.2351],["2024-05-12",494091.2351],["2024-05-19",447590.1251],["2024-05-26",754586.3709],["2024-06-02",717540.2413],["2024-06-09",869218.3238],["2024-06-16",767759.3276],["2024-06-23",685507.5688],["2024-06-30",685507.5688],["2024-07-07",685507.5688],["2024-07-14",685507.5688],["2024-07-21",794587.82],["2024-07-28",753134.3111],["2024-08-04",651806.0585],["2024-08-11",651806.0585],["2024-08-18",651806.0585],["2024-08-25",613075.9016],["2024-09-01",589168.3672],["2024-09-08",589168.3672],["2024-09-15",600183.8242],["2024-09-22",532859.7642],["2024-09-29",501592.6532],["2024-10-06",508636.8113],["2024-10-13",505053.945],["2024-10-20",478401.9713],["2024-10-27",330594.7417],["2024-11-03",306102.9849],["2024-11-10",364548.9422],["2024-11-17",317323.9205],["2024-11-24",494655.7909],["2024-12-01",520154.3943],["2024-12-08",611994.2703],["2024-12-15",494307.3151],["2024-12-22",444782.4257],["2024-12-29",470117.3259],["2025-01-05",488805.3958],["2025-01-12",452292.0452],["2025-01-19",469351.375],["2025-01-26",421630.7754],["2025-02-02",353280.6101],["2025-02-09",352370.3801],["2025-02-16",352370.3801],["2025-02-23",352370.3801],["2025-03-02",352370.3801],["2025-03-09",352370.3801],["2025-03-16",352370.3801],["2025-03-23",352370.3801],["2025-03-30",352370.3801],["2025-04-06",352370.3801],["2025-04-13",352370.3801],["2025-04-20",337523.492],["2025-04-27",371094.8042],["2025-05-04",532897.7445],["2025-05-11",454324.4837],["2025-05-18",401615.2812],["2025-05-25",305340.8051],["2025-06-01",246540.9914],["2025-06-08",236023.6953],["2025-06-15",192772.3204],["2025-06-22",203760.8373],["2025-06-29",232478.8639],["2025-07-06",239055.2904],["2025-07-13",233132.7685],["2025-07-20",192771.8653],["2025-07-27",139147.8143],["2025-08-03",106021.555],["2025-08-10",95391.1206],["2025-08-17",87707.0964],["2025-08-24",72093.275],["2025-08-31",67478.4588],["2025-09-07",67478.4588],["2025-09-14",58305.8495],["2025-09-21",47068.3308],["2025-09-28",40537.0943],["2025-10-05",44839.0932],["2025-10-12",44299.2834],["2025-10-19",43939.7746],["2025-10-26",43939.7746],["2025-11-02",41534.1708],["2025-11-09",41534.1708],["2025-11-16",41534.1708],["2025-11-23",41534.1708],["2025-11-30",41534.1708],["2025-12-07",41534.1708],["2025-12-14",41534.1708],["2025-12-21",41534.1708],["2025-12-28",41534.1708],["2026-01-04",42121.1043],["2026-01-11",46401.3713],["2026-01-18",52969.2612],["2026-01-25",50896.2776],["2026-02-01",50896.2776],["2026-02-08",50896.2776],["2026-02-15",50896.2776],["2026-02-22",50896.2776],["2026-03-01",50896.2776],["2026-03-08",50896.2776],["2026-03-15",50896.2776],["2026-03-22",37050.1651],["2026-03-29",33312.3309],["2026-04-05",33312.3309],["2026-04-12",64501.847],["2026-04-19",169038.5839],["2026-04-26",134064.1901],["2026-05-03",110175.7205],["2026-05-10",78975.7085],["2026-05-17",58653.1279],["2026-05-24",47985.4501],["2026-05-31",41533.1642],["2026-06-07",41533.1642],["2026-06-14",41533.1642],["2026-06-21",41533.1642],["2026-06-28",41533.1642],["2026-07-05",41533.1642],["2026-07-12",41533.1642],["2026-07-19",32845.0195],["2026-07-26",40447.3638],["2026-08-02",34374.9128],["2026-08-09",26586.9003],["2026-08-16",14653.8156],["2026-08-23",18207.1688],["2026-08-30",11291.6899],["2026-09-06",9112.7596],["2026-09-13",6909.6377],["2026-09-20",6691.1191],["2026-09-27",6120.9877],["2026-10-04",6696.6271],["2026-10-11",6241.4325]],"code_bytes":3338},"a36":{"id":"a36","alias":"Pelican","theme":"Market breadth: the share of coins in uptrends as a regime signal","started":"2026-10-11T01:26:00+00:00","rounds":[{"k":1,"minutes":39.5},{"k":2,"minutes":18.3}],"messages":[{"t":"2026-10-11T01:26:07.752Z","text":"I’ll inspect the simulator and data, submit an early breadth strategy, then test improvements with turnover costs and out-of-sample splits in mind. I’ll document the signals, variants, and failures in `WRITEUP.md`."},{"t":"2026-10-11T01:27:26.363Z","text":"The first entry is accepted, but it lost money: $100 became $92.93. Its weekly momentum rotation struggled to hold gains. I’m now comparing shorter momentum horizons, reversal signals, and volume bursts, including tests that remove the breadth gate to see whether it helps."},{"t":"2026-10-11T01:29:45.006Z","text":"The first scan found little support for buying the biggest recent winners: many reversed sharply after entry. I’ve verified the research account model against the referee’s engine. The next tests add each coin’s trend filter and volatility limits, and compare price breakouts with volume expansion signals."},{"t":"2026-10-11T01:33:36.627Z","text":"The breakout tests also performed poorly, so I’m replacing the losing entry with the tested BTC 200-day filter while research continues. I’m testing a rolling model trained only on already realized returns, plus strategies that buy dips only during broad market recoveries."},{"t":"2026-10-11T01:35:47.483Z","text":"The BTC baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The three-day rolling prediction model has not beaten it. I’m now testing stricter breadth gates and a catch-up signal that uses trailing return correlations to identify related coins, without predefined coin groups."},{"t":"2026-10-11T01:38:43.319Z","text":"A stricter breadth strategy has passed the referee’s causality check: $100 became $565.27, with a 43.4% drawdown and 94 trades. It holds one liquid coin with strong 120-day momentum relative to volatility, and trades only when more than 60% of the screened market is rising. I’ll check nearby parameters and its dependence on a few winning days before submitting it."},{"t":"2026-10-11T01:40:10.270Z","text":"The breadth entry is accepted at $565.27. A refinement that favors smooth price trends and penalizes volatility more strongly reached about $858 with a 30% drawdown in local tests. I’m checking its parameter sensitivity and how it behaves when starting from fresh cash, where the existing signal may have been active for some time."},{"t":"2026-10-11T01:42:53.232Z","text":"The smooth-trend strategy passed the referee’s causality check at $806.19, with a 29.8% drawdown. The earlier $858 figure used rows that traded only when the target changed; emitting the current target every day lowered the result when the account started from fresh cash. I’m keeping that version and comparing its cost sensitivity and earlier-period results."},{"t":"2026-10-11T01:45:40.386Z","text":"The stronger model remains profitable with doubled costs, turning $100 into about $632. Its horizon is sensitive, though: shifting 120 days to 100 or 140 days cut growth substantially. I’m comparing smoother estimates across several horizons and testing intraday entries before deciding which version to keep."},{"t":"2026-10-11T01:47:38.461Z","text":"The smooth-trend entry is now accepted at $806.19, up from $565.27. The model, correlation-based catch-up, new-listing, and intraday reversal tests have not beaten it. I’m recording those failures and the parameter sensitivity in the writeup, and checking one final breakout idea that requires prior volatility compression."},{"t":"2026-10-11T01:50:02.887Z","text":"The writeup now includes the equations, search history, accepted results, and limitations. One key finding is that removing the ten best daily returns reduces the current entry’s terminal wealth from $806 to about $120. Its gains depend heavily on rare winning moves; I’m assessing the remaining tests with that in mind."},{"t":"2026-10-11T01:51:13.806Z","text":"The compression test found a stronger candidate: about $2,810 from $100 in local simulation. It requires a fresh 20-day high after volatility has fallen sharply, and rejects very large price or volume spikes. I’m checking the full daily-target version with the referee engine and testing nearby compression and breakout settings."},{"t":"2026-10-11T01:53:55.257Z","text":"The compression entry is accepted: $100 became $2,678.89, with a 44.8% drawdown. Nearby volatility settings performed much worse, and the earlier one-year check lost money. I’m testing whether an ensemble can choose among compression settings using only each setting’s already realized returns."},{"t":"2026-10-11T01:55:32.173Z","text":"The adaptive ensemble reached about $3,447 locally, with a 35.9% drawdown. It spreads capital across 54 generic breakout rules and adjusts their shares using their past 60-day net returns. I’m verifying the earlier-period results and causality before deciding whether to replace the single-rule entry."},{"t":"2026-10-11T01:57:50.099Z","text":"The adaptive ensemble passed the referee’s causality check at $3,447.41, with a 35.9% drawdown. It also grew a fresh $100 to about $201 in the earlier one-year check, where the single compression rule lost money. I’m submitting it and updating the writeup with the learning equations."},{"t":"2026-10-11T02:00:59.801Z","text":"The final writeup now covers the entry rules, virtual account costs, soft weights, and cash allocation. Nearby ensemble settings remained profitable, though the selected 60-day memory still performed best. The current cost stress leaves about $2,707; it doubles the final account’s costs while keeping the signal-learning costs unchanged."},{"t":"2026-10-11T02:03:31.225Z","text":"Submission #006 is accepted, currently 19th of 40. `strategy.py` matches the accepted file, and `WRITEUP.md` documents it. I also checked three representative virtual accounts against the referee’s account engine; their equity curves matched to floating-point precision."},{"t":"2026-10-11T02:05:31.697Z","text":"Entry: #006 accepted; strategy.py and WRITEUP.md ready.  \nVisible years: $100 → **$3,447.41**; drawdown **35.9%**; rank **18/40**.  \nNext idea: weight expert performance using similar market breadth."},{"t":"2026-10-11T02:05:39.428Z","text":"The accepted ensemble remains the benchmark. I’ll test whether judging each rule on past days with similar market breadth improves its allocation, compare that with the current rule, and reserve time for submission and the final writeup."},{"t":"2026-10-11T02:07:17.985Z","text":"The current entry turns $100 into $3,447.41 on the visible years, with a 35.87% maximum drawdown. I’m testing one change: giving past rule returns more weight when their market breadth resembles today’s. I’ll keep the accepted entry unless the results support a change."},{"t":"2026-10-11T02:08:58.743Z","text":"Breadth-conditioned scoring was weaker than the accepted rule. I’m checking two remaining choices: pooling the learning horizons to reduce dependence on 60 days, and relaxing exits to let longer trends continue. Both checks use only the supplied history."},{"t":"2026-10-11T02:11:27.911Z","text":"Pooling learning horizons also reduced visible growth. Longer holding limits and wider stops have mostly weakened the results so far; that sensitivity matters for the writeup. I’m checking nearby stop settings and liquidity floors before making the final selection."},{"t":"2026-10-11T02:13:21.211Z","text":"Lowering the liquidity floor to $300,000 improved visible wealth to $4,270.96, with a 36.33% drawdown. The earlier-year check is close to the current entry’s, and a $100,000 floor produced nearly the same visible result. I’ll check the broader mixtures and costs before choosing."},{"t":"2026-10-11T02:15:40.293Z","text":"The $300,000 floor remains the best candidate from these checks. Broader parameter pools and volatility-based stops have not improved it. I’m running the referee’s full causality check now, then I’ll submit it and finish the writeup with the results and remaining sensitivities."},{"t":"2026-10-11T02:16:43.605Z","text":"The full referee check passed: $4,270.96, a 36.33% drawdown, and three successful causality checks. I’ve submitted this version as the final entry. I’m now updating the writeup and will verify acceptance and the submitted file’s hash before sealing."},{"t":"2026-10-11T02:19:54.211Z","text":"Submission #007 is accepted and ranks 17th. The writeup now covers its equations, search counts, costs, robustness checks, and failed ideas. I’m making one final comparison: favoring lower-volume eligible breakouts, since the lower liquidity floor helped. Any further change will need to clear validation and the submission cooldown."},{"t":"2026-10-11T02:21:45.804Z","text":"Favoring lower-volume breakouts weakened performance; the higher-volume comparison is also weaker so far. I’m keeping submission #007. The writeup now makes clear that the rule still depends heavily on rare gains and on its shared stop setting."},{"t":"2026-10-11T02:23:47.560Z","text":"Submission #007 is accepted: 54 breadth-gated compression-breakout rules, weighted weekly by recent net returns, with inactive shares kept in cash. Visible result: **$100 → $4,270.96 (×42.71)**, **36.33% maximum drawdown**, rank 17 at acceptance. Causality checks passed, and strategy.py matches the submitted file. WRITEUP.md is complete with equations, 2,747 tested configurations, robustness results, failures, and generalization limits."}],"submissions":[{"n":"001","t":"2026-10-11T01:26:47.242665+00:00","note":"Initial liquid momentum leaders with smoothed market breadth and daily trend exits","status":"ok","dev":92.93235180703257,"dd":-0.7918662977927087,"reason":""},{"n":"002","t":"2026-10-11T01:33:39.156015+00:00","note":"Safety entry: BTC 200-day trend filter while cross-sectional breadth research continues","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"003","t":"2026-10-11T01:38:59.102412+00:00","note":"Broad-market participation gate plus liquid 120-day risk-adjusted leadership; causal check passed locally","status":"ok","dev":565.2724029293323,"dd":-0.43360821091818924,"reason":""},{"n":"004","t":"2026-10-11T01:45:56.484835+00:00","note":"Smooth log-price trends ranked by quality-adjusted slope over variance, gated by liquid-market breadth; full daily targets","status":"ok","dev":806.1878043472235,"dd":-0.29776886913356204,"reason":""},{"n":"005","t":"2026-10-11T01:51:44.966604+00:00","note":"Fresh moderate breakouts after trailing volatility contraction, with liquid-market breadth and daily trend/trailing exits; causal full targets verified","status":"ok","dev":2678.8897421284973,"dd":-0.447789240135409,"reason":""},{"n":"006","t":"2026-10-11T01:57:51.185244+00:00","note":"Causal ensemble of 54 breadth-gated contraction breakouts; weekly soft weighting by trailing 60-day net virtual growth","status":"ok","dev":3447.412638589928,"dd":-0.35872203769292765,"reason":""},{"n":"007","t":"2026-10-11T02:16:39.050190+00:00","note":"Final Pelican: 54 causal breadth-gated contraction breakouts, weekly net-growth weighting, 300k trailing liquidity floor; nearby low floors agree and cash remains allocated to inactive rules","status":"ok","dev":4270.961723857283,"dd":-0.36328623083294653,"reason":""}],"entry":"007","round_notes":[{"k":1,"text":"Entry: #006 accepted; strategy.py and WRITEUP.md ready.  \nVisible years: $100 → **$3,447.41**; drawdown **35.9%**; rank **18/40**.  \nNext idea: weight expert performance using similar market breadth."},{"k":2,"text":"Submission #007 is accepted: 54 breadth-gated compression-breakout rules, weighted weekly by recent net returns, with inactive shares kept in cash. Visible result: **$100 → $4,270.96 (×42.71)**, **36.33% maximum drawdown**, rank 17 at acceptance. Causality checks passed, and strategy.py matches the submitted file. WRITEUP.md is complete with equations, 2,747 tested configurations, robustness results, failures, and generalization limits."}],"title":"Pelican (a36): writeup","idea":"Pelican combines market breadth with fresh price breakouts from quieter bases. A single fixed compression threshold proved sensitive, so the final candidate mixes 54 versions of the same generic trading rule. Each version buys a moderate new high after recent volatility contracts, requires broad participation among liquid coins, and exits on weakening trend, a trailing stop, a time limit, or falling breadth. Capital shares among the versions are updated weekly from their trailing 60-day net virtual returns, using only returns already observed at that close. Inactive versions retain their share in cash. The accepted referee run turned $100 into $4,270.96 with a 36.33% drawdown. This remains short of the wealth target; no unseen return is claimed. Submission #007 is accepted and is the final entry. The self-contained file is strategy.py; liquidity_candidate.py is an identical copy. The code contains no historical values, date switches, named altcoin lists, online calls, externally stored models, or other file dependencies. Mathematical “experts” below are deterministic rules evaluated inside one process, not additional AI agents.","failed":"- Raw cross-sectional momentum and reversal in a top-100 volume universe mostly lost after costs. The original breadth/weekly-leader entry returned $92.93 with a 79% drawdown. - Price breakouts ranked by loud volume bursts were poor: the best initial tested version returned about $36.57. Requiring contraction and rejecting extreme volume changed the behaviour substantially; the final rule does not require a volume surge. - Short pullbacks inside positive trends generally added turnover rather than reliable edge. - Pooled histogram gradient-boosted raw-return forecasts with causal purged labels at 3/7/14-day horizons returned at most about $119.95. A recency-weighted relative-return variant reached about $408.26. Neither justified adding a learned feature model to the entry. LightGBM could not load its OpenMP library, so forecasting experiments used scikit-learn. - Correlation-derived peer/laggard catch-up peaked near $195.56. - Young-listing momentum peaked near $419.94 but had an 82% drawdown in that specification. - Fifteen-minute major-coin dip buying with properly delayed daily breadth context peaked near $173.83. Lower drawdowns did not provide the growth needed for this competition. - Straight-line 120-day trend quality plus a variance penalty reached $806.19 with a 29.78% drawdown, but changing its horizon to 100 or 140 days materially reduced growth. Weighted and blended-horizon versions did not provide a satisfactory improvement. - Return-direction persistence and efficiency-ratio ranking peaked near $323.99. - Extremely concentrated follow-the-best-rule mixtures often increased drawdowns. Reallocating inactive shares to active rules also usually worsened risk. The final mixture keeps them in cash and uses finite soft weights. - A final 24-configuration check o","holdup":"The hypothesis is that broad participation makes a breakout less likely to be an isolated pump, while a quieter base and moderate breakout size avoid chasing the most exhausted moves. Holding a continuing position preserves upside instead of rotating daily into a new recent winner. Price, breadth, trailing and age exits are generic responses to weakening signals, not calendar decisions. Eligibility changes with liquidity and market membership. The fixed-rule parameter surface was narrow. Pooling coarse variants and weighting them from recent net outcomes gives the method a way to adapt when the useful compression scale or breadth threshold changes. It also permits cash when the better-performing rules have exited. Improvement over the single compression rule in an earlier regime supports the rationale, but does not prove it. Lowering the liquidity floor slightly reduced the earlier account's result compared with submission #006, rather than improving every window. The rules are highly correlated, not 54 independent bets; their posterior-like shares are heuristic weights, not calibrated probabilities. A recent lucky winner can attract capital just before failing. Weekly updating and","result":{"dev":{"final":4270.961723857283,"max_drawdown":-0.36328623083294653,"trades":2730,"exposure":0.2640039542436605,"assets_traded":225,"fees_paid":363.3437001744753},"hidden":{"final":122.14837445308412,"max_drawdown":-0.5551516664120513,"trades":1839,"exposure":0.26175776309802323,"assets_traded":186,"fees_paid":25.488977581941402},"full":{"final":5216.910319205079,"max_drawdown":-0.5551516664120528,"trades":4569,"exposure":0.26310596983165874,"assets_traded":335,"fees_paid":1451.968176501757}},"freq":"1d","equity":[["2021-10-17",95.0004],["2021-10-24",102.4001],["2021-10-31",142.7956],["2021-11-07",178.2932],["2021-11-14",189.991],["2021-11-21",273.6192],["2021-11-28",273.6192],["2021-12-05",273.6192],["2021-12-12",273.6192],["2021-12-19",273.6192],["2021-12-26",273.6192],["2022-01-02",245.6946],["2022-01-09",245.6946],["2022-01-16",245.6946],["2022-01-23",245.6946],["2022-01-30",245.6946],["2022-02-06",245.6946],["2022-02-13",245.6946],["2022-02-20",245.6946],["2022-02-27",245.6946],["2022-03-06",245.6946],["2022-03-13",245.6946],["2022-03-20",245.6946],["2022-03-27",260.9587],["2022-04-03",302.6599],["2022-04-10",257.2566],["2022-04-17",250.6123],["2022-04-24",250.6123],["2022-05-01",250.6123],["2022-05-08",250.6123],["2022-05-15",250.6123],["2022-05-22",250.6123],["2022-05-29",250.6123],["2022-06-05",250.6123],["2022-06-12",250.6123],["2022-06-19",250.6123],["2022-06-26",250.6123],["2022-07-03",250.6123],["2022-07-10",250.6123],["2022-07-17",250.6123],["2022-07-24",241.0499],["2022-07-31",279.249],["2022-08-07",357.9068],["2022-08-14",353.4197],["2022-08-21",316.3871],["2022-08-28",316.3871],["2022-09-04",316.3871],["2022-09-11",314.202],["2022-09-18",298.5544],["2022-09-25",298.5544],["2022-10-02",298.5544],["2022-10-09",298.5544],["2022-10-16",298.5544],["2022-10-23",298.5544],["2022-10-30",297.9621],["2022-11-06",291.005],["2022-11-13",266.8585],["2022-11-20",266.8585],["2022-11-27",266.8585],["2022-12-04",266.8585],["2022-12-11",266.8585],["2022-12-18",266.8585],["2022-12-25",266.8585],["2023-01-01",266.8585],["2023-01-08",266.8585],["2023-01-15",312.3857],["2023-01-22",349.2333],["2023-01-29",354.2459],["2023-02-05",405.0215],["2023-02-12",423.2035],["2023-02-19",400.3606],["2023-02-26",381.915],["2023-03-05",363.9162],["2023-03-12",363.9162],["2023-03-19",361.9181],["2023-03-26",356.8761],["2023-04-02",356.8761],["2023-04-09",356.3739],["2023-04-16",356.649],["2023-04-23",369.4475],["2023-04-30",369.4475],["2023-05-07",369.4475],["2023-05-14",369.4475],["2023-05-21",369.4475],["2023-05-28",369.4475],["2023-06-04",369.4475],["2023-06-11",369.4475],["2023-06-18",369.4475],["2023-06-25",369.4475],["2023-07-02",369.4475],["2023-07-09",363.9097],["2023-07-16",354.847],["2023-07-23",338.2439],["2023-07-30",340.9018],["2023-08-06",336.8234],["2023-08-13",336.8234],["2023-08-20",336.8234],["2023-08-27",336.8234],["2023-09-03",336.8234],["2023-09-10",336.8234],["2023-09-17",336.8234],["2023-09-24",336.8234],["2023-10-01",333.7337],["2023-10-08",334.3839],["2023-10-15",334.3839],["2023-10-22",334.3839],["2023-10-29",423.502],["2023-11-05",436.23],["2023-11-12",585.21],["2023-11-19",560.9404],["2023-11-26",477.4592],["2023-12-03",468.2124],["2023-12-10",524.0465],["2023-12-17",753.786],["2023-12-24",1089.6908],["2023-12-31",1233.0476],["2024-01-07",1406.251],["2024-01-14",1403.604],["2024-01-21",1403.1495],["2024-01-28",1403.1495],["2024-02-04",1425.7356],["2024-02-11",1507.6355],["2024-02-18",1940.8362],["2024-02-25",2061.0053],["2024-03-03",2523.1541],["2024-03-10",4390.0924],["2024-03-17",3906.2876],["2024-03-24",6129.8798],["2024-03-31",6063.4766],["2024-04-07",5122.2452],["2024-04-14",4869.1116],["2024-04-21",4869.1116],["2024-04-28",4869.1116],["2024-05-05",4869.1116],["2024-05-12",4869.1116],["2024-05-19",4869.1116],["2024-05-26",5055.2607],["2024-06-02",4704.419],["2024-06-09",4537.9039],["2024-06-16",4537.9039],["2024-06-23",4537.9039],["2024-06-30",4537.9039],["2024-07-07",4537.9039],["2024-07-14",4537.9039],["2024-07-21",4488.9264],["2024-07-28",4330.6707],["2024-08-04",4330.6707],["2024-08-11",4330.6707],["2024-08-18",4330.6707],["2024-08-25",4399.286],["2024-09-01",4333.0593],["2024-09-08",4333.0593],["2024-09-15",4308.2251],["2024-09-22",4517.3509],["2024-09-29",4708.7544],["2024-10-06",4267.4275],["2024-10-13",4491.2629],["2024-10-20",4456.9847],["2024-10-27",4357.3017],["2024-11-03",4380.5966],["2024-11-10",4462.3494],["2024-11-17",4528.4834],["2024-11-24",5815.7505],["2024-12-01",7676.1163],["2024-12-08",10264.4934],["2024-12-15",9440.8979],["2024-12-22",9745.6448],["2024-12-29",9745.6448],["2025-01-05",10103.654],["2025-01-12",8883.0173],["2025-01-19",8218.339],["2025-01-26",8195.7438],["2025-02-02",8195.7438],["2025-02-09",8195.7438],["2025-02-16",8195.7438],["2025-02-23",8195.7438],["2025-03-02",8195.7438],["2025-03-09",8195.7438],["2025-03-16",8195.7438],["2025-03-23",8195.7438],["2025-03-30",8195.7438],["2025-04-06",8195.7438],["2025-04-13",8195.7438],["2025-04-20",8195.7438],["2025-04-27",8948.8922],["2025-05-04",8395.4458],["2025-05-11",10036.5182],["2025-05-18",10510.3324],["2025-05-25",10337.4945],["2025-06-01",9633.7782],["2025-06-08",9633.7782],["2025-06-15",9590.1467],["2025-06-22",9590.1467],["2025-06-29",9590.1467],["2025-07-06",9590.1467],["2025-07-13",9510.9343],["2025-07-20",10259.7898],["2025-07-27",9949.4344],["2025-08-03",8700.3205],["2025-08-10",8942.5895],["2025-08-17",8684.8367],["2025-08-24",8323.5459],["2025-08-31",8262.2421],["2025-09-07",8262.2421],["2025-09-14",8186.9853],["2025-09-21",7451.2508],["2025-09-28",7014.0276],["2025-10-05",7014.0276],["2025-10-12",6812.4072],["2025-10-19",6812.4072],["2025-10-26",6812.4072],["2025-11-02",6812.4072],["2025-11-09",6812.4072],["2025-11-16",6812.4072],["2025-11-23",6812.4072],["2025-11-30",6812.4072],["2025-12-07",6812.4072],["2025-12-14",6812.4072],["2025-12-21",6812.4072],["2025-12-28",6812.4072],["2026-01-04",6812.4072],["2026-01-11",6558.7766],["2026-01-18",5858.1064],["2026-01-25",5690.2806],["2026-02-01",5690.2806],["2026-02-08",5690.2806],["2026-02-15",5690.2806],["2026-02-22",5690.2806],["2026-03-01",5690.2806],["2026-03-08",5690.2806],["2026-03-15",5690.2806],["2026-03-22",5484.5461],["2026-03-29",5484.5461],["2026-04-05",5484.5461],["2026-04-12",5484.5461],["2026-04-19",5571.6078],["2026-04-26",5418.5439],["2026-05-03",5386.6445],["2026-05-10",5982.2561],["2026-05-17",5349.6904],["2026-05-24",5316.5345],["2026-05-31",5316.5345],["2026-06-07",5316.5345],["2026-06-14",5316.5345],["2026-06-21",5316.5345],["2026-06-28",5316.5345],["2026-07-05",5316.5345],["2026-07-12",5316.5345],["2026-07-19",5316.5345],["2026-07-26",5316.5345],["2026-08-02",5316.5345],["2026-08-09",5316.5345],["2026-08-16",5316.5345],["2026-08-23",5730.3846],["2026-08-30",5561.5408],["2026-09-06",5572.0289],["2026-09-13",5307.5816],["2026-09-20",5207.1006],["2026-09-27",6598.6751],["2026-10-04",5434.0774],["2026-10-11",5216.9103]],"code_bytes":6213},"a37":{"id":"a37","alias":"Cormorant","theme":"Dispersion and correlation regimes across coins","started":"2026-10-11T01:27:25+00:00","rounds":[{"k":1,"minutes":36.8},{"k":2,"minutes":15.1}],"messages":[{"t":"2026-10-11T01:27:32.953Z","text":"I’ll inspect the simulator and data, submit a simple causal baseline early, then test whether dispersion and correlation improve coin selection and risk control. I’ll keep a record of the variants and maths for the writeup."},{"t":"2026-10-11T01:28:13.074Z","text":"The BTC trend baseline is with the referee. I’m now testing whether low-correlation markets support persistent coin-specific trends, and whether rising correlation warns that those trades are becoming exposed to a broad selloff."},{"t":"2026-10-11T01:29:44.970Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first coin-rotation tests show severe losses, so I’m tracing the losing trades before trusting the ranking signal or adding more filters."},{"t":"2026-10-11T01:31:10.832Z","text":"The losses cluster around sharp coin-specific pumps that reverse soon afterward. I’m testing two responses: rank trends after discounting large jumps, and test short-term pullbacks as a separate signal. Each will be checked across three annual windows with fresh starting cash."},{"t":"2026-10-11T01:34:42.838Z","text":"Neither raw momentum nor indiscriminate pullback buying is beating the baseline. I’m keeping that entry while testing slower trend rules and a walk-forward model that can learn when a move tends to continue or reverse. The model will train only on outcomes already known at each refit."},{"t":"2026-10-11T01:38:59.339Z","text":"The major-coin rotation was accepted: $100 became $516.08, with a 39.7% maximum drawdown. The broad-universe tests still look weak, including the first learned model. I’m now comparing steadier liquidity screens and faster trend exits to see whether either improves growth across all three annual windows."},{"t":"2026-10-11T01:44:34.984Z","text":"Sentiment confirmation is helping in the faster strategy: the current candidate turns $100 into about $1,318 with a 28.8% drawdown, and gains in each annual window. It trades a major coin only when its price trend is positive, BTC is above its 50-day average, and Fear & Greed is above its trailing mean. I’m checking nearby settings before submitting it."},{"t":"2026-10-11T01:50:02.803Z","text":"The sentiment-confirmed entry was accepted at $1,318.21. Nearby settings retain substantial gains, particularly across the fast-trend and sentiment windows. I’m checking doubled costs and an extra fill delay now. The broad expert ensemble and most rebound overlays have underperformed the simpler rule."},{"t":"2026-10-11T01:54:41.644Z","text":"The cost checks passed: doubled costs leave about $1,008, and an extra four-hour fill delay leaves about $1,023. The largest limitation is rally dependence: the ten best days supply 62% of log growth. I’m testing smoother sentiment confirmation to reduce daily gate flips, then I’ll finalize the entry and writeup."},{"t":"2026-10-11T01:57:33.034Z","text":"A slower common gate improves the tradeoff: BTC above its 100-day average and Fear & Greed above a six-week average produce about $1,654 with a 26.0% drawdown. All nearby settings tested remain above $1,000, and every annual slice gains. I’ll submit this version and update the writeup with its exact results."},{"t":"2026-10-11T02:03:01.801Z","text":"The final entry was accepted: $100 became $1,653.95, with a 26.0% maximum drawdown. Doubled costs retain $1,329.98; an extra four-hour delay retains $1,179.63. The writeup now has the equations, parameters, trial counts and failed ideas. I’m checking that the delivered file matches the accepted submission."},{"t":"2026-10-11T02:04:16.531Z","text":"Entry: accepted #004; strategy and writeup.  \nVisible years: **$100 → $1,653.95**, maximum drawdown **−26.0%**.  \nNext idea: adapt trend horizons using a volatility clock."},{"t":"2026-10-11T02:04:24.434Z","text":"I’ll use the remaining time to test whether a volatility clock improves trend timing. I’ll keep the accepted strategy unless a challenger shows a clear improvement across annual slices, then finalize the entry and writeup before sealing."},{"t":"2026-10-11T02:06:17.161Z","text":"The accepted entry is intact: $100 became $1,653.95, with a 26.04% maximum drawdown. The writeup already documents its signals and validation. I’m testing a final idea: shortening trend memory when volatility rises and lengthening it when markets quiet down, while checking each annual slice."},{"t":"2026-10-11T02:07:17.291Z","text":"The volatility clock produced one promising challenger: $1,982.65 with a 26.44% drawdown. Its improvement came mainly from the middle year; it also traded more and paid higher costs. I’ll check nearby settings and execution stress before deciding whether that gain justifies the added rule."},{"t":"2026-10-11T02:08:22.078Z","text":"The nearby volatility-clock settings were uneven, so the visible gain is not enough to replace the simpler accepted rule. Excluding recent pumps from broad-coin momentum also failed. I’m checking one remaining gap: newly listed coins, which the earlier screens excluded until they had 30–90 days of history."},{"t":"2026-10-11T02:10:07.057Z","text":"New-listing momentum failed after costs. I’m now checking whether volatility-adjusted rankings and using sentiment only for entry improve the major-coin rule. The accepted submission remains the fallback; I’ll replace it only if a challenger improves growth without depending on one narrow setting."},{"t":"2026-10-11T02:11:33.202Z","text":"The strongest challenger keeps the existing filters and ranks eligible majors by momentum multiplied by the square root of their volatility relative to BTC. It returned $1,881.75, improved all three annual slices, and reduced drawdown to 25.06%. Nearby settings generally retained the gain, and the causal check passed. I’m submitting this version and updating the writeup."},{"t":"2026-10-11T02:13:00.561Z","text":"Submission #005 has been accepted and is now the entry: $100 → $1,881.75, with a 25.06% drawdown. The working file matches the accepted source. I’m finishing the writeup with the exact volatility formula, updated results and search count, and the limits of the evidence."},{"t":"2026-10-11T02:15:32.084Z","text":"I found one useful execution fix: weekly target refreshes can leave a fresh account waiting to enter. Emitting the current target every bar removes that delay; because the portfolio holds one coin at 100% or cash, the test added no trades. It returned $1,942.08. I’ll submit this correction once the required five-minute interval has passed."},{"t":"2026-10-11T02:18:24.212Z","text":"Submission #006 is accepted: $100 → $1,942.08, with a 25.06% drawdown and 228 trades. The writeup now covers the final equations, 930 logged configuration evaluations, failed ideas and validation limits. I’m checking that its reported results and source fingerprint match the accepted entry."},{"t":"2026-10-11T02:19:19.610Z","text":"Submission **#006 is my final accepted entry**, matching entry.py. It rotates BTC, ETH and SOL using trend and sentiment confirmation, volatility-adjusted momentum, and a switching buffer. Visible result: **$100 → $1,942.08 (19.42×)**, with **25.06% maximum drawdown**. Causal checks passed. WRITEUP.md is complete with precise equations, parameters, 930 logged configuration evaluations, validation results, failed ideas and limitations."}],"submissions":[{"n":"001","t":"2026-10-11T01:27:57.015940+00:00","note":"Causal BTC 200-day filter baseline; research in progress.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:35:41.477075+00:00","note":"28-day momentum in BTC/ETH/SOL with individual 50-day trend filters; more stable than raw all-coin ranks.","status":"ok","dev":516.076749383975,"dd":-0.3972465767746518,"reason":""},{"n":"003","t":"2026-10-11T01:46:24.347139+00:00","note":"4h major-coin trend rotation confirmed by BTC 50-day trend and rising Fear & Greed; nearby settings retain gains.","status":"ok","dev":1318.2053846079777,"dd":-0.2883846654866684,"reason":""},{"n":"004","t":"2026-10-11T01:58:16.092729+00:00","note":"Final 4h trend rotation with BTC 100-day and Fear & Greed six-week confirmation; profitable annual slices and parameter neighborhood, reduced drawdown.","status":"ok","dev":1653.9529213996989,"dd":-0.26035683816883126,"reason":""},{"n":"005","t":"2026-10-11T02:11:34.755071+00:00","note":"Confirmed major trends; weekly relative-volatility ranking; causal, annual-slice and cost checks.","status":"ok","dev":1881.748916786542,"dd":-0.2506288866770916,"reason":""},{"n":"006","t":"2026-10-11T02:16:53.858998+00:00","note":"Same confirmed volatility-ranked major trends; emit every bar to initialize fresh accounts promptly.","status":"ok","dev":1942.0761425079395,"dd":-0.2506288866770917,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: accepted #004; strategy and writeup.  \nVisible years: **$100 → $1,653.95**, maximum drawdown **−26.0%**.  \nNext idea: adapt trend horizons using a volatility clock."},{"k":2,"text":"Submission **#006 is my final accepted entry**, matching entry.py. It rotates BTC, ETH and SOL using trend and sentiment confirmation, volatility-adjusted momentum, and a switching buffer. Visible result: **$100 → $1,942.08 (19.42×)**, with **25.06% maximum drawdown**. Causal checks passed. WRITEUP.md is complete with precise equations, parameters, 930 logged configuration evaluations, validation results, failed ideas and limitations."}],"title":"Cormorant (a37): writeup","idea":"The accepted entry is a fully invested or cash, long-only trend rotation among BTC, ETH and SOL at four-hour frequency. A coin must have a positive medium-term return, a rising fast/slow price trend, and a close above its slower average. A common gate additionally requires BTC above its daily moving average and the daily Fear & Greed index above its own trailing exponential average. Among eligible coins the strategy ranks twenty-day momentum multiplied by the square root of its trailing volatility relative to BTC, retaining the incumbent until a challenger has a small score advantage. This is a mild preference for greater upside participation during a confirmed uptrend; it does not increase position size or permit leverage. My assigned direction was dispersion and correlation across coins. Extreme, low-correlation moves in the broader spot universe were usually poor trades after the next-open fill and costs. Correlation screens and broad dispersion gates did not improve the major-coin strategy consistently; the final volatility adjustment uses dispersion in the permitted liquid majors instead. Final accepted submission: #006, entry_complete.py, byte-identical working copy entry.py. Referee result: $100 to $1,942.08 over the visible years. This remains far below the growth rate needed for the stated million-dollar goal. The accepted source has SHA-256 c4c8361c6c9b113e3dc18ea33ce","failed":"- Raw cross-sectional momentum: 7–63-day leaders selected from dynamically liquid coins, with 1 or 3 holdings and several market gates. The best initial batch ended at $27.67. Concentrated versions suffered repeated severe reversals after pumps. - Risk-normalized and jump-discounted momentum: dividing by volatility, subtracting the largest up day, clipping daily jumps, and ranking median returns did not produce a competitive broad-universe strategy. The corresponding batches' best results were $83.80 and $20.45. - Indiscriminate short-term reversal: buying the worst one- or three-day performers often caught continuing collapses. The best batch ended at $12.33. A separate short-horizon momentum batch also failed, topping out at $16.53. - Persistent breakout holdings: fresh 20- or 60-day highs with volume confirmation and 15% or 25% trailing stops did not overcome the weak subsequent returns. Best result: $93.57. - Stable-liquidity screens: trailing median volume, longer listing age and longer own trend filters improved the failures but topped out at $219.38. A later quiet-trend family combined a volatility cap, jump concentration penalty, median-volume universe and sentiment confirmation; it reached $419.34 in 32 configurations, still below the major-coin strategy. - Correlation and beta screens: higher BTC correlation was associated with less adverse subsequent returns than low correlation in the pooled observations, but actual high-correlation rotations topped out at $189.32. This was insufficient for inclusion. - Cross-sectional dispersion/correlation gates on the major trend strategy: excluding elevated dispersion or requiring a high common-correlation proxy generally removed profitable exposure. Fear & Greed confirmation was more useful in this sample. - Walk-forwar","holdup":"The proposed mechanism is persistent demand and relative leadership among liquid spot majors, with agreement between price and sentiment before risking capital. A rising sentiment reading is insufficient alone: BTC must have an established positive price trend, and the selected coin must satisfy both its absolute trend and positive return conditions. The volatility factor mildly favors participation in a stronger-amplitude trend, conditional on these confirmations. It can increase downside risk and is not a claim that high volatility by itself earns a premium. A strong coin-specific pump is insufficient without common confirmation. Hysteresis reduces fees from small leadership reversals. Cash provides the permitted defense against persistent declines. These claims refer only to the available data and general market mechanisms. The rule contains no event dates, manually selected alternate coins, forecasts from later years, or assumptions about later listings. Nearby parameters, doubled costs and delayed execution retain gains, which is preferable to one isolated winning configuration. Nevertheless, the search was broad, there are only three visible evaluation years, much of the prof","result":{"dev":{"final":1942.0761425079395,"max_drawdown":-0.2506288866770917,"trades":228,"exposure":0.30885036496350365,"assets_traded":3,"fees_paid":168.48787183041324},"hidden":{"final":195.06050441626306,"max_drawdown":-0.24792182875575297,"trades":170,"exposure":0.3150684931506849,"assets_traded":3,"fees_paid":24.402311736261346},"full":{"final":3788.2235197238797,"max_drawdown":-0.2506288866770917,"trades":398,"exposure":0.31133625410733845,"assets_traded":3,"fees_paid":642.3993462807583}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",116.7007],["2021-10-31",106.2081],["2021-11-07",133.9054],["2021-11-14",120.4765],["2021-11-21",115.9431],["2021-11-28",115.9431],["2021-12-05",115.9431],["2021-12-12",115.9431],["2021-12-19",115.9431],["2021-12-26",115.9431],["2022-01-02",115.9431],["2022-01-09",115.9431],["2022-01-16",115.9431],["2022-01-23",115.9431],["2022-01-30",115.9431],["2022-02-06",115.9431],["2022-02-13",115.9431],["2022-02-20",115.9431],["2022-02-27",115.9431],["2022-03-06",115.9431],["2022-03-13",115.9431],["2022-03-20",115.9431],["2022-03-27",115.8893],["2022-04-03",151.8808],["2022-04-10",131.9231],["2022-04-17",131.9231],["2022-04-24",131.9231],["2022-05-01",131.9231],["2022-05-08",131.9231],["2022-05-15",131.9231],["2022-05-22",131.9231],["2022-05-29",131.9231],["2022-06-05",131.9231],["2022-06-12",131.9231],["2022-06-19",131.9231],["2022-06-26",131.9231],["2022-07-03",131.9231],["2022-07-10",131.9231],["2022-07-17",131.9231],["2022-07-24",131.9231],["2022-07-31",131.9231],["2022-08-07",131.9231],["2022-08-14",131.9231],["2022-08-21",131.9231],["2022-08-28",131.9231],["2022-09-04",131.9231],["2022-09-11",131.9231],["2022-09-18",131.9231],["2022-09-25",131.9231],["2022-10-02",131.9231],["2022-10-09",131.9231],["2022-10-16",131.9231],["2022-10-23",131.9231],["2022-10-30",131.9231],["2022-11-06",124.5625],["2022-11-13",118.9607],["2022-11-20",118.9607],["2022-11-27",118.9607],["2022-12-04",118.9607],["2022-12-11",118.9607],["2022-12-18",118.9607],["2022-12-25",118.9607],["2023-01-01",118.9607],["2023-01-08",118.9607],["2023-01-15",167.3384],["2023-01-22",178.3702],["2023-01-29",188.5423],["2023-02-05",166.6828],["2023-02-12",159.3134],["2023-02-19",163.839],["2023-02-26",143.522],["2023-03-05",144.0048],["2023-03-12",144.0048],["2023-03-19",164.5229],["2023-03-26",162.0313],["2023-04-02",162.7062],["2023-04-09",160.8874],["2023-04-16",171.5263],["2023-04-23",173.3806],["2023-04-30",170.9826],["2023-05-07",163.4427],["2023-05-14",163.4427],["2023-05-21",163.4427],["2023-05-28",163.4427],["2023-06-04",163.4427],["2023-06-11",163.4427],["2023-06-18",163.4427],["2023-06-25",165.6523],["2023-07-02",170.5581],["2023-07-09",184.527],["2023-07-16",237.6005],["2023-07-23",232.5706],["2023-07-30",232.5706],["2023-08-06",232.5706],["2023-08-13",230.0064],["2023-08-20",234.7149],["2023-08-27",234.7149],["2023-09-03",234.7149],["2023-09-10",234.7149],["2023-09-17",234.7149],["2023-09-24",234.7149],["2023-10-01",234.7149],["2023-10-08",234.7149],["2023-10-15",234.7149],["2023-10-22",280.8869],["2023-10-29",323.0836],["2023-11-05",398.8614],["2023-11-12",548.5568],["2023-11-19",676.6854],["2023-11-26",609.8414],["2023-12-03",651.2708],["2023-12-10",707.6997],["2023-12-17",641.6605],["2023-12-24",1045.9869],["2023-12-31",963.3673],["2024-01-07",929.5818],["2024-01-14",1062.228],["2024-01-21",1062.228],["2024-01-28",1062.228],["2024-02-04",1054.2423],["2024-02-11",1156.9773],["2024-02-18",1174.0127],["2024-02-25",1279.3552],["2024-03-03",1438.6104],["2024-03-10",1541.7222],["2024-03-17",2075.2101],["2024-03-24",1943.6281],["2024-03-31",1977.733],["2024-04-07",1853.0557],["2024-04-14",1886.4402],["2024-04-21",1886.4402],["2024-04-28",1886.4402],["2024-05-05",1886.4402],["2024-05-12",1912.2533],["2024-05-19",2029.4996],["2024-05-26",2193.7026],["2024-06-02",2139.1324],["2024-06-09",2085.9792],["2024-06-16",2064.1131],["2024-06-23",2064.1131],["2024-06-30",2064.1131],["2024-07-07",2064.1131],["2024-07-14",2064.1131],["2024-07-21",2113.1526],["2024-07-28",2193.8889],["2024-08-04",2036.5518],["2024-08-11",2036.5518],["2024-08-18",2036.5518],["2024-08-25",2014.6368],["2024-09-01",1968.162],["2024-09-08",1968.162],["2024-09-15",1968.162],["2024-09-22",1933.224],["2024-09-29",2130.1809],["2024-10-06",1996.9664],["2024-10-13",1916.8554],["2024-10-20",2091.6458],["2024-10-27",2144.8787],["2024-11-03",1990.4442],["2024-11-10",2221.0051],["2024-11-17",2291.3834],["2024-11-24",2419.6775],["2024-12-01",2350.8748],["2024-12-08",2380.6511],["2024-12-15",2190.2389],["2024-12-22",2016.2658],["2024-12-29",2016.2658],["2025-01-05",2016.2658],["2025-01-12",1993.8883],["2025-01-19",2542.3447],["2025-01-26",2417.633],["2025-02-02",2228.4281],["2025-02-09",2228.4281],["2025-02-16",2228.4281],["2025-02-23",2228.4281],["2025-03-02",2228.4281],["2025-03-09",2228.4281],["2025-03-16",2228.4281],["2025-03-23",2228.4281],["2025-03-30",2228.4281],["2025-04-06",2228.4281],["2025-04-13",2228.4281],["2025-04-20",2228.4281],["2025-04-27",2244.9057],["2025-05-04",2126.3578],["2025-05-11",2851.5071],["2025-05-18",2738.0162],["2025-05-25",2764.7031],["2025-06-01",2814.2016],["2025-06-08",2749.3623],["2025-06-15",2676.1524],["2025-06-22",2676.1524],["2025-06-29",2703.0444],["2025-07-06",2617.6005],["2025-07-13",2889.7802],["2025-07-20",3624.9054],["2025-07-27",3700.6818],["2025-08-03",3525.4838],["2025-08-10",3732.181],["2025-08-17",3895.7399],["2025-08-24",3895.7399],["2025-08-31",3895.7399],["2025-09-07",3895.7399],["2025-09-14",3891.167],["2025-09-21",3795.4391],["2025-09-28",3795.4391],["2025-10-05",3918.8417],["2025-10-12",3800.8268],["2025-10-19",3800.8268],["2025-10-26",3800.8268],["2025-11-02",3800.8268],["2025-11-09",3800.8268],["2025-11-16",3800.8268],["2025-11-23",3800.8268],["2025-11-30",3800.8268],["2025-12-07",3800.8268],["2025-12-14",3800.8268],["2025-12-21",3800.8268],["2025-12-28",3800.8268],["2026-01-04",3800.8268],["2026-01-11",3800.8268],["2026-01-18",3800.8268],["2026-01-25",3800.8268],["2026-02-01",3800.8268],["2026-02-08",3800.8268],["2026-02-15",3800.8268],["2026-02-22",3800.8268],["2026-03-01",3800.8268],["2026-03-08",3800.8268],["2026-03-15",3800.8268],["2026-03-22",3800.8268],["2026-03-29",3800.8268],["2026-04-05",3800.8268],["2026-04-12",3800.8268],["2026-04-19",3718.588],["2026-04-26",3687.9876],["2026-05-03",3604.5037],["2026-05-10",3814.7832],["2026-05-17",3362.606],["2026-05-24",3362.606],["2026-05-31",3362.606],["2026-06-07",3362.606],["2026-06-14",3362.606],["2026-06-21",3362.606],["2026-06-28",3362.606],["2026-07-05",3362.606],["2026-07-12",3362.606],["2026-07-19",3362.606],["2026-07-26",3362.606],["2026-08-02",3362.606],["2026-08-09",3362.606],["2026-08-16",3362.606],["2026-08-23",3645.7686],["2026-08-30",4021.6249],["2026-09-06",3967.2844],["2026-09-13",3754.6832],["2026-09-20",3541.8868],["2026-09-27",3945.2995],["2026-10-04",3883.4087],["2026-10-11",3788.2235]],"code_bytes":1546},"a38":{"id":"a38","alias":"Albatross","theme":"Nearest-neighbour and analog forecasting (pattern matching on past market states)","started":"2026-10-11T01:27:26+00:00","rounds":[{"k":1,"minutes":37.8},{"k":2,"minutes":14.4}],"messages":[{"t":"2026-10-11T01:27:32.349Z","text":"I’ll inspect the data and simulator, submit a causal baseline early, then test whether historical analogs improve its returns. I’ll record the experiments and signal equations in the writeup."},{"t":"2026-10-11T01:28:30.946Z","text":"The baseline is accepted: $100 became $230.64, with a 25.7% maximum drawdown. The data includes 459 assets and delistings. I’ll use that changing universe to test momentum rotation, then compare each candidate with similar past states whose outcomes were already known."},{"t":"2026-10-11T01:31:07.923Z","text":"The first rotation tests performed poorly: chasing recent winners led to severe losses. Across the visible period, even the most liquid coins’ strongest recent moves tended to reverse over the following week. I’ll test rebound signals and market filters, and let the analog model estimate where continuation has historically worked."},{"t":"2026-10-11T01:32:52.397Z","text":"Market filters reduced the losses, but broad coin rotation still trails the BTC baseline. I’m checking BTC/ETH/SOL as a comparison while the analog forecasts run. The large gap between ordinary rotation and the leaderboard makes it especially important to avoid selecting a fragile rule on its visible result alone."},{"t":"2026-10-11T01:35:05.887Z","text":"BTC/ETH/SOL rotation is the first improvement: $100 became $613.98, with a 44.5% maximum drawdown, and the causal checks passed. I’ll submit it now. I’m also testing analogs normalized by volatility, so patterns from these three assets can be compared without treating all return sizes as equivalent."},{"t":"2026-10-11T01:36:50.180Z","text":"The small-universe analog tests trail simple momentum so far. I’m testing two other ways to avoid chasing isolated jumps: rank coins by gains from a causal paper trend strategy, and look for early breakouts with rising volume. Both select coins from the data."},{"t":"2026-10-11T01:38:07.320Z","text":"The $613.98 entry is accepted. Breakout and paper-trend ranking tests have not improved on it. I’ll now test shorter BTC/ETH/SOL signals; the full-universe nearest-neighbour run is taking longer, so I’m also preparing a lower-dimensional version to reduce its runtime."},{"t":"2026-10-11T01:39:51.314Z","text":"The broad one-day analog model lost money, so I stopped its longer run. A relative-return version looks more promising: it forecasts ETH and SOL against BTC, which makes the task choosing a leader rather than predicting the whole market. Its first tested variant reached about $647, close to the simpler baseline."},{"t":"2026-10-11T01:41:58.481Z","text":"Shorter trend signals reached about $711, but neither broad analogs nor skipped-period momentum has produced a substantial improvement. I’m testing a small walk-forward tree model next. It may capture interactions between candle shape, volume and trend that nearest-neighbour distance missed."},{"t":"2026-10-11T01:44:06.332Z","text":"The tree forecasts also trailed the major-asset trend rule. I’ll use analogs for a narrower decision now: forecast the payoff of a fixed momentum policy and decide whether to deploy it. That avoids asking the model to identify the next winning coin across the whole universe."},{"t":"2026-10-11T01:46:12.743Z","text":"The policy analogs produced a meaningful improvement. Using them to veto weak setups inside the BTC trend filter reached roughly $989–$1,030, with drawdowns around 30–32%. Several nearby settings also improved on the baseline. I’ll check the stronger variants’ stability and causality before submitting one."},{"t":"2026-10-11T01:49:16.850Z","text":"The self-contained policy-analog entry passed the causal checks: $100 became $1,030.01, with 29.8% maximum drawdown and 144 trades. Doubling costs left about $892. I’ve written the equations and failed experiments into the writeup, and I’m submitting this version now."},{"t":"2026-10-11T01:50:50.235Z","text":"The $1,030.01 version is accepted. Its gains are concentrated in the recovery year; it stayed in cash throughout the visible bear year. I’m using the remaining round time to test whether the analog result survives changes to its state features and training samples."},{"t":"2026-10-11T01:53:32.711Z","text":"The robustness checks are mixed. Removing one state feature produced $614–$1,037; wider spacing between training episodes reduced the selected result to about $705–$871. Recent-error corrections also weakened it. I’ll keep the accepted version while testing one final extension: let analogs choose among a small set of momentum policies."},{"t":"2026-10-11T01:58:02.395Z","text":"A useful execution change emerged: keeping the daily analog gate, trading at four-hour closes, and requiring a 1% momentum advantage before switching coins raised the result to about $1,389. Nearby settings also improved. I’ll verify this version for submission; it reduces switches between nearly tied leaders."},{"t":"2026-10-11T02:00:01.568Z","text":"The four-hour version is accepted: $100 became $1,389.20, with 28.7% maximum drawdown. At double costs it reached $1,105.98; fresh starts in the later visible years also gained. I’m updating the writeup with the separate daily reference policy, four-hour trading rule and final results."},{"t":"2026-10-11T02:03:24.887Z","text":"The writeup now reflects the accepted entry and its limitations. Most of the gains came from a few strong days: removing the ten best daily returns leaves about $287. I’m checking hourly execution with the same lookback lengths as a final comparison; the four-hour submission remains the current entry."},{"t":"2026-10-11T02:05:16.666Z","text":"Entry: accepted #004; strategy and writeup.  \nVisible result: $100 → **$1,389.20** (13.89×), 28.7% max drawdown; causal checks passed.  \nNext idea: estimate analog distance scales separately by market regime."},{"t":"2026-10-11T02:05:25.243Z","text":"I’ll use the remaining time to test adaptive distance scales for the analog model, check that the judged file matches the accepted submission, and finish the writeup. I’ll replace the current entry only if an improvement survives causal and cost checks."},{"t":"2026-10-11T02:07:38.339Z","text":"The current accepted entry has a causal check and a completed draft writeup. Its main weakness is that nearby analogs often come from overlapping episodes. I’m testing whether adapting the distance metric or requiring separated episodes improves the forecast while keeping the trading rule fixed."},{"t":"2026-10-11T02:08:34.881Z","text":"The adaptive distance metrics and separated-episode tests did not beat the accepted rule. Recent-outcome weighting also weakened it. I’m keeping the analog model unchanged and testing a small range of momentum horizons and trend filters around the four-hour execution rule."},{"t":"2026-10-11T02:10:54.719Z","text":"The execution tests also favor the accepted settings; adding coin-level relative-return forecasts did not improve them. I’m trying one final idea: a separate analog library for young listings, where the relevant patterns may differ from mature coins. I’ll assess it separately before considering a replacement."},{"t":"2026-10-11T02:13:57.453Z","text":"The young-listing model was much weaker than the accepted entry. After I corrected its outcome labels to retain delisted coins at their last known value, the best quality-filtered version reached $133.29. I’m rejecting it and making one final check of percentile-based and market-trajectory analog distances."},{"t":"2026-10-11T02:15:23.474Z","text":"None of the final-round variants beat the accepted strategy. I’m retaining the four-hour momentum rule with the daily analog veto and 1% switching margin. I’ll resubmit that exact file to confirm the final entry, then finish the writeup with the additional trials and their results."},{"t":"2026-10-11T02:17:58.186Z","text":"Submission #005 is accepted and is now the entry for judging. The unchanged file returned $1,389.20 from $100 on the visible years, with 28.67% drawdown and 228 trades after costs. The writeup now includes all 2,347 research evaluations, the failed final-round ideas, and the strategy’s generalization limits."},{"t":"2026-10-11T02:19:42.360Z","text":"Submission **#005 is accepted and final**. entry.py uses daily nearest-neighbour policy forecasts and a BTC trend gate to control four-hour BTC/ETH/SOL momentum, with a 1% switching margin. Visible results: **$100 → $1,389.20**, **28.67% maximum drawdown**, and **228 trades**, after costs. Causal checks passed. WRITEUP.md is complete with equations, parameters, all **2,347 research evaluations**, robustness results, failed ideas, and the rationale and limitations for unseen years."}],"submissions":[{"n":"001","t":"2026-10-11T01:27:46.422683+00:00","note":"Initial causal BTC 200-day trend baseline; analog research follows.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:35:08.341220+00:00","note":"Major-asset relative strength with own 50-day trend and BTC 100-day market filter; all causal checks passed.","status":"ok","dev":613.9840487750861,"dd":-0.4454794644142003,"reason":""},{"n":"003","t":"2026-10-11T01:49:18.295292+00:00","note":"Purged historical analogs forecast a fixed major-asset momentum policy; negative forecasts veto positions inside a BTC 200-day trend gate. Causal checks passed; double-cost result about $892.","status":"ok","dev":1030.010040495169,"dd":-0.29840521036250534,"reason":""},{"n":"004","t":"2026-10-11T01:58:39.473741+00:00","note":"Daily purged policy analogs gate four-hour major-asset momentum. A 1% leader-switch margin reduces turnover; causal daily-context alignment and referee prefix checks passed.","status":"ok","dev":1389.200825340662,"dd":-0.28671488339151807,"reason":""},{"n":"005","t":"2026-10-11T02:15:25.305928+00:00","note":"Final Albatross: unchanged causal daily policy-analog veto, BTC long-trend gate, four-hour major momentum and 1% leader-switch margin; final-round alternatives rejected.","status":"ok","dev":1389.200825340662,"dd":-0.28671488339151807,"reason":""}],"entry":"005","round_notes":[{"k":1,"text":"Entry: accepted #004; strategy and writeup.  \nVisible result: $100 → **$1,389.20** (13.89×), 28.7% max drawdown; causal checks passed.  \nNext idea: estimate analog distance scales separately by market regime."},{"k":2,"text":"Submission **#005 is accepted and final**. entry.py uses daily nearest-neighbour policy forecasts and a BTC trend gate to control four-hour BTC/ETH/SOL momentum, with a 1% switching margin. Visible results: **$100 → $1,389.20**, **28.67% maximum drawdown**, and **228 trades**, after costs. Causal checks passed. WRITEUP.md is complete with equations, parameters, all **2,347 research evaluations**, robustness results, failed ideas, and the rationale and limitations for unseen years."}],"title":"Albatross (a38): writeup","idea":"The entry is a historical-analog veto on a momentum policy, with friction in leader switching. A fixed daily reference policy holds the strongest eligible asset among BTC, ETH and SOL, the three expressly permitted named assets. Its mature historical outcomes label the analog library. BTC's long-term trend and the analog prediction jointly gate trading. The executed policy updates every four hours on the same fourteen-day momentum and approximately twenty-day EMA timescales, and switches leaders only when the new coin has a one-percentage-point advantage in log momentum. This reduces trades between nearly tied leaders. Pooled forecasts of individual coins failed; forecasting the payoff of a fixed policy worked better. The strategy is long-only, holds one coin at a time, and contains no calendar overrides, stored fitted model or embedded historical prices. A negligible numerical BTC target in cash states avoids an observed simulator error on wholly cash subwindows.","failed":"- Chasing the highest 7-56-day returns across dynamically liquid coins, with or without volatility normalisation, lost most of the starting account. Diversification and slower rebalancing did not fix it. - Buying recent losers across the same universe also failed after costs and during sustained declines. - Restricting broad rotation to BTC trend regimes improved it, but the best tested broad variants still trailed the initial BTC baseline. - Pooled nearest-neighbour coin forecasts at 1-14-day horizons failed, including relative-return targets, downside penalties, current trend filters and a lower-dimensional, faster model. The full eight-dimensional one-day run was stopped after its completed forecast family lost money. - Major-only absolute-return analogs trailed simple momentum. Relative-to-BTC analogs were better, reaching roughly $647, but gave only a small gain for their complexity. - Selecting coins by historical paper-trend profits, abnormal volume, new highs, listing age, or momentum with the most recent days skipped did not beat major-only trends. - Four-hour major-asset trend rotation without the analog gate reached about $711. Four-hour execution with the daily analog gate but no leader margin reached $1,073; the margin improved it to $1,389. Single-coin and short-term dip-adjusted major rules also failed to improve it. - A shallow, regularly refitted histogram boosting model with momentum, candle shape, volume, listing age and market context did not beat the trend baseline. LightGBM could not import because its OpenMP shared library was unavailable; the histogram model used the installed scikit-learn implementation instead. - A stand-alone positive-forecast analog timing rule, and allowing analogs to override a bearish long-term market gate, were weaker tha","holdup":"The reference and executed policies rely on persistent trends and rotate mechanically into relative strength. The switching margin reduces turnover when the three majors have similar recent performance. Its named universe is limited to the expressly permitted low-cost majors; broad coin-level strategies were rejected on their actual results. The 200-day gate addresses sustained market declines. Analogs supply a separate, adaptive check on circumstances in which that fixed policy has historically struggled, including deteriorating breadth and volatile rallies. The library rolls forward and new labels enter only after their outcomes mature. Daily reference forecasts transfer to a closely related intraday policy, which remains an approximation and a separate source of model risk. Thus the same code can learn from newly observed regimes without a researcher changing the file. The causal separation is explicit: query features use data through today's close; labels refer to old episodes whose full forward horizon is already over; training libraries stay fixed between scheduled rebuilds. Prefix tests check that adding later data cannot alter any previous target. The implementation is smal","result":{"dev":{"final":1389.200825340662,"max_drawdown":-0.28671488339151807,"trades":228,"exposure":0.33424574209295527,"assets_traded":3,"fees_paid":167.30783776671825},"hidden":{"final":191.74168093375985,"max_drawdown":-0.3344882140710086,"trades":210,"exposure":0.371461187215098,"assets_traded":3,"fees_paid":28.656477797998242},"full":{"final":2663.6770140534895,"max_drawdown":-0.3344882140710198,"trades":438,"exposure":0.3491237677989624,"assets_traded":3,"fees_paid":565.403863850062}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",122.759],["2021-10-31",125.5663],["2021-11-07",158.3119],["2021-11-14",150.8578],["2021-11-21",139.8099],["2021-11-28",139.8099],["2021-12-05",139.8099],["2021-12-12",139.8099],["2021-12-19",139.8099],["2021-12-26",141.0848],["2022-01-02",131.816],["2022-01-09",131.816],["2022-01-16",131.816],["2022-01-23",131.816],["2022-01-30",131.816],["2022-02-06",131.816],["2022-02-13",131.816],["2022-02-20",131.816],["2022-02-27",131.816],["2022-03-06",131.816],["2022-03-13",131.816],["2022-03-20",131.816],["2022-03-27",131.816],["2022-04-03",131.816],["2022-04-10",131.816],["2022-04-17",131.816],["2022-04-24",131.816],["2022-05-01",131.816],["2022-05-08",131.816],["2022-05-15",131.816],["2022-05-22",131.816],["2022-05-29",131.816],["2022-06-05",131.816],["2022-06-12",131.816],["2022-06-19",131.816],["2022-06-26",131.816],["2022-07-03",131.816],["2022-07-10",131.816],["2022-07-17",131.816],["2022-07-24",131.816],["2022-07-31",131.816],["2022-08-07",131.816],["2022-08-14",131.816],["2022-08-21",131.816],["2022-08-28",131.816],["2022-09-04",131.816],["2022-09-11",131.816],["2022-09-18",131.816],["2022-09-25",131.816],["2022-10-02",131.816],["2022-10-09",131.816],["2022-10-16",131.816],["2022-10-23",131.816],["2022-10-30",131.816],["2022-11-06",131.816],["2022-11-13",131.816],["2022-11-20",131.816],["2022-11-27",131.816],["2022-12-04",131.816],["2022-12-11",131.816],["2022-12-18",131.816],["2022-12-25",131.816],["2023-01-01",131.816],["2023-01-08",131.816],["2023-01-15",168.1873],["2023-01-22",179.2749],["2023-01-29",181.9338],["2023-02-05",172.3491],["2023-02-12",161.5505],["2023-02-19",171.0292],["2023-02-26",150.09],["2023-03-05",144.8237],["2023-03-12",144.8237],["2023-03-19",151.9291],["2023-03-26",149.6282],["2023-04-02",150.2514],["2023-04-09",155.1219],["2023-04-16",174.0659],["2023-04-23",174.0659],["2023-04-30",174.0659],["2023-05-07",171.2413],["2023-05-14",168.2435],["2023-05-21",168.2435],["2023-05-28",163.3184],["2023-06-04",172.1785],["2023-06-11",149.6644],["2023-06-18",149.6644],["2023-06-25",172.0507],["2023-07-02",184.7867],["2023-07-09",185.5554],["2023-07-16",231.8721],["2023-07-23",231.8721],["2023-07-30",231.8721],["2023-08-06",231.8721],["2023-08-13",231.8721],["2023-08-20",231.8721],["2023-08-27",231.8721],["2023-09-03",231.8721],["2023-09-10",231.8721],["2023-09-17",231.8721],["2023-09-24",231.8721],["2023-10-01",231.8721],["2023-10-08",231.8721],["2023-10-15",231.8721],["2023-10-22",272.4205],["2023-10-29",313.3452],["2023-11-05",386.8389],["2023-11-12",532.0222],["2023-11-19",581.018],["2023-11-26",543.9208],["2023-12-03",540.8187],["2023-12-10",627.3962],["2023-12-17",627.994],["2023-12-24",983.4075],["2023-12-31",886.1427],["2024-01-07",816.3697],["2024-01-14",831.389],["2024-01-21",822.5],["2024-01-28",782.5766],["2024-02-04",772.2673],["2024-02-11",852.3184],["2024-02-18",915.6829],["2024-02-25",997.8459],["2024-03-03",1122.0585],["2024-03-10",1123.3878],["2024-03-17",1515.0368],["2024-03-24",1363.2235],["2024-03-31",1525.3218],["2024-04-07",1388.9445],["2024-04-14",1281.1736],["2024-04-21",1281.1736],["2024-04-28",1311.3316],["2024-05-05",1241.3298],["2024-05-12",1183.6712],["2024-05-19",1281.8019],["2024-05-26",1461.2815],["2024-06-02",1424.9308],["2024-06-09",1403.4373],["2024-06-16",1339.1788],["2024-06-23",1339.1788],["2024-06-30",1314.8868],["2024-07-07",1280.7889],["2024-07-14",1309.538],["2024-07-21",1604.9551],["2024-07-28",1666.2748],["2024-08-04",1498.8055],["2024-08-11",1498.8055],["2024-08-18",1498.8055],["2024-08-25",1491.4],["2024-09-01",1468.3348],["2024-09-08",1468.3348],["2024-09-15",1468.3348],["2024-09-22",1468.3348],["2024-09-29",1453.597],["2024-10-06",1389.2008],["2024-10-13",1389.2008],["2024-10-20",1404.2692],["2024-10-27",1551.6308],["2024-11-03",1358.8711],["2024-11-10",1531.8148],["2024-11-17",1631.9938],["2024-11-24",1603.7016],["2024-12-01",1651.6404],["2024-12-08",1777.2653],["2024-12-15",1715.2455],["2024-12-22",1679.3469],["2024-12-29",1665.2204],["2025-01-05",1711.9904],["2025-01-12",1644.1306],["2025-01-19",2020.1476],["2025-01-26",1921.0517],["2025-02-02",1532.035],["2025-02-09",1532.035],["2025-02-16",1532.035],["2025-02-23",1513.2144],["2025-03-02",1463.308],["2025-03-09",1463.308],["2025-03-16",1463.308],["2025-03-23",1463.308],["2025-03-30",1495.3001],["2025-04-06",1495.3001],["2025-04-13",1495.3001],["2025-04-20",1495.3001],["2025-04-27",1506.3568],["2025-05-04",1493.2129],["2025-05-11",1986.8602],["2025-05-18",1907.7825],["2025-05-25",1996.5832],["2025-06-01",1861.7727],["2025-06-08",1866.9607],["2025-06-15",1777.7679],["2025-06-22",1680.9345],["2025-06-29",1659.1072],["2025-07-06",1595.9248],["2025-07-13",1877.5383],["2025-07-20",2355.1613],["2025-07-27",2404.3945],["2025-08-03",2207.5506],["2025-08-10",2556.8634],["2025-08-17",2683.6766],["2025-08-24",2830.7514],["2025-08-31",2705.3414],["2025-09-07",2661.4665],["2025-09-14",2877.7445],["2025-09-21",2824.108],["2025-09-28",2630.3057],["2025-10-05",2758.2237],["2025-10-12",2485.258],["2025-10-19",2485.258],["2025-10-26",2496.9632],["2025-11-02",2389.5304],["2025-11-09",2389.5304],["2025-11-16",2389.5304],["2025-11-23",2389.5304],["2025-11-30",2389.5304],["2025-12-07",2389.5304],["2025-12-14",2389.5304],["2025-12-21",2389.5304],["2025-12-28",2389.5304],["2026-01-04",2389.5304],["2026-01-11",2389.5304],["2026-01-18",2389.5304],["2026-01-25",2389.5304],["2026-02-01",2389.5304],["2026-02-08",2389.5304],["2026-02-15",2389.5304],["2026-02-22",2389.5304],["2026-03-01",2389.5304],["2026-03-08",2389.5304],["2026-03-15",2389.5304],["2026-03-22",2389.5304],["2026-03-29",2389.5304],["2026-04-05",2389.5304],["2026-04-12",2389.5304],["2026-04-19",2389.5304],["2026-04-26",2389.5304],["2026-05-03",2389.5304],["2026-05-10",2389.5304],["2026-05-17",2389.5304],["2026-05-24",2389.5304],["2026-05-31",2389.5304],["2026-06-07",2389.5304],["2026-06-14",2389.5304],["2026-06-21",2389.5304],["2026-06-28",2389.5304],["2026-07-05",2389.5304],["2026-07-12",2389.5304],["2026-07-19",2389.5304],["2026-07-26",2389.5304],["2026-08-02",2389.5304],["2026-08-09",2389.5304],["2026-08-16",2389.5304],["2026-08-23",2590.751],["2026-08-30",2748.0754],["2026-09-06",2739.2458],["2026-09-13",2558.9637],["2026-09-20",2633.459],["2026-09-27",2865.1428],["2026-10-04",2820.1967],["2026-10-11",2663.677]],"code_bytes":3872},"a39":{"id":"a39","alias":"Shearwater","theme":"Bayesian model averaging across many trend horizons and filters","started":"2026-10-11T01:28:55+00:00","rounds":[{"k":1,"minutes":37.3},{"k":2,"minutes":12.5}],"messages":[{"t":"2026-10-11T01:29:02.838Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test whether adaptive averaging across trend horizons improves returns across different market regimes. I’ll keep a record of the variants for the writeup."},{"t":"2026-10-11T01:29:33.864Z","text":"The BTC baseline is submitted. The leaderboard’s large returns suggest the main opportunity may lie in rotating among the broader coin universe. I’ll test that directly, with liquidity and trading-history requirements, alongside an ensemble that learns which trend horizons are working."},{"t":"2026-10-11T01:31:07.468Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. I’m now comparing momentum horizons, basket sizes and rebalance intervals, including fresh-start annual tests to see whether gains survive across the bear market and recovery."},{"t":"2026-10-11T01:32:38.238Z","text":"The first momentum sweep failed badly: selecting the strongest recent gainers lost money across all 270 variants. I’m testing the opposite signal—short-term rebounds—while retaining the idea of adapting across horizons and filters. I’ll also distinguish ordinary pullbacks from sustained collapses."},{"t":"2026-10-11T01:35:06.925Z","text":"Unfiltered rebound buying also failed. The data shows a sharp difference between moderate selloffs and extreme losses: the largest drops often continue falling. I’m narrowing the rules to established, liquid coins and testing a walk-forward forecast that uses only outcomes already known at each decision."},{"t":"2026-10-11T01:36:33.643Z","text":"I’ve added a walk-forward model using returns at several horizons, volatility, volume changes and candle shape. It retrains every 28 days on completed next-day outcomes. This tests whether combinations of signals can distinguish follow-through from the extremes that defeated the simple rules."},{"t":"2026-10-11T01:37:35.348Z","text":"So far, simple rotations—including pullback filters—have mostly lost money after costs. The walk-forward forecast is still running. Once its predictions are ready, I’ll test allocation and cost choices separately so I can see whether any improvement comes from the forecast or from portfolio rules."},{"t":"2026-10-11T01:38:41.047Z","text":"The best major-coin rotation returned about $456 in research with a 32.7% drawdown. I’ve submitted it as an interim improvement. It trades the strongest of BTC, ETH and SOL only when both its own trend and BTC’s market trend are positive."},{"t":"2026-10-11T01:40:11.262Z","text":"The interim entry was accepted at $456.50, with a 32.7% drawdown. The learned forecast failed out of sample within the visible period, particularly in the final year, so I’ve rejected it. I’m now testing smoother trends and volume-led entries that may catch moves earlier than raw momentum rankings."},{"t":"2026-10-11T01:43:29.607Z","text":"Four-hour trading produced a stronger result: a 16-day trend filter with an 8-day momentum rotation returned about $824. Averaging many horizons reduced that result, so I’m testing whether stronger evidence weighting or smoother crossover signals can retain more of the gains."},{"t":"2026-10-11T01:45:00.836Z","text":"Smoother four-hour crossovers reached about $932, with a 33.6% drawdown. Nearby settings also made money, and the leading rule gained in each fresh-start annual test. I’m testing a narrower ensemble around those settings, plus a small rebound strategy for periods when the trend rule holds cash."},{"t":"2026-10-11T01:47:24.190Z","text":"The adaptive voting ensemble returned about $804 with a 32.5% drawdown, and gained in all three annual tests. It uses cost-adjusted paper portfolios to weight twenty trend models. That trails the best fixed rule, but gives the strategy a way to adapt when a different horizon starts working."},{"t":"2026-10-11T01:49:16.152Z","text":"The ensemble is now the accepted entry: $100 became $804.06, with a 32.5% drawdown, and it passed the referee’s causality checks. The rebound overlay added only about 14% from a handful of trades; I’m leaving it out because that evidence is too thin to justify the extra rule."},{"t":"2026-10-11T01:50:16.019Z","text":"The ensemble’s fresh-start annual results were $132.36, $154.44 and $393.37 from $100. I’m testing whether adding a permanent cash expert helps when all trend models have recently performed poorly. I’ll also check doubled costs and nearby parameter choices before settling the entry."},{"t":"2026-10-11T01:51:27.923Z","text":"Adding a cash expert improved the result. A two-week evidence window produced about $953 with a 27.4% drawdown when cash received a 20% prior weight. I’m checking nearby settings to see whether that improvement survives modest parameter changes."},{"t":"2026-10-11T01:53:27.733Z","text":"The local parameter grid returned a median of $868; the chosen $953 rule still returned about $650 with doubled trading costs. I’m making one final comparison: a second ensemble that averages choices for the learning window and evidence strength, so adaptation can cover those parameters too."},{"t":"2026-10-11T01:54:39.783Z","text":"The cash-augmented ensemble is accepted at $952.67, with a 27.4% drawdown. The second ensemble added only about $10, so I’m keeping the simpler version. I’ll finish the validation and writeup, including the large search count and how much the result depends on a few strong days."},{"t":"2026-10-11T01:55:50.121Z","text":"Costs remain material: the accepted entry paid about $201 and averaged 37% exposure. The robustness checks now use the exact referee engine. I’m also measuring daily return concentration; a few large days may explain much of the gain despite the positive annual results."},{"t":"2026-10-11T01:59:02.431Z","text":"The exact checks confirmed $650.19 with doubled costs and $852–$1,029 across the six ±20% parameter changes. The best ten days account for about 70% of total log growth, a material limitation. The writeup now includes the equations, roughly 2,900 evaluated configurations, and the approaches that failed."},{"t":"2026-10-11T02:00:05.061Z","text":"The entry and writeup are complete and match the accepted snapshot. There’s time for one focused comparison: apply the cash-augmented learning rule to daily altcoin experts, to test whether it can rescue strategies whose unconditional results were poor. The accepted $952.67 entry remains in place during that test."},{"t":"2026-10-11T02:01:40.026Z","text":"The daily altcoin comparison failed across all 72 settings. Adaptive weighting did not rescue those weak experts, so the accepted $952.67 entry stays in place. I’m adding that result to the writeup and checking that the final artifact and recorded search count agree."},{"t":"2026-10-11T02:02:53.129Z","text":"One remaining cost question is whether small changes in the asset vote cause avoidable switches. I’m testing a minimum lead for a new asset before switching, plus separate thresholds for entering and leaving cash. This checks whether fewer trades improve the result without substantially delaying entries."},{"t":"2026-10-11T02:03:56.801Z","text":"A 10-percentage-point switching margin improved the research result to about $1,035, reduced trades from 382 to 342, and kept drawdown at 27.4%. I’m validating that version now. It keeps the cash rule unchanged and requires a clearer asset preference before paying to switch."},{"t":"2026-10-11T02:05:18.609Z","text":"The final version was accepted: $100 became $1,035.29, with a 27.4% drawdown and 342 trades. It returned $735.42 with doubled costs and $865–$1,118 across the parameter checks. The writeup now reflects the switching-margin rule and final results."},{"t":"2026-10-11T02:06:14.868Z","text":"Entry: #005 accepted — strategy.py; WRITEUP.md.  \nVisible years: **$100 → $1,035.29**, **27.4% drawdown**, 342 trades; causality passed.  \nNext idea: scale the switching margin with volatility and trading costs."},{"t":"2026-10-11T02:06:20.894Z","text":"I’ll use the remaining time to test whether a switching margin tied to volatility improves the accepted strategy. Then I’ll check the final entry and writeup match before the workspace is sealed."},{"t":"2026-10-11T02:08:14.081Z","text":"The accepted entry turns $100 into $1,035 on the visible years, with a 27.4% maximum drawdown. Its result stayed near that level under modest parameter changes. I’m testing whether volatility can guide the switching threshold and reduce unnecessary trades."},{"t":"2026-10-11T02:09:31.437Z","text":"The volatility adjustments made only modest changes; the best rose to $1,130. I’m testing a separate hourly trend ensemble to see whether faster decisions improve growth after fees."},{"t":"2026-10-11T02:10:26.028Z","text":"The hourly ensemble lost much of its advantage to turnover and fees, so I’m keeping the four-hour entry as the fallback. I’m also checking a daily selection rule that favours established, liquid coins with steadier gains."},{"t":"2026-10-11T02:11:31.688Z","text":"Adding a 50-day BTC filter alongside the 100- and 200-day filters improved the ensemble to $1,332 and lowered maximum drawdown to 24.2%. It also improved the first two annual tests. I’ll check that result with the referee and examine nearby parameters before choosing it as the final entry."},{"t":"2026-10-11T02:12:59.555Z","text":"Submission 006 is accepted at $1,332.29, now ranked 35th. Its nearby parameter tests stayed between $1,149 and $1,426; doubling execution costs left $950. I’m updating the writeup and checking one remaining extension: whether favouring mild pullbacks improves asset selection within a rising trend."},{"t":"2026-10-11T02:14:29.459Z","text":"The pullback extension did little; its best variant favoured acceleration and added trades and drawdown for a small gain. I’m leaving it out. My last comparison tests whether the fraction of liquid coins in rising trends adds useful market information beyond the BTC filters."},{"t":"2026-10-11T02:16:07.912Z","text":"The breadth and sentiment filters underperformed the accepted entry. The medium BTC filter remains the clearest improvement, with lower drawdown and solid nearby results. I’m completing the research counts and checking that the writeup’s equations match submission 006."},{"t":"2026-10-11T02:18:05.639Z","text":"The files now match accepted submission 006, and the writeup includes every signal and sizing equation, 3,524 configuration evaluations, failed approaches, and robustness results. I’m keeping the $1,332 entry: the small gains from extra complexity did not justify the additional assumptions."},{"t":"2026-10-11T02:18:44.096Z","text":"Accepted submission **006** in strategy.py averages 30 cost-aware trend experts across BTC, ETH and SOL, with a competing cash model and a margin before switching assets. Visible performance: **$100 → $1,332.29**, **24.2% maximum drawdown**, and 338 trades; referee causality checks passed. Doubling execution costs still produced $950.18. WRITEUP.md is complete with equations, parameters, 3,524 configuration evaluations, robustness results, failed approaches, and the rationale for unseen years."}],"submissions":[{"n":"001","t":"2026-10-11T01:29:20.873716+00:00","note":"Initial causal BTC trend baseline; adaptive horizon research underway.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:38:23.595311+00:00","note":"Liquid BTC/ETH/SOL rotation with 28-day strength, 50-day local and 100-day BTC trend filters; broader momentum trials rejected.","status":"ok","dev":456.4999352842199,"dd":-0.32704051983020344,"reason":""},{"n":"003","t":"2026-10-11T01:48:17.325625+00:00","note":"Discounted-wealth posterior voting across twenty 4h trend/rotation experts; known-close daily BTC filters; cost-aware causal expert updates.","status":"ok","dev":804.0626502097246,"dd":-0.3252892107583555,"reason":""},{"n":"004","t":"2026-10-11T01:53:45.607630+00:00","note":"Twenty cost-adjusted trend experts plus a permanent cash expert; two-week discounted evidence and separate majority exposure/asset votes. Broad neighbourhood and doubled-cost checks passed.","status":"ok","dev":952.6688199526158,"dd":-0.2735663397976452,"reason":""},{"n":"005","t":"2026-10-11T02:04:08.854834+00:00","note":"Final cost-aware Bayesian trend ensemble: unchanged cash majority rule; require a 10-point posterior lead to switch assets. Fewer trades, positive annual tests and cost/parameter robustness.","status":"ok","dev":1035.2916163628959,"dd":-0.273566339797645,"reason":""},{"n":"006","t":"2026-10-11T02:12:05.851211+00:00","note":"Thirty cost-aware trend experts; add a 50-day BTC filter while preserving the cash prior share and switching margin.","status":"ok","dev":1332.286103285517,"dd":-0.24153209930528796,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: #005 accepted — strategy.py; WRITEUP.md.  \nVisible years: **$100 → $1,035.29**, **27.4% drawdown**, 342 trades; causality passed.  \nNext idea: scale the switching margin with volatility and trading costs."},{"k":2,"text":"Accepted submission **006** in strategy.py averages 30 cost-aware trend experts across BTC, ETH and SOL, with a competing cash model and a margin before switching assets. Visible performance: **$100 → $1,332.29**, **24.2% maximum drawdown**, and 338 trades; referee causality checks passed. Doubling execution costs still produced $950.18. WRITEUP.md is complete with equations, parameters, 3,524 configuration evaluations, robustness results, failed approaches, and the rationale for unseen years."}],"title":"Shearwater (a39): writeup","idea":"The submitted strategy uses discounted wealth evidence to average thirty trend models and a permanent cash model, then makes two decisions: whether the ensemble has a majority in favour of being invested, and which asset receives the strongest conditional vote. It trades four-hour BTC, ETH and SOL bars. Each trend model combines a fast/slow price EMA crossover, positive 7- or 14-day momentum, and a 50-, 100- or 200-day BTC market filter. Each model runs an independent virtual spot account with the arena's actual execution timing and 0.10% transaction costs. Models that recently made money gain posterior weight; a cash model gives the ensemble an alternative when the whole trend family loses. The final account takes one full position or holds cash, requiring a ten-percentage-point posterior lead before switching an existing holding. Separating exposure from asset choice avoids treating disagreement between two bullish assets as a reason to hold cash. This is a generalized Bayesian expert algorithm, not a claim that its votes are calibrated probabilities of future returns. The latest accepted entry is submission 006, implemented by strategy.py. Its visible result is $100 → $1,332.29, with 24.15% maximum drawdown. No hidden prices were available or used. This is a substantial historical improvement over the BTC references, but remains far below the million-dollar goal's requ","failed":"- Raw winner chasing failed most clearly. Ranking liquid coins on positive recent log returns, optionally above an EMA, and holding the top one, three or five lost heavily. The largest recent winners often reversed. All 270 initial configurations lost money. - Buying the largest losers was not the answer. The worst falls frequently continued. Pooled visible observations distinguished moderate selloffs from extreme declines, but a generic bottom-rank strategy still lost after costs. - Longer momentum, volatility normalization, pullback caps and “early trend” ranking did not rescue broad daily altcoin rotation. Apparent gains in one recovery were offset by losses or turnover elsewhere. No hand-picked alternative coin list was adopted. - A causal learned forecast failed. I trained one small histogram gradient-boosting regressor every 28 days on the previous 365 days of completed next-day open-to-close log returns. It used 33 price, volatility, volume, candle, BTC and sentiment features; 60 trees, 15 leaves, fixed seed 39, and substantial regularization. Its 64 allocation variants all lost over the full visible interval; its best ended at $58.83. The final annual interval was particularly weak. Complexity did not establish transferable alpha. - Volume-led absorption entries were interesting but unstable. A score resembling \\log(V_t/\\operatorname{median}_{28}(V)_{t-1})-10\\log(H_t/L_t), combined with small daily price moves and liquidity checks, reached $376.57 and a strong last annual interval. Earlier losses and drawdowns outweighed that appeal. - Diffuse Bayesian mixtures diluted the useful trend family. Mixing very short and very long horizons and several market gates left the broad averaging family at best $486.46. Very aggressive evidence temperatures also created switc","holdup":"The justification uses only the observed data through 2024-10-09 and general trading principles: 1. Adaptation is online. Trend horizon and daily filter weights respond to already-realized, cost-adjusted performance. The entry does not commit all capital to the single crossover that happened to win development. 2. Cash is an actual competing model. A universally losing trend family cannot look attractive merely because one member is the least bad. The cash benchmark contributes to the exposure vote independently of current trend gates. 3. The market filter and local trend have different jobs. Daily BTC filters identify broad risk conditions; four-hour local trends and relative momentum decide which major currently carries the strongest move. The visibility shift is explicit. 4. Execution is part of learning. Frequent whipsaws lower an expert's evidence because virtual returns include costs at the next open. This reduces the advantage of a noisy, superficially fast model. The real account further avoids marginal asset switches through its explicit posterior margin. 5. Liquidity and cost assumptions are relatively credible here. Trading is confined to the three liquid, permitted majo","result":{"dev":{"final":1332.286103285517,"max_drawdown":-0.24153209930528796,"trades":338,"exposure":0.3771289537712895,"assets_traded":3,"fees_paid":236.26036195851563},"hidden":{"final":114.89935000708692,"max_drawdown":-0.414291680624871,"trades":288,"exposure":0.3618721461187215,"assets_traded":3,"fees_paid":32.651087756726454},"full":{"final":1530.7880729098047,"max_drawdown":-0.41429168062487176,"trades":626,"exposure":0.37102957283680177,"assets_traded":3,"fees_paid":671.2662667129403}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",124.7791],["2021-10-31",120.0878],["2021-11-07",147.5278],["2021-11-14",140.5815],["2021-11-21",134.2115],["2021-11-28",134.2115],["2021-12-05",128.2742],["2021-12-12",128.2742],["2021-12-19",128.2742],["2021-12-26",128.2742],["2022-01-02",128.2742],["2022-01-09",128.2742],["2022-01-16",128.2742],["2022-01-23",128.2742],["2022-01-30",128.2742],["2022-02-06",128.2742],["2022-02-13",128.2742],["2022-02-20",128.2742],["2022-02-27",128.2742],["2022-03-06",128.2742],["2022-03-13",128.2742],["2022-03-20",128.2742],["2022-03-27",132.219],["2022-04-03",173.2821],["2022-04-10",150.5122],["2022-04-17",150.5122],["2022-04-24",150.5122],["2022-05-01",150.5122],["2022-05-08",150.5122],["2022-05-15",150.5122],["2022-05-22",150.5122],["2022-05-29",150.5122],["2022-06-05",150.5122],["2022-06-12",150.5122],["2022-06-19",150.5122],["2022-06-26",150.5122],["2022-07-03",150.5122],["2022-07-10",150.5122],["2022-07-17",150.5122],["2022-07-24",150.5122],["2022-07-31",150.5122],["2022-08-07",150.5122],["2022-08-14",150.5122],["2022-08-21",150.5122],["2022-08-28",150.5122],["2022-09-04",150.5122],["2022-09-11",150.5122],["2022-09-18",150.5122],["2022-09-25",150.5122],["2022-10-02",150.5122],["2022-10-09",150.5122],["2022-10-16",150.5122],["2022-10-23",150.5122],["2022-10-30",153.4242],["2022-11-06",151.2233],["2022-11-13",144.0907],["2022-11-20",144.0907],["2022-11-27",144.0907],["2022-12-04",144.0907],["2022-12-11",144.0907],["2022-12-18",144.0907],["2022-12-25",144.0907],["2023-01-01",144.0907],["2023-01-08",144.0907],["2023-01-15",209.5075],["2023-01-22",218.4526],["2023-01-29",223.9061],["2023-02-05",200.9021],["2023-02-12",198.2964],["2023-02-19",197.5996],["2023-02-26",182.9052],["2023-03-05",182.9052],["2023-03-12",182.9052],["2023-03-19",211.8088],["2023-03-26",198.679],["2023-04-02",196.4973],["2023-04-09",201.674],["2023-04-16",216.8616],["2023-04-23",202.6947],["2023-04-30",195.8532],["2023-05-07",194.6659],["2023-05-14",194.6659],["2023-05-21",194.6659],["2023-05-28",194.6659],["2023-06-04",200.6762],["2023-06-11",179.8051],["2023-06-18",179.8051],["2023-06-25",203.6169],["2023-07-02",218.6897],["2023-07-09",240.4909],["2023-07-16",309.1703],["2023-07-23",279.4352],["2023-07-30",279.4352],["2023-08-06",279.4352],["2023-08-13",280.6416],["2023-08-20",269.5409],["2023-08-27",269.5409],["2023-09-03",269.5409],["2023-09-10",269.5409],["2023-09-17",269.5409],["2023-09-24",263.1442],["2023-10-01",278.5757],["2023-10-08",286.0551],["2023-10-15",269.625],["2023-10-22",333.8149],["2023-10-29",378.1304],["2023-11-05",440.4851],["2023-11-12",605.8022],["2023-11-19",661.5927],["2023-11-26",593.4974],["2023-12-03",590.332],["2023-12-10",673.8383],["2023-12-17",674.4804],["2023-12-24",1056.2029],["2023-12-31",951.7382],["2024-01-07",909.5112],["2024-01-14",926.244],["2024-01-21",924.0757],["2024-01-28",902.5213],["2024-02-04",914.1541],["2024-02-11",940.2747],["2024-02-18",986.5387],["2024-02-25",1075.0594],["2024-03-03",1232.4054],["2024-03-10",1157.2338],["2024-03-17",1560.6826],["2024-03-24",1386.6299],["2024-03-31",1541.8509],["2024-04-07",1361.2308],["2024-04-14",1265.2383],["2024-04-21",1265.2383],["2024-04-28",1265.2383],["2024-05-05",1265.2383],["2024-05-12",1258.6104],["2024-05-19",1342.8581],["2024-05-26",1520.6111],["2024-06-02",1474.2611],["2024-06-09",1444.7513],["2024-06-16",1426.3833],["2024-06-23",1426.3833],["2024-06-30",1426.3833],["2024-07-07",1426.3833],["2024-07-14",1426.3833],["2024-07-21",1590.5134],["2024-07-28",1651.2814],["2024-08-04",1508.2817],["2024-08-11",1508.2817],["2024-08-18",1508.2817],["2024-08-25",1499.0286],["2024-09-01",1381.5675],["2024-09-08",1381.5675],["2024-09-15",1381.5675],["2024-09-22",1354.8495],["2024-09-29",1421.1608],["2024-10-06",1332.2861],["2024-10-13",1332.2861],["2024-10-20",1297.6435],["2024-10-27",1394.8347],["2024-11-03",1296.3437],["2024-11-10",1374.7664],["2024-11-17",1273.266],["2024-11-24",1241.9157],["2024-12-01",1384.2759],["2024-12-08",1440.936],["2024-12-15",1271.5631],["2024-12-22",1232.7813],["2024-12-29",1232.7813],["2025-01-05",1221.6179],["2025-01-12",1173.1955],["2025-01-19",1386.2974],["2025-01-26",1308.4114],["2025-02-02",1256.0949],["2025-02-09",1256.0949],["2025-02-16",1256.0949],["2025-02-23",1224.1732],["2025-03-02",1224.1732],["2025-03-09",1224.1732],["2025-03-16",1224.1732],["2025-03-23",1224.1732],["2025-03-30",1224.1732],["2025-04-06",1224.1732],["2025-04-13",1224.1732],["2025-04-20",1224.1732],["2025-04-27",1324.1398],["2025-05-04",1283.184],["2025-05-11",1705.3393],["2025-05-18",1637.4662],["2025-05-25",1714.9396],["2025-06-01",1568.8738],["2025-06-08",1574.7377],["2025-06-15",1531.2386],["2025-06-22",1461.3378],["2025-06-29",1453.5296],["2025-07-06",1394.689],["2025-07-13",1572.2054],["2025-07-20",1972.1555],["2025-07-27",1953.7566],["2025-08-03",1843.6062],["2025-08-10",2135.33],["2025-08-17",2241.2363],["2025-08-24",2082.2414],["2025-08-31",1895.1558],["2025-09-07",1780.2921],["2025-09-14",2012.8212],["2025-09-21",1975.3054],["2025-09-28",1930.3793],["2025-10-05",1912.3972],["2025-10-12",1809.909],["2025-10-19",1809.909],["2025-10-26",1809.909],["2025-11-02",1733.9342],["2025-11-09",1733.9342],["2025-11-16",1733.9342],["2025-11-23",1733.9342],["2025-11-30",1733.9342],["2025-12-07",1733.9342],["2025-12-14",1733.9342],["2025-12-21",1733.9342],["2025-12-28",1733.9342],["2026-01-04",1733.9342],["2026-01-11",1733.9342],["2026-01-18",1692.6047],["2026-01-25",1692.6047],["2026-02-01",1692.6047],["2026-02-08",1692.6047],["2026-02-15",1692.6047],["2026-02-22",1692.6047],["2026-03-01",1692.6047],["2026-03-08",1692.6047],["2026-03-15",1692.6047],["2026-03-22",1688.7689],["2026-03-29",1688.7689],["2026-04-05",1688.7689],["2026-04-12",1662.8642],["2026-04-19",1515.281],["2026-04-26",1510.5451],["2026-05-03",1471.943],["2026-05-10",1560.4666],["2026-05-17",1405.2925],["2026-05-24",1405.2925],["2026-05-31",1405.2925],["2026-06-07",1405.2925],["2026-06-14",1405.2925],["2026-06-21",1405.2925],["2026-06-28",1405.2925],["2026-07-05",1405.2925],["2026-07-12",1405.2925],["2026-07-19",1405.2925],["2026-07-26",1405.2925],["2026-08-02",1405.2925],["2026-08-09",1405.2925],["2026-08-16",1405.2925],["2026-08-23",1585.3989],["2026-08-30",1671.6682],["2026-09-06",1588.1762],["2026-09-13",1474.7192],["2026-09-20",1494.4781],["2026-09-27",1654.3724],["2026-10-04",1594.4655],["2026-10-11",1530.7881]],"code_bytes":4607},"a40":{"id":"a40","alias":"Fulmar","theme":"Free choice: invent an approach nobody else in the arena would try","started":"2026-10-11T01:29:27+00:00","rounds":[{"k":1,"minutes":39.5},{"k":2,"minutes":11.5}],"messages":[{"t":"2026-10-11T01:29:34.999Z","text":"I’ll inspect the simulator and data, submit a causal baseline early, then test a few distinct ideas. I’ll keep a record of the variants and results for the writeup."},{"t":"2026-10-11T01:30:12.987Z","text":"The baseline is awaiting the referee. My first research candidate will follow coins with strengthening price and volume, while retaining a winner until its trend weakens. I’ll use the earlier history and visible subperiods to check whether any gains depend on a few isolated rallies."},{"t":"2026-10-11T01:32:21.179Z","text":"The baseline was accepted: $100 became $230.64, with a 25.7% maximum drawdown. The first broad momentum sweep failed badly: it repeatedly bought extreme price spikes before reversals. I’m now testing moderate, persistent trends and pullbacks, with explicit limits on recent price jumps."},{"t":"2026-10-11T01:34:33.844Z","text":"The broad altcoin signals remain weak, even after filtering out sharp spikes. An adaptive predictor also failed to beat the baseline. I’m shifting the next tests toward the most liquid coins and four-hour BTC, ETH and SOL signals, where costs are lower and price moves may be easier to trade."},{"t":"2026-10-11T01:39:14.262Z","text":"The weekly BTC/ETH/SOL rotation was accepted: $100 became $571.64, with a 41.9% drawdown. A four-hour trend candidate also made gains in each visible year. I’ll compare it with a strategy that buys brief selloffs within rising trends, then test the stronger candidate’s sensitivity to parameters and costs."},{"t":"2026-10-11T01:42:01.879Z","text":"The brief selloff strategies did not beat the accepted entry. I’m testing two final alternatives: an online selector of trend rules, and a small model trained only on completed past returns. These will show whether adaptation helps enough to justify the added complexity."},{"t":"2026-10-11T01:43:21.064Z","text":"The online selector produced $704.97, with a 39.2% drawdown and gains in each visible year. It ranks nine BTC/ETH/SOL trend rules using recent returns after estimated fees. I’ll test nearby settings and submit it if that result holds up."},{"t":"2026-10-11T01:48:00.984Z","text":"The selector’s nearby settings also made gains: 15–30 days of memory produced roughly $700–$990. Softer allocations earned less. The best candidate’s latest visible year was about 5.7×, but its three-year total is still far below the goal’s pace. I’ll check higher costs and reliance on the largest winning days next."},{"t":"2026-10-11T01:50:54.111Z","text":"The version that preserves a chosen expert’s cash signal looks more balanced: $100 became $869.53, with gains in all three years and a 38.6% drawdown. Doubling costs left $642.88; removing the ten best daily returns left $167.09. I’ve submitted it and will test whether penalizing uncertain performance improves that balance."},{"t":"2026-10-11T01:55:34.508Z","text":"Giving each expert credit for avoiding losses in its underlying coin improved the result to $908.64, with fewer trades and a 36.1% drawdown. The smaller credit setting gave a similar result; doubling the credit reduced growth. I’ll verify this candidate and compare the same rule at hourly bars."},{"t":"2026-10-11T01:57:43.623Z","text":"Hourly bars earned less after costs, so I’m retaining four-hour decisions. I’m checking the timing-credit candidate’s nearby settings and testing whether estimated growth should set the position size. The writeup will include the full search count and the failed approaches."},{"t":"2026-10-11T02:00:53.182Z","text":"The timing-credit candidate kept $719.07 under doubled costs and $174.90 after removing its ten best daily returns. Its neighboring settings remain profitable. A larger switch buffer improved all three visible years and reduced trades; the sizing test added little, so I’m keeping binary positions and verifying that buffer change."},{"t":"2026-10-11T02:05:40.806Z","text":"Submission #006 was accepted: $100 became $973.26, with a 35.4% drawdown. It kept $784.19 under doubled costs. Jointly changing the main timing parameters by ±20% still made gains, though returns fell substantially. I’m recording those limits alongside the result in the writeup."},{"t":"2026-10-11T02:08:56.721Z","text":"Entry: accepted #006, strategy and writeup.  \nVisible years: $100 → $973.26; max drawdown 35.4%; rank 40/40.  \nNext idea: test cross-asset lead/lag signals for earlier trend entries."},{"t":"2026-10-11T02:09:04.543Z","text":"I’ll keep the accepted strategy in place while I test a few alternatives, then leave enough time for scoring and the final writeup. The biggest open question is whether I can identify useful price moves earlier without adding too much turnover or fragility."},{"t":"2026-10-11T02:11:26.581Z","text":"The accepted entry turns $100 into $973.26 on the visible years, with a 35.4% maximum drawdown. I’m testing whether cross-coin lag signals or slower, less spike-sensitive coin selection can improve that result. The writeup already documents the current rule and its substantial dependence on a few winning days."},{"t":"2026-10-11T02:13:02.375Z","text":"Neither new direction improved the entry: twelve cross-coin variants peaked at $173.65, and twenty-four slower daily rotations peaked at $118.14. I’m now testing whether shorter trend confirmation and different expert-ranking measures help within the liquid-coin strategy."},{"t":"2026-10-11T02:14:31.787Z","text":"The accepted rule also held up against the alternative ranking measures; none improved its visible return. Breakout and recovery experts increased drawdowns and added turnover. I’ll make one final test of an adaptive return forecast, then verify the submission and finish recording the full search."},{"t":"2026-10-11T02:17:08.636Z","text":"The forecast variants peaked at $178.57, and the new-listing rules also lost money. I’m keeping the current nine-expert entry as the candidate for submission. The remaining test checks whether broad market weakness provides a useful reason to stay in cash."},{"t":"2026-10-11T02:18:45.380Z","text":"The market filters also failed to improve the entry; the best fear-and-greed filter produced exactly the same decisions. I’m freezing `fulmar.py` and finishing the writeup. The final-round research added 117 evaluations, with no better candidate, and the source hash still matches the saved $973.26 result."},{"t":"2026-10-11T02:20:28.148Z","text":"Final entry remains accepted #006: fulmar.py, selecting among nine BTC/ETH/SOL trend experts while rewarding loss avoidance and preserving cash decisions. Visible result: **$100 → $973.26**, with **35.4% maximum drawdown**. Another 117 evaluations found no improvement; final causality checks passed. WRITEUP.md is complete with equations, parameters, research counts, results, failed ideas and limitations. Hidden-year performance remains unknown."}],"submissions":[{"n":"001","t":"2026-10-11T01:29:57.853237+00:00","note":"Early baseline: causal BTC 200-day trend filter.","status":"ok","dev":230.63853700985013,"dd":-0.2567657431362347,"reason":""},{"n":"002","t":"2026-10-11T01:36:24.734698+00:00","note":"Weekly BTC/ETH/SOL relative trend rotation with causal daily risk and exit filters.","status":"ok","dev":571.6356999094256,"dd":-0.4192886489572957,"reason":""},{"n":"003","t":"2026-10-11T01:43:34.418161+00:00","note":"Online net-growth selection among nine BTC/ETH/SOL trend experts; 30-day memory, switch buffer, cash when evidence is negative.","status":"ok","dev":704.9703251519761,"dd":-0.39190826306624105,"reason":""},{"n":"004","t":"2026-10-11T01:49:11.805050+00:00","note":"Twenty-day online growth selector; preserve the winning expert cash state, 2% switch buffer, nine major-coin trend experts.","status":"ok","dev":869.528566581745,"dd":-0.3860616777972503,"reason":""},{"n":"005","t":"2026-10-11T01:55:39.466479+00:00","note":"Online net-growth and timing-credit selector: nine trend experts, positive growth admission, preserve cash, causal benchmark comparison.","status":"ok","dev":908.643159872858,"dd":-0.3609515065952231,"reason":""},{"n":"006","t":"2026-10-11T02:00:56.193574+00:00","note":"Timing-credit trend selector with 3% switch buffer: lower turnover; positive returns in all visible years and broad neighbouring settings.","status":"ok","dev":973.2554414709997,"dd":-0.35444019668696314,"reason":""}],"entry":"006","round_notes":[{"k":1,"text":"Entry: accepted #006, strategy and writeup.  \nVisible years: $100 → $973.26; max drawdown 35.4%; rank 40/40.  \nNext idea: test cross-asset lead/lag signals for earlier trend entries."},{"k":2,"text":"Final entry remains accepted #006: fulmar.py, selecting among nine BTC/ETH/SOL trend experts while rewarding loss avoidance and preserving cash decisions. Visible result: **$100 → $973.26**, with **35.4% maximum drawdown**. Another 117 evaluations found no improvement; final causality checks passed. WRITEUP.md is complete with equations, parameters, research counts, results, failed ideas and limitations. Hidden-year performance remains unknown."}],"title":"Fulmar (a40): writeup","idea":"The entry selects among nine simple trend traders for BTC, ETH and SOL. It measures each trader's recent net log growth and adds credit for returns saved by its cash decisions, measured against holding the same coin. A trader must have positive estimated net growth to qualify. The selected trader's current cash decision is respected; profitable history does not force a purchase. A switch buffer reduces turnover. This combines price trends, an online comparison of trading rules, and a benchmark that gives loss avoidance economic value. The final strategy uses four-hour open and close prices, invests the entire account in one coin or holds cash, and continually updates its statistics from completed bars. Current entry: fulmar.py, official accepted submission #006. It is self-contained, imports only numpy and pandas, and is 1,936 bytes. Its visible result is $100 → $973.26. This remains far below the growth rate required for the million-dollar objective.","failed":"- Broad daily momentum: 7/14/30/60-day momentum, risk adjustment, one- or three-coin concentration, and rank-based retention. The initial 30-variant sweep peaked at only $0.39. It repeatedly entered extreme spikes before reversals. - Spike caps, median and trimmed trends: thirty variants capped recent gains and used smoother momentum or pullbacks. Best visible result was $10.92. - Volume surprises and candle efficiency: twenty variants combined bounded positive returns with unusual volume and close location. Best result was $1.27. - Broad reversals, quiet-volume trends, drawdown and low-risk selection: thirty-two variants; best was $54.58. Isolated rebounds did not offset sustained declines and costs. - Adaptive ridge regression: three memory lengths (60/180/540 days), cross-sectional price/volume features, and four allocation variants each. Best was $94.51. - Walk-forward boosted regression: two return targets, a 540-day training window, 60-day refits, fixed seed, and four allocation variants per target. Best was $57.81. Labels used for each fit were completed before that fit. - Daily virtual experts and robust liquidity universes: sixteen daily-expert variants peaked at $11.40. Eighteen weekly strategies using volume floors, medians or lagged liquidity peaked at $156.00. - Hourly shocks, wick recovery and reversal: thirty-six variants peaked at $201.79. Short holding periods incurred costs without enough continuation or reversion. - Faster hourly expert selection: eight variants peaked at $774.96, below the retained four-hour policy. - Soft expert mixtures: generally reduced growth. Strong uncertainty penalties also reduced returns. Forcing an extra fast-price exit increased churn. Estimated Kelly sizing was usually capped at one; fractional versions added trades with","holdup":"The economic hypothesis is persistent buying and selling in liquid crypto markets, coupled with periods when staying in cash preserves capital. Three trend horizons let the algorithm observe different lengths of persistence. Online statistics continually replace old evidence, while the benchmark credit rewards trading rules that retain favorable returns and avoid adverse exposure. Preserving the selected rule's cash state limits the tendency to seek another purchase immediately after an exit. The buffer and low-cost universe make the policy less dependent on frequent trading. Selection is based on completed virtual trading rewards rather than a fixed favored coin. The same formulas can move among the three eligible assets as their trend quality changes. Simpler online estimates also keep runtime and implementation uncertainty small. Evidence is limited to a few market cycles. Extensive parameter searching makes the selected result optimistic, and causality alone does not protect against economic overfitting. The narrower liquid universe limits attainable upside; concentration creates meaningful drawdown risk; sharp gaps can defeat moving-average exits. The cost and neighborhood che","result":{"dev":{"final":973.2554414709997,"max_drawdown":-0.35444019668696314,"trades":216,"exposure":0.37636861313868614,"assets_traded":3,"fees_paid":81.53674009214693},"hidden":{"final":225.72965139923974,"max_drawdown":-0.26182373634034495,"trades":174,"exposure":0.4363013698630137,"assets_traded":3,"fees_paid":28.97450941750961},"full":{"final":2196.926115256622,"max_drawdown":-0.35444019668696314,"trades":390,"exposure":0.400328587075575,"assets_traded":3,"fees_paid":363.53272963758695}},"freq":"4h","equity":[["2021-10-10",100.8101],["2021-10-17",110.0261],["2021-10-24",113.4769],["2021-10-31",114.4914],["2021-11-07",140.6526],["2021-11-14",127.1336],["2021-11-21",121.373],["2021-11-28",121.373],["2021-12-05",113.4363],["2021-12-12",113.4363],["2021-12-19",113.4363],["2021-12-26",113.4602],["2022-01-02",111.7815],["2022-01-09",111.7815],["2022-01-16",111.7815],["2022-01-23",111.7815],["2022-01-30",111.7815],["2022-02-06",113.2792],["2022-02-13",112.8458],["2022-02-20",111.5768],["2022-02-27",111.5768],["2022-03-06",111.5768],["2022-03-13",111.5768],["2022-03-20",109.7116],["2022-03-27",120.4674],["2022-04-03",154.1916],["2022-04-10",126.1968],["2022-04-17",126.1968],["2022-04-24",126.1968],["2022-05-01",126.1968],["2022-05-08",126.1968],["2022-05-15",126.1968],["2022-05-22",126.1968],["2022-05-29",126.1968],["2022-06-05",126.1968],["2022-06-12",126.1968],["2022-06-19",126.1968],["2022-06-26",126.1968],["2022-07-03",126.1968],["2022-07-10",125.2242],["2022-07-17",128.3977],["2022-07-24",152.675],["2022-07-31",145.3522],["2022-08-07",129.6262],["2022-08-14",146.5437],["2022-08-21",132.2389],["2022-08-28",132.2389],["2022-09-04",132.2389],["2022-09-11",132.2389],["2022-09-18",122.0139],["2022-09-25",122.0139],["2022-10-02",122.0139],["2022-10-09",120.9467],["2022-10-16",120.9467],["2022-10-23",120.9467],["2022-10-30",132.8754],["2022-11-06",134.8066],["2022-11-13",123.3899],["2022-11-20",123.3899],["2022-11-27",123.3899],["2022-12-04",123.3899],["2022-12-11",121.5115],["2022-12-18",119.048],["2022-12-25",119.048],["2023-01-01",119.048],["2023-01-08",111.6963],["2023-01-15",180.0839],["2023-01-22",191.9558],["2023-01-29",201.4064],["2023-02-05",183.083],["2023-02-12",178.575],["2023-02-19",171.4016],["2023-02-26",161.7241],["2023-03-05",161.7241],["2023-03-12",161.7241],["2023-03-19",177.0674],["2023-03-26",174.3858],["2023-04-02",172.4427],["2023-04-09",166.5483],["2023-04-16",182.0843],["2023-04-23",166.3395],["2023-04-30",166.3395],["2023-05-07",166.3395],["2023-05-14",166.3395],["2023-05-21",166.3395],["2023-05-28",166.3395],["2023-06-04",167.4551],["2023-06-11",162.1658],["2023-06-18",162.1658],["2023-06-25",182.4057],["2023-07-02",183.1727],["2023-07-09",184.6191],["2023-07-16",237.3427],["2023-07-23",212.4159],["2023-07-30",200.9338],["2023-08-06",200.9338],["2023-08-13",207.4589],["2023-08-20",198.9731],["2023-08-27",198.9731],["2023-09-03",198.9731],["2023-09-10",198.9731],["2023-09-17",198.9731],["2023-09-24",191.7582],["2023-10-01",203.0034],["2023-10-08",208.4538],["2023-10-15",196.702],["2023-10-22",217.1805],["2023-10-29",249.8067],["2023-11-05",308.3978],["2023-11-12",424.1416],["2023-11-19",463.2023],["2023-11-26",404.3873],["2023-12-03",438.7488],["2023-12-10",496.3484],["2023-12-17",496.8213],["2023-12-24",777.9976],["2023-12-31",701.0491],["2024-01-07",598.0119],["2024-01-14",573.5471],["2024-01-21",541.934],["2024-01-28",541.934],["2024-02-04",537.0905],["2024-02-11",593.9859],["2024-02-18",636.7242],["2024-02-25",684.6346],["2024-03-03",763.8969],["2024-03-10",836.1758],["2024-03-17",1017.0454],["2024-03-24",915.1329],["2024-03-31",1040.0679],["2024-04-07",947.8566],["2024-04-14",947.8566],["2024-04-21",947.8566],["2024-04-28",947.8566],["2024-05-05",947.8566],["2024-05-12",947.8566],["2024-05-19",914.9533],["2024-05-26",1046.9157],["2024-06-02",1020.8727],["2024-06-09",971.7523],["2024-06-16",971.7523],["2024-06-23",971.7523],["2024-06-30",971.7523],["2024-07-07",955.657],["2024-07-14",955.657],["2024-07-21",1046.7225],["2024-07-28",1086.7141],["2024-08-04",1008.7792],["2024-08-11",1008.7792],["2024-08-18",1008.7792],["2024-08-25",1016.6182],["2024-09-01",994.1874],["2024-09-08",994.1874],["2024-09-15",994.1874],["2024-09-22",1015.7086],["2024-09-29",1029.0106],["2024-10-06",973.2554],["2024-10-13",973.2554],["2024-10-20",994.6137],["2024-10-27",1058.8293],["2024-11-03",977.5668],["2024-11-10",1047.5269],["2024-11-17",1177.9283],["2024-11-24",1166.851],["2024-12-01",1223.1864],["2024-12-08",1316.2228],["2024-12-15",1196.0978],["2024-12-22",1125.9441],["2024-12-29",1106.0933],["2025-01-05",1106.5204],["2025-01-12",1043.4747],["2025-01-19",1248.9832],["2025-01-26",1187.7158],["2025-02-02",1080.7379],["2025-02-09",1080.7379],["2025-02-16",1080.7379],["2025-02-23",1080.7379],["2025-03-02",1080.7379],["2025-03-09",1080.7379],["2025-03-16",1080.7379],["2025-03-23",1080.7379],["2025-03-30",1062.1368],["2025-04-06",1062.1368],["2025-04-13",1062.1368],["2025-04-20",1047.0207],["2025-04-27",1111.1812],["2025-05-04",1124.7197],["2025-05-11",1420.7888],["2025-05-18",1364.241],["2025-05-25",1438.2467],["2025-06-01",1471.9911],["2025-06-08",1488.1187],["2025-06-15",1432.5968],["2025-06-22",1432.5968],["2025-06-29",1432.5968],["2025-07-06",1432.5968],["2025-07-13",1614.9382],["2025-07-20",2025.7589],["2025-07-27",1983.8867],["2025-08-03",1821.4691],["2025-08-10",2088.2023],["2025-08-17",2191.7713],["2025-08-24",2284.4789],["2025-08-31",2099.1844],["2025-09-07",2026.1573],["2025-09-14",2290.7995],["2025-09-21",2248.1027],["2025-09-28",2093.8284],["2025-10-05",2138.5045],["2025-10-12",2074.104],["2025-10-19",2074.104],["2025-10-26",2074.104],["2025-11-02",2074.104],["2025-11-09",2074.104],["2025-11-16",2074.104],["2025-11-23",2074.104],["2025-11-30",2074.104],["2025-12-07",2074.104],["2025-12-14",2074.104],["2025-12-21",2074.104],["2025-12-28",2074.104],["2026-01-04",2091.9684],["2026-01-11",2183.911],["2026-01-18",2228.3943],["2026-01-25",2105.9611],["2026-02-01",2105.9611],["2026-02-08",2105.9611],["2026-02-15",2105.9611],["2026-02-22",2105.9611],["2026-03-01",2105.9611],["2026-03-08",2071.8593],["2026-03-15",2060.4102],["2026-03-22",1973.8913],["2026-03-29",1973.8913],["2026-04-05",1973.8913],["2026-04-12",1941.554],["2026-04-19",1906.79],["2026-04-26",1962.9006],["2026-05-03",1928.5023],["2026-05-10",1994.063],["2026-05-17",1891.8639],["2026-05-24",1891.8639],["2026-05-31",1891.8639],["2026-06-07",1891.8639],["2026-06-14",1891.8639],["2026-06-21",1891.8639],["2026-06-28",1891.8639],["2026-07-05",1833.0651],["2026-07-12",1773.5405],["2026-07-19",1792.1463],["2026-07-26",1832.3774],["2026-08-02",1769.4543],["2026-08-09",1769.4543],["2026-08-16",1769.4543],["2026-08-23",2072.04],["2026-08-30",2201.4432],["2026-09-06",2194.3699],["2026-09-13",2095.7321],["2026-09-20",2080.4532],["2026-09-27",2317.412],["2026-10-04",2281.0582],["2026-10-11",2196.9261]],"code_bytes":1936}},"refs":{"hold_btc":{"dev":110.23794944804915,"hidden":136.14516687855453,"full":150.23369910346588,"equity":[["2021-10-17",110.6735],["2021-10-24",111.421],["2021-10-31",112.4617],["2021-11-07",111.7553],["2021-11-14",117.0446],["2021-11-21",108.5499],["2021-11-28",99.4843],["2021-12-05",89.3605],["2021-12-12",89.7924],["2021-12-19",85.1464],["2021-12-26",91.628],["2022-01-02",86.7611],["2022-01-09",75.7749],["2022-01-16",78.3284],["2022-01-23",63.7608],["2022-01-30",69.3884],["2022-02-06",75.2347],["2022-02-13",76.7532],["2022-02-20",72.865],["2022-02-27",71.1153],["2022-03-06",71.6266],["2022-03-13",70.5528],["2022-03-20",76.7228],["2022-03-27",80.9227],["2022-04-03",83.2857],["2022-04-10",77.7279],["2022-04-17",73.4096],["2022-04-24",71.7059],["2022-05-01",68.4138],["2022-05-08",64.4898],["2022-05-15",54.6985],["2022-05-22",53.5319],["2022-05-29",52.7798],["2022-06-05",54.2936],["2022-06-12",51.6769],["2022-06-19",34.4894],["2022-06-26",39.0716],["2022-07-03",35.0021],["2022-07-10",39.2545],["2022-07-17",38.5342],["2022-07-24",40.8167],["2022-07-31",42.9846],["2022-08-07",41.7314],["2022-08-14",44.4351],["2022-08-21",38.4332],["2022-08-28",36.4289],["2022-09-04",36.0549],["2022-09-11",39.3573],["2022-09-18",36.5671],["2022-09-25",34.398],["2022-10-02",35.1078],["2022-10-09",35.3024],["2022-10-16",34.6687],["2022-10-23",34.914],["2022-10-30",37.8326],["2022-11-06",38.7228],["2022-11-13",30.5648],["2022-11-20",30.3623],["2022-11-27",29.9221],["2022-12-04",30.6978],["2022-12-11",31.1382],["2022-12-18",30.5002],["2022-12-25",30.6085],["2023-01-01",30.0745],["2023-01-08",30.8039],["2023-01-15",38.0966],["2023-01-22",41.4211],["2023-01-29",41.8557],["2023-02-05",42.4088],["2023-02-12",39.7467],["2023-02-19",44.7816],["2023-02-26",42.1002],["2023-03-05",40.6267],["2023-03-12",37.1891],["2023-03-19",48.9186],["2023-03-26",49.9284],["2023-04-02",51.7278],["2023-04-09",50.7927],["2023-04-16",55.0773],["2023-04-23",50.5718],["2023-04-30",53.1418],["2023-05-07",52.4468],["2023-05-14",48.6782],["2023-05-21",49.273],["2023-05-28",48.8218],["2023-06-04",49.2126],["2023-06-11",46.98],["2023-06-18",48.2086],["2023-06-25",55.4997],["2023-07-02",55.606],["2023-07-09",55.0583],["2023-07-16",55.0672],["2023-07-23",54.1663],["2023-07-30",53.3644],["2023-08-06",52.8539],["2023-08-13",53.5049],["2023-08-20",47.4505],["2023-08-27",47.3003],["2023-09-03",47.0315],["2023-09-10",47.0898],["2023-09-17",48.2862],["2023-09-24",48.3158],["2023-10-01",49.0187],["2023-10-08",50.826],["2023-10-15",48.8185],["2023-10-22",54.3768],["2023-10-29",61.9602],["2023-11-05",63.7438],["2023-11-12",67.5034],["2023-11-19",66.4818],["2023-11-26",68.6863],["2023-12-03",71.7218],["2023-12-10",79.4725],["2023-12-17",76.8626],["2023-12-24",79.4517],["2023-12-31",76.6122],["2024-01-07",79.9356],["2024-01-14",77.8988],["2024-01-21",75.8045],["2024-01-28",76.5765],["2024-02-04",78.1953],["2024-02-11",86.8128],["2024-02-18",93.8877],["2024-02-25",93.7524],["2024-03-03",112.6946],["2024-03-10",124.1954],["2024-03-17",118.7183],["2024-03-24",116.3356],["2024-03-31",126.5023],["2024-04-07",125.2548],["2024-04-14",116.2165],["2024-04-21",118.0638],["2024-04-28",115.3756],["2024-05-05",116.1575],["2024-05-12",110.5833],["2024-05-19",121.6537],["2024-05-26",125.9722],["2024-06-02",123.202],["2024-06-09",126.0083],["2024-06-16",120.4048],["2024-06-23",116.8301],["2024-06-30",110.8755],["2024-07-07",105.864],["2024-07-14",107.6345],["2024-07-21",122.0623],["2024-07-28",123.4377],["2024-08-04",110.3506],["2024-08-11",110.7606],["2024-08-18",108.1581],["2024-08-25",116.6392],["2024-09-01",107.2163],["2024-09-08",98.4659],["2024-09-15",109.069],["2024-09-22",115.1701],["2024-09-29",119.7317],["2024-10-06",112.8231],["2024-10-13",114.9106],["2024-10-20",124.3131],["2024-10-27",121.9765],["2024-11-03",126.1252],["2024-11-10",139.4017],["2024-11-17",164.6895],["2024-11-24",177.5711],["2024-12-01",175.2724],["2024-12-08",181.4973],["2024-12-15",184.3843],["2024-12-22",176.8795],["2024-12-29",173.258],["2025-01-05",178.5676],["2025-01-12",171.9854],["2025-01-19",190.0861],["2025-01-26",190.4326],["2025-02-02",182.9584],["2025-02-09",175.3392],["2025-02-16",177.3843],["2025-02-23",175.5324],["2025-03-02",156.4677],["2025-03-09",156.7548],["2025-03-16",153.3296],["2025-03-23",152.4245],["2025-03-30",150.2573],["2025-04-06",151.8744],["2025-04-13",155.0357],["2025-04-20",154.6723],["2025-04-27",172.0363],["2025-05-04",174.2696],["2025-05-11",190.5466],["2025-05-18",187.4871],["2025-05-25",195.9141],["2025-06-01",190.1509],["2025-06-08",191.8967],["2025-06-15",191.6467],["2025-06-22",185.657],["2025-06-29",195.0685],["2025-07-06",196.7072],["2025-07-13",213.4728],["2025-07-20",214.2364],["2025-07-27",214.3818],["2025-08-03",204.6124],["2025-08-10",211.7316],["2025-08-17",213.4013],["2025-08-24",209.8695],["2025-08-31",197.8311],["2025-09-07",200.3248],["2025-09-14",210.7426],["2025-09-21",210.3197],["2025-09-28",199.321],["2025-10-05",222.5102],["2025-10-12",201.1546],["2025-10-19",194.8653],["2025-10-26",202.976],["2025-11-02",200.1614],["2025-11-09",186.0077],["2025-11-16",173.7966],["2025-11-23",154.0592],["2025-11-30",165.0813],["2025-12-07",162.2349],["2025-12-14",164.0588],["2025-12-21",160.6425],["2025-12-28",159.7628],["2026-01-04",164.7642],["2026-01-11",164.54],["2026-01-18",172.9813],["2026-01-25",162.2141],["2026-02-01",143.1535],["2026-02-08",125.97],["2026-02-15",126.94],["2026-02-22",123.5821],["2026-03-01",121.7592],["2026-03-08",122.2858],["2026-03-15",129.4653],["2026-03-22",125.295],["2026-03-29",120.6753],["2026-04-05",122.354],["2026-04-12",132.7945],["2026-04-19",137.6097],["2026-04-26",141.1244],["2026-05-03",143.0549],["2026-05-10",146.6755],["2026-05-17",142.0753],["2026-05-24",139.5373],["2026-05-31",134.3238],["2026-06-07",110.6899],["2026-06-14",117.1864],["2026-06-21",116.8955],["2026-06-28",109.1344],["2026-07-05",114.7975],["2026-07-12",116.0247],["2026-07-19",117.8704],["2026-07-26",117.0355],["2026-08-02",114.2151],["2026-08-09",118.1038],["2026-08-16",114.6921],["2026-08-23",140.1243],["2026-08-30",142.2243],["2026-09-06",145.1363],["2026-09-13",140.4949],["2026-09-20",147.7147],["2026-09-27",153.5017],["2026-10-04",154.0843],["2026-10-11",150.2337]]},"hold_eth":{"dev":66.24376258736494,"hidden":104.84059217010909,"full":69.51980332532695,"equity":[["2021-10-17",107.0298],["2021-10-24",116.4519],["2021-10-31",120.7991],["2021-11-07",126.2574],["2021-11-14",129.7863],["2021-11-21",123.2999],["2021-11-28",114.4446],["2021-12-05",115.0583],["2021-12-12",114.2791],["2021-12-19",110.6619],["2021-12-26",114.4317],["2022-01-02",105.2296],["2022-01-09",86.0984],["2022-01-16",92.9476],["2022-01-23",67.4094],["2022-01-30",72.698],["2022-02-06",84.1897],["2022-02-13",81.5693],["2022-02-20",77.2308],["2022-02-27",77.6734],["2022-03-06",74.4787],["2022-03-13",71.7862],["2022-03-20",82.4624],["2022-03-27",87.8883],["2022-04-03",96.2376],["2022-04-10",91.0621],["2022-04-17",85.4934],["2022-04-24",81.9647],["2022-05-01",76.1976],["2022-05-08",73.6457],["2022-05-15",57.4692],["2022-05-22",55.1941],["2022-05-29",50.0843],["2022-06-05",50.4297],["2022-06-12",42.8372],["2022-06-19",27.8093],["2022-06-26",34.7183],["2022-07-03",29.818],["2022-07-10",34.0101],["2022-07-17",37.8898],["2022-07-24",43.2782],["2022-07-31",47.4234],["2022-08-07",47.2445],["2022-08-14",55.4311],["2022-08-21",44.0431],["2022-08-28",41.6683],["2022-09-04",43.5239],["2022-09-11",49.5785],["2022-09-18",41.0471],["2022-09-25",36.8077],["2022-10-02",36.6585],["2022-10-09",36.7616],["2022-10-16",35.6282],["2022-10-23",36.7094],["2022-10-30",45.259],["2022-11-06",45.4666],["2022-11-13",35.1344],["2022-11-20",34.0283],["2022-11-27",33.691],["2022-12-04",34.6666],["2022-12-11",35.3887],["2022-12-18",33.1757],["2022-12-25",34.1049],["2023-01-01",33.4263],["2023-01-08",35.325],["2023-01-15",43.3126],["2023-01-22",45.4577],["2023-01-29",43.933],["2023-02-05",46.5677],["2023-02-12",42.9943],["2023-02-19",47.273],["2023-02-26",44.5612],["2023-03-05",43.7829],["2023-03-12",41.1348],["2023-03-19",49.1515],["2023-03-26",48.6776],["2023-04-02",50.8827],["2023-04-09",51.666],["2023-04-16",58.4227],["2023-04-23",52.3742],["2023-04-30",53.3143],["2023-05-07",52.9937],["2023-05-14",50.1651],["2023-05-21",50.8444],["2023-05-28",51.137],["2023-06-04",52.8744],["2023-06-11",48.9472],["2023-06-18",48.283],["2023-06-25",52.3965],["2023-07-02",53.7809],["2023-07-09",52.1249],["2023-07-16",53.974],["2023-07-23",52.1607],["2023-07-30",52.5502],["2023-08-06",51.3212],["2023-08-13",51.6971],["2023-08-20",46.6596],["2023-08-27",46.0272],["2023-09-03",45.7475],["2023-09-10",45.7058],["2023-09-17",45.674],["2023-09-24",44.5411],["2023-10-01",46.6937],["2023-10-08",45.6508],["2023-10-15",43.4534],["2023-10-22",45.5214],["2023-10-29",49.6366],["2023-11-05",51.8538],["2023-11-12",57.3767],["2023-11-19",54.8457],["2023-11-26",58.2128],["2023-12-03",60.4946],["2023-12-10",65.406],["2023-12-17",62.2892],["2023-12-24",64.5036],["2023-12-31",64.0419],["2024-01-07",62.6195],["2024-01-14",72.0486],["2024-01-21",69.0813],["2024-01-28",63.3785],["2024-02-04",64.1764],["2024-02-11",69.8702],["2024-02-18",77.854],["2024-02-25",83.63],["2024-03-03",95.6124],["2024-03-10",109.1327],["2024-03-17",98.4542],["2024-03-24",93.0451],["2024-03-31",97.9666],["2024-04-07",93.6616],["2024-04-14",84.0321],["2024-04-21",88.1899],["2024-04-28",90.978],["2024-05-05",87.1123],["2024-05-12",81.3896],["2024-05-19",87.2718],["2024-05-26",104.7746],["2024-06-02",106.6347],["2024-06-09",102.883],["2024-06-16",99.7299],["2024-06-23",97.6902],["2024-06-30",94.422],["2024-07-07",85.7038],["2024-07-14",88.7527],["2024-07-21",98.298],["2024-07-28",90.7949],["2024-08-04",81.1434],["2024-08-11",72.9353],["2024-08-18",73.0636],["2024-08-25",77.3529],["2024-09-01",70.2271],["2024-09-08",63.5361],["2024-09-15",67.5661],["2024-09-22",73.0046],["2024-09-29",74.7598],["2024-10-06",67.4787],["2024-10-13",69.204],["2024-10-20",74.005],["2024-10-27",69.3748],["2024-11-03",69.7023],["2024-11-10",87.3632],["2024-11-17",87.5493],["2024-11-24",94.8442],["2024-12-01",103.4986],["2024-12-08",111.676],["2024-12-15",108.1569],["2024-12-22",93.3075],["2024-12-29",95.1262],["2025-01-05",102.193],["2025-01-12",91.74],["2025-01-19",92.4353],["2025-01-26",92.7444],["2025-02-02",87.1209],["2025-02-09",73.5652],["2025-02-16",75.2581],["2025-02-23",77.2193],["2025-03-02",61.9659],["2025-03-09",61.58],["2025-03-16",54.135],["2025-03-23",55.3512],["2025-03-30",51.0865],["2025-04-06",50.4697],["2025-04-13",45.9473],["2025-04-20",45.0835],["2025-04-27",50.8852],["2025-05-04",51.2385],["2025-05-11",72.1894],["2025-05-18",69.1671],["2025-05-25",70.7131],["2025-06-01",70.6477],["2025-06-08",70.5518],["2025-06-15",70.7231],["2025-06-22",64.1551],["2025-06-29",68.0644],["2025-07-06",70.3221],["2025-07-13",82.2512],["2025-07-20",100.3802],["2025-07-27",104.5466],["2025-08-03",94.8451],["2025-08-10",119.0648],["2025-08-17",123.5743],["2025-08-24",133.5344],["2025-08-31",122.2249],["2025-09-07",119.4147],["2025-09-14",130.4081],["2025-09-21",125.2072],["2025-09-28",112.2953],["2025-10-05",125.3953],["2025-10-12",104.7056],["2025-10-19",108.6856],["2025-10-26",110.4911],["2025-11-02",108.2541],["2025-11-09",95.0566],["2025-11-16",88.5277],["2025-11-23",77.4121],["2025-11-30",83.5333],["2025-12-07",84.8998],["2025-12-14",87.0399],["2025-12-21",83.2195],["2025-12-28",82.4124],["2026-01-04",87.3884],["2026-01-11",86.2597],["2026-01-18",92.515],["2026-01-25",82.529],["2026-02-01",68.5208],["2026-02-08",58.3243],["2026-02-15",58.3106],["2026-02-22",55.1461],["2026-03-01",54.899],["2026-03-08",55.041],["2026-03-15",58.5781],["2026-03-22",58.2626],["2026-03-29",55.7438],["2026-04-05",57.7218],["2026-04-12",63.855],["2026-04-19",65.6807],["2026-04-26",64.8194],["2026-05-03",64.7487],["2026-05-10",65.0133],["2026-05-17",60.9425],["2026-05-24",59.1794],["2026-05-31",56.5179],["2026-06-07",43.8656],["2026-06-14",46.9813],["2026-06-21",48.6552],["2026-06-28",43.9858],["2026-07-05",49.7607],["2026-07-12",49.9597],["2026-07-19",52.0514],["2026-07-26",52.3946],["2026-08-02",51.5576],["2026-08-09",53.5641],["2026-08-16",52.6112],["2026-08-23",67.7006],["2026-08-30",68.6806],["2026-09-06",69.3141],["2026-09-13",70.5882],["2026-09-20",73.5716],["2026-09-27",75.3501],["2026-10-04",75.0958],["2026-10-11",69.5198]]},"sma200_btc":{"dev":230.63853700985013,"hidden":139.75554324083487,"full":322.3301403208302,"equity":[["2021-10-17",100.0],["2021-10-24",100.0],["2021-10-31",100.0],["2021-11-07",100.0],["2021-11-14",100.0],["2021-11-21",100.0],["2021-11-28",100.0],["2021-12-05",100.0],["2021-12-12",100.0],["2021-12-19",95.2515],["2021-12-26",98.0995],["2022-01-02",92.4481],["2022-01-09",92.4481],["2022-01-16",92.4481],["2022-01-23",92.4481],["2022-01-30",92.4481],["2022-02-06",92.4481],["2022-02-13",92.4481],["2022-02-20",92.4481],["2022-02-27",92.4481],["2022-03-06",92.4481],["2022-03-13",92.4481],["2022-03-20",92.4481],["2022-03-27",92.4481],["2022-04-03",92.4481],["2022-04-10",92.4481],["2022-04-17",92.4481],["2022-04-24",92.4481],["2022-05-01",92.4481],["2022-05-08",92.4481],["2022-05-15",92.4481],["2022-05-22",92.4481],["2022-05-29",92.4481],["2022-06-05",92.4481],["2022-06-12",92.4481],["2022-06-19",92.4481],["2022-06-26",92.4481],["2022-07-03",92.4481],["2022-07-10",92.4481],["2022-07-17",92.4481],["2022-07-24",92.4481],["2022-07-31",92.4481],["2022-08-07",92.4481],["2022-08-14",92.4481],["2022-08-21",92.4481],["2022-08-28",92.4481],["2022-09-04",92.4481],["2022-09-11",92.4481],["2022-09-18",92.4481],["2022-09-25",92.4481],["2022-10-02",92.4481],["2022-10-09",92.4481],["2022-10-16",92.4481],["2022-10-23",92.4481],["2022-10-30",92.4481],["2022-11-06",92.4481],["2022-11-13",92.4481],["2022-11-20",92.4481],["2022-11-27",92.4481],["2022-12-04",92.4481],["2022-12-11",92.4481],["2022-12-18",92.4481],["2022-12-25",92.4481],["2023-01-01",92.4481],["2023-01-08",92.4481],["2023-01-15",97.1051],["2023-01-22",105.579],["2023-01-29",106.6868],["2023-02-05",108.0966],["2023-02-12",101.3111],["2023-02-19",114.1445],["2023-02-26",107.3099],["2023-03-05",103.554],["2023-03-12",94.7919],["2023-03-19",124.6894],["2023-03-26",127.2634],["2023-04-02",131.85],["2023-04-09",129.4665],["2023-04-16",140.3875],["2023-04-23",128.9033],["2023-04-30",135.454],["2023-05-07",133.6826],["2023-05-14",124.0767],["2023-05-21",125.5927],["2023-05-28",124.4427],["2023-06-04",125.4388],["2023-06-11",119.7482],["2023-06-18",122.8798],["2023-06-25",141.4642],["2023-07-02",141.7351],["2023-07-09",140.339],["2023-07-16",140.3617],["2023-07-23",138.0655],["2023-07-30",136.0214],["2023-08-06",134.7203],["2023-08-13",136.3795],["2023-08-20",123.2496],["2023-08-27",123.2496],["2023-09-03",121.1556],["2023-09-10",121.1556],["2023-09-17",121.1556],["2023-09-24",121.1556],["2023-10-01",121.1556],["2023-10-08",121.1556],["2023-10-15",121.1556],["2023-10-22",127.0183],["2023-10-29",144.7322],["2023-11-05",148.8985],["2023-11-12",157.6804],["2023-11-19",155.2942],["2023-11-26",160.4436],["2023-12-03",167.5342],["2023-12-10",185.639],["2023-12-17",179.5426],["2023-12-24",185.5905],["2023-12-31",178.9576],["2024-01-07",186.7208],["2024-01-14",181.963],["2024-01-21",177.071],["2024-01-28",178.8742],["2024-02-04",182.6557],["2024-02-11",202.7851],["2024-02-18",219.3114],["2024-02-25",218.9953],["2024-03-03",263.2421],["2024-03-10",290.1067],["2024-03-17",277.3129],["2024-03-24",271.7471],["2024-03-31",295.4954],["2024-04-07",292.5814],["2024-04-14",271.4689],["2024-04-21",275.7839],["2024-04-28",269.5047],["2024-05-05",271.331],["2024-05-12",258.3104],["2024-05-19",284.1695],["2024-05-26",294.257],["2024-06-02",287.7862],["2024-06-09",294.3415],["2024-06-16",281.2522],["2024-06-23",272.9022],["2024-06-30",258.9928],["2024-07-07",242.0327],["2024-07-14",242.0327],["2024-07-21",274.2015],["2024-07-28",277.2913],["2024-08-04",247.8923],["2024-08-11",247.6445],["2024-08-18",247.6445],["2024-08-25",247.8598],["2024-09-01",242.5058],["2024-09-08",242.5058],["2024-09-15",242.5058],["2024-09-22",242.5058],["2024-09-29",240.0943],["2024-10-06",230.6385],["2024-10-13",230.6385],["2024-10-20",238.4064],["2024-10-27",233.9253],["2024-11-03",241.8816],["2024-11-10",267.3432],["2024-11-17",315.8399],["2024-11-24",340.5441],["2024-12-01",336.1356],["2024-12-08",348.0737],["2024-12-15",353.6104],["2024-12-22",339.2177],["2024-12-29",332.2725],["2025-01-05",342.4551],["2025-01-12",329.8318],["2025-01-19",364.5452],["2025-01-26",365.2098],["2025-02-02",350.8757],["2025-02-09",336.2637],["2025-02-16",340.1859],["2025-02-23",336.6342],["2025-03-02",300.0721],["2025-03-09",300.6227],["2025-03-16",272.9227],["2025-03-23",258.3689],["2025-03-30",252.8859],["2025-04-06",252.8859],["2025-04-13",252.8859],["2025-04-20",252.8859],["2025-04-27",255.8371],["2025-05-04",259.1582],["2025-05-11",283.3639],["2025-05-18",278.8141],["2025-05-25",291.346],["2025-06-01",282.7755],["2025-06-08",285.3717],["2025-06-15",284.9999],["2025-06-22",276.0925],["2025-06-29",290.0885],["2025-07-06",292.5254],["2025-07-13",317.4577],["2025-07-20",318.5932],["2025-07-27",318.8095],["2025-08-03",304.2813],["2025-08-10",314.8683],["2025-08-17",317.3514],["2025-08-24",312.0993],["2025-08-31",294.1968],["2025-09-07",297.9051],["2025-09-14",313.3977],["2025-09-21",312.7687],["2025-09-28",296.4124],["2025-10-05",330.8973],["2025-10-12",299.1391],["2025-10-19",287.4618],["2025-10-26",287.8065],["2025-11-02",279.9274],["2025-11-09",270.7193],["2025-11-16",270.7193],["2025-11-23",270.7193],["2025-11-30",270.7193],["2025-12-07",270.7193],["2025-12-14",270.7193],["2025-12-21",270.7193],["2025-12-28",270.7193],["2026-01-04",270.7193],["2026-01-11",270.7193],["2026-01-18",270.7193],["2026-01-25",270.7193],["2026-02-01",270.7193],["2026-02-08",270.7193],["2026-02-15",270.7193],["2026-02-22",270.7193],["2026-03-01",270.7193],["2026-03-08",270.7193],["2026-03-15",270.7193],["2026-03-22",270.7193],["2026-03-29",270.7193],["2026-04-05",270.7193],["2026-04-12",270.7193],["2026-04-19",270.7193],["2026-04-26",270.7193],["2026-05-03",270.7193],["2026-05-10",270.7193],["2026-05-17",270.7193],["2026-05-24",270.7193],["2026-05-31",270.7193],["2026-06-07",270.7193],["2026-06-14",270.7193],["2026-06-21",270.7193],["2026-06-28",270.7193],["2026-07-05",270.7193],["2026-07-12",270.7193],["2026-07-19",270.7193],["2026-07-26",270.7193],["2026-08-02",270.7193],["2026-08-09",270.7193],["2026-08-16",270.7193],["2026-08-23",300.6403],["2026-08-30",305.1457],["2026-09-06",311.3936],["2026-09-13",301.4352],["2026-09-20",316.9256],["2026-09-27",329.3417],["2026-10-04",330.5917],["2026-10-11",322.3301]]}},"board":[{"time":"2026-10-11T00:01:33+00:00","agent":"a02","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:01:38+00:00","agent":"a01","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:02:03+00:00","agent":"a05","n":"001","status":"ok","final":195.63220430662435},{"time":"2026-10-11T00:02:24+00:00","agent":"a06","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:02:39+00:00","agent":"a07","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:02:44+00:00","agent":"a03","n":"001","status":"ok","final":580.9942833608492},{"time":"2026-10-11T00:03:13+00:00","agent":"a04","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:03:38+00:00","agent":"a08","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:06:49+00:00","agent":"a01","n":"002","status":"ok","final":847.4906239849815},{"time":"2026-10-11T00:07:36+00:00","agent":"a05","n":"002","status":"ok","final":9486.23825477442},{"time":"2026-10-11T00:07:55+00:00","agent":"a03","n":"002","status":"ok","final":724.9344425326136},{"time":"2026-10-11T00:07:57+00:00","agent":"a06","n":"002","status":"ok","final":855.0070241385579},{"time":"2026-10-11T00:08:04+00:00","agent":"a02","n":"002","status":"ok","final":394.1268504757573},{"time":"2026-10-11T00:09:04+00:00","agent":"a07","n":"002","status":"ok","final":852.5377619338178},{"time":"2026-10-11T00:09:08+00:00","agent":"a08","n":"002","status":"ok","final":775.3038971660925},{"time":"2026-10-11T00:11:59+00:00","agent":"a04","n":"002","status":"ok","final":1003.9368722721639},{"time":"2026-10-11T00:12:55+00:00","agent":"a05","n":"003","status":"ok","final":95065.191677513},{"time":"2026-10-11T00:13:49+00:00","agent":"a03","n":"003","status":"ok","final":1301.9166738222118},{"time":"2026-10-11T00:14:18+00:00","agent":"a02","n":"003","status":"ok","final":1748.815475399441},{"time":"2026-10-11T00:15:39+00:00","agent":"a07","n":"003","status":"ok","final":2081.1820995139497},{"time":"2026-10-11T00:18:35+00:00","agent":"a01","n":"003","status":"ok","final":1752.9433423982277},{"time":"2026-10-11T00:19:29+00:00","agent":"a02","n":"004","status":"ok","final":3083.4294181629475},{"time":"2026-10-11T00:20:45+00:00","agent":"a05","n":"004","status":"ok","final":97728.08912314363},{"time":"2026-10-11T00:23:22+00:00","agent":"a06","n":"003","status":"ok","final":2255.122113212569},{"time":"2026-10-11T00:24:18+00:00","agent":"a03","n":"004","status":"ok","final":1526.8816197017704},{"time":"2026-10-11T00:26:11+00:00","agent":"a01","n":"004","status":"ok","final":3354.4014532938413},{"time":"2026-10-11T00:27:45+00:00","agent":"a04","n":"003","status":"ok","final":1522.7794867035561},{"time":"2026-10-11T00:27:54+00:00","agent":"a08","n":"003","status":"ok","final":1357.5498405670726},{"time":"2026-10-11T00:27:56+00:00","agent":"a05","n":"005","status":"ok","final":128010.99398593127},{"time":"2026-10-11T00:29:59+00:00","agent":"a10","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:30:09+00:00","agent":"a09","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:30:15+00:00","agent":"a11","n":"001","status":"ok","final":6.7010784932997325},{"time":"2026-10-11T00:30:40+00:00","agent":"a12","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:31:04+00:00","agent":"a02","n":"005","status":"ok","final":3415.1939311053757},{"time":"2026-10-11T00:31:07+00:00","agent":"a16","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:31:09+00:00","agent":"a13","n":"001","status":"ok","final":1.8905847779942555},{"time":"2026-10-11T00:31:20+00:00","agent":"a17","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:31:20+00:00","agent":"a14","n":"001","status":"ok","final":122.4224776062087},{"time":"2026-10-11T00:31:35+00:00","agent":"a15","n":"001","status":"ok","final":0.002415035628257502},{"time":"2026-10-11T00:31:55+00:00","agent":"a03","n":"005","status":"ok","final":1777.330694881204},{"time":"2026-10-11T00:32:12+00:00","agent":"a20","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:32:12+00:00","agent":"a18","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:32:19+00:00","agent":"a21","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:32:34+00:00","agent":"a22","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:32:40+00:00","agent":"a19","n":"001","status":"ok","final":171.0181109430941},{"time":"2026-10-11T00:32:45+00:00","agent":"a23","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:33:02+00:00","agent":"a24","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:36:00+00:00","agent":"a11","n":"002","status":"ok","final":549.2410965045042},{"time":"2026-10-11T00:36:35+00:00","agent":"a13","n":"002","status":"ok","final":425.4868284581931},{"time":"2026-10-11T00:36:53+00:00","agent":"a17","n":"002","status":"ok","final":1021.5414394283893},{"time":"2026-10-11T00:36:57+00:00","agent":"a15","n":"002","status":"ok","final":230.63853700985013},{"time":"2026-10-11T00:37:34+00:00","agent":"a16","n":"002","status":"ok","final":669.1603875387693},{"time":"2026-10-11T00:37:56+00:00","agent":"a01","n":"005","status":"ok","final":4254.430828567441},{"time":"2026-10-11T00:38:01+00:00","agent":"a14","n":"002","status":"ok","final":648.7685949005431},{"time":"2026-10-11T00:38:23+00:00","agent":"a09","n":"002","status":"ok","final":984.2093517139671},{"time":"2026-10-11T00:38:32+00:00","agent":"a24","n":"002","status":"ok","final":567.7506065246486},{"time":"2026-10-11T00:38:50+00:00","agent":"a07","n":"004","status":"ok","final":5205.576771765608},{"time":"2026-10-11T00:39:06+00:00","agent":"a18","n":"002","status":"ok","final":485.8562839282218},{"time":"2026-10-11T00:39:18+00:00","agent":"a19","n":"002","status":"ok","final":464.4770846481689},{"time":"2026-10-11T00:39:19+00:00","agent":"a12","n":"002","status":"ok","final":812.0834868040389},{"time":"2026-10-11T00:39:32+00:00","agent":"a20","n":"002","status":"ok","final":516.0767493839754},{"time":"2026-10-11T00:39:58+00:00","agent":"a10","n":"002","status":"ok","final":1000.0662243637282},{"time":"2026-10-11T00:40:13+00:00","agent":"a21","n":"002","status":"ok","final":661.919638782842},{"time":"2026-10-11T00:41:57+00:00","agent":"a15","n":"003","status":"ok","final":1007.875952071391},{"time":"2026-10-11T00:42:40+00:00","agent":"a13","n":"003","status":"ok","final":1137.5600330898408},{"time":"2026-10-11T00:43:41+00:00","agent":"a16","n":"003","status":"ok","final":1741.540032456692},{"time":"2026-10-11T00:43:52+00:00","agent":"a24","n":"003","status":"ok","final":1046.2095638548703},{"time":"2026-10-11T00:44:25+00:00","agent":"a11","n":"003","status":"ok","final":743.2938321655027},{"time":"2026-10-11T00:44:33+00:00","agent":"a14","n":"003","status":"ok","final":842.404416957649},{"time":"2026-10-11T00:44:53+00:00","agent":"a12","n":"003","status":"ok","final":2545.9150039856013},{"time":"2026-10-11T00:44:59+00:00","agent":"a18","n":"003","status":"ok","final":3834.582181506042},{"time":"2026-10-11T00:45:10+00:00","agent":"a03","n":"006","status":"ok","final":1913.4592191261152},{"time":"2026-10-11T00:45:33+00:00","agent":"a20","n":"003","status":"ok","final":1138.7571133396855},{"time":"2026-10-11T00:45:52+00:00","agent":"a22","n":"002","status":"rejected","final":null},{"time":"2026-10-11T00:46:20+00:00","agent":"a09","n":"003","status":"ok","final":1726.013692640102},{"time":"2026-10-11T00:47:46+00:00","agent":"a21","n":"003","status":"ok","final":1257.7838932653854},{"time":"2026-10-11T00:48:15+00:00","agent":"a10","n":"003","status":"ok","final":1402.8971986254423},{"time":"2026-10-11T00:48:15+00:00","agent":"a17","n":"003","status":"ok","final":1497.78586315637},{"time":"2026-10-11T00:48:33+00:00","agent":"a06","n":"004","status":"ok","final":4739.406588980694},{"time":"2026-10-11T00:48:50+00:00","agent":"a23","n":"002","status":"ok","final":886.5425573980995},{"time":"2026-10-11T00:49:32+00:00","agent":"a24","n":"004","status":"ok","final":1319.7375254883773},{"time":"2026-10-11T00:49:49+00:00","agent":"a14","n":"004","status":"ok","final":1398.7581300016698},{"time":"2026-10-11T00:49:53+00:00","agent":"a16","n":"004","status":"rejected","final":null},{"time":"2026-10-11T00:49:59+00:00","agent":"a11","n":"004","status":"ok","final":798.6242930716215},{"time":"2026-10-11T00:50:38+00:00","agent":"a15","n":"004","status":"ok","final":1389.107050514133},{"time":"2026-10-11T00:50:38+00:00","agent":"a18","n":"004","status":"ok","final":8996.258180020943},{"time":"2026-10-11T00:50:39+00:00","agent":"a12","n":"004","status":"ok","final":5395.141023970818},{"time":"2026-10-11T00:51:34+00:00","agent":"a22","n":"003","status":"ok","final":925.5458914941535},{"time":"2026-10-11T00:51:56+00:00","agent":"a19","n":"003","status":"ok","final":750.125093441099},{"time":"2026-10-11T00:53:08+00:00","agent":"a01","n":"006","status":"ok","final":4809.002824505},{"time":"2026-10-11T00:53:22+00:00","agent":"a13","n":"004","status":"ok","final":1155.5933183165403},{"time":"2026-10-11T00:54:17+00:00","agent":"a05","n":"006","status":"ok","final":122614.1693703855},{"time":"2026-10-11T00:54:30+00:00","agent":"a10","n":"004","status":"ok","final":2659.551203234806},{"time":"2026-10-11T00:54:57+00:00","agent":"a17","n":"004","status":"ok","final":1780.8304012839596},{"time":"2026-10-11T00:55:03+00:00","agent":"a24","n":"005","status":"ok","final":1556.5285338791193},{"time":"2026-10-11T00:55:22+00:00","agent":"a16","n":"005","status":"ok","final":3288.6484866558294},{"time":"2026-10-11T00:56:02+00:00","agent":"a23","n":"003","status":"ok","final":2078.6594258813266},{"time":"2026-10-11T00:56:22+00:00","agent":"a11","n":"005","status":"ok","final":1320.335623389158},{"time":"2026-10-11T00:57:11+00:00","agent":"a15","n":"005","status":"ok","final":1899.7589434792048},{"time":"2026-10-11T00:58:37+00:00","agent":"a21","n":"004","status":"ok","final":1636.920008049947},{"time":"2026-10-11T00:59:42+00:00","agent":"a14","n":"005","status":"ok","final":867.7333642619166},{"time":"2026-10-11T01:00:13+00:00","agent":"a24","n":"006","status":"ok","final":1673.3750112694097},{"time":"2026-10-11T01:00:17+00:00","agent":"a19","n":"004","status":"ok","final":1424.1641735226622},{"time":"2026-10-11T01:01:34+00:00","agent":"a10","n":"005","status":"ok","final":2947.3041026168016},{"time":"2026-10-11T01:01:47+00:00","agent":"a17","n":"005","status":"ok","final":1947.8372644222486},{"time":"2026-10-11T01:01:57+00:00","agent":"a18","n":"005","status":"ok","final":8996.258180020943},{"time":"2026-10-11T01:02:04+00:00","agent":"a11","n":"006","status":"ok","final":1016.7298325991762},{"time":"2026-10-11T01:03:03+00:00","agent":"a23","n":"004","status":"ok","final":4559.298352394875},{"time":"2026-10-11T01:03:33+00:00","agent":"a08","n":"004","status":"ok","final":7964.565967381761},{"time":"2026-10-11T01:03:58+00:00","agent":"a07","n":"005","status":"ok","final":6181.0602844968},{"time":"2026-10-11T01:04:02+00:00","agent":"a13","n":"005","status":"ok","final":1275.3160590872048},{"time":"2026-10-11T01:04:27+00:00","agent":"a20","n":"004","status":"ok","final":789.7035166206674},{"time":"2026-10-11T01:04:29+00:00","agent":"a21","n":"005","status":"ok","final":1715.1663333397125},{"time":"2026-10-11T01:04:57+00:00","agent":"a04","n":"004","status":"ok","final":2438.9931961712273},{"time":"2026-10-11T01:05:03+00:00","agent":"a05","n":"007","status":"ok","final":131685.0142714035},{"time":"2026-10-11T01:05:19+00:00","agent":"a22","n":"004","status":"ok","final":1031.9928079783965},{"time":"2026-10-11T01:06:09+00:00","agent":"a14","n":"006","status":"ok","final":1302.7745216872863},{"time":"2026-10-11T01:09:58+00:00","agent":"a09","n":"004","status":"ok","final":2316.774579384113},{"time":"2026-10-11T01:14:17+00:00","agent":"a12","n":"005","status":"ok","final":5395.141023970818},{"time":"2026-10-11T01:14:35+00:00","agent":"a18","n":"006","status":"ok","final":7286.453834705852},{"time":"2026-10-11T01:14:42+00:00","agent":"a14","n":"007","status":"ok","final":1525.0390426477582},{"time":"2026-10-11T01:15:29+00:00","agent":"a13","n":"006","status":"ok","final":1439.123544620006},{"time":"2026-10-11T01:15:36+00:00","agent":"a11","n":"007","status":"ok","final":1436.7170601472337},{"time":"2026-10-11T01:15:49+00:00","agent":"a20","n":"005","status":"ok","final":1010.0137475642217},{"time":"2026-10-11T01:16:39+00:00","agent":"a21","n":"006","status":"ok","final":2199.340710400624},{"time":"2026-10-11T01:17:03+00:00","agent":"a02","n":"006","status":"ok","final":3827.05044671903},{"time":"2026-10-11T01:17:04+00:00","agent":"a08","n":"005","status":"ok","final":8616.306582805486},{"time":"2026-10-11T01:17:10+00:00","agent":"a17","n":"006","status":"ok","final":2062.741028161006},{"time":"2026-10-11T01:17:10+00:00","agent":"a25","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:17:16+00:00","agent":"a01","n":"007","status":"ok","final":8891.69684150335},{"time":"2026-10-11T01:18:48+00:00","agent":"a28","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:19:03+00:00","agent":"a26","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:19:56+00:00","agent":"a27","n":"001","status":"ok","final":187.4511072173092},{"time":"2026-10-11T01:20:02+00:00","agent":"a05","n":"008","status":"ok","final":147427.6230599583},{"time":"2026-10-11T01:20:10+00:00","agent":"a29","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:20:13+00:00","agent":"a19","n":"005","status":"ok","final":2779.9397676900244},{"time":"2026-10-11T01:21:44+00:00","agent":"a30","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:21:54+00:00","agent":"a31","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:22:03+00:00","agent":"a21","n":"007","status":"ok","final":3141.397939979284},{"time":"2026-10-11T01:22:10+00:00","agent":"a22","n":"005","status":"ok","final":1360.389934939222},{"time":"2026-10-11T01:22:29+00:00","agent":"a09","n":"005","status":"ok","final":9941.460292568889},{"time":"2026-10-11T01:22:40+00:00","agent":"a32","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:22:55+00:00","agent":"a23","n":"005","status":"ok","final":5530.300523925712},{"time":"2026-10-11T01:23:11+00:00","agent":"a25","n":"002","status":"ok","final":733.5533144671215},{"time":"2026-10-11T01:23:15+00:00","agent":"a33","n":"001","status":"ok","final":13.622760861369475},{"time":"2026-10-11T01:23:24+00:00","agent":"a34","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:24:10+00:00","agent":"a35","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:24:44+00:00","agent":"a26","n":"002","status":"ok","final":1250.0035634656278},{"time":"2026-10-11T01:25:52+00:00","agent":"a27","n":"002","status":"ok","final":934.2464972245282},{"time":"2026-10-11T01:27:06+00:00","agent":"a36","n":"001","status":"ok","final":92.93235180703257},{"time":"2026-10-11T01:27:32+00:00","agent":"a01","n":"008","status":"ok","final":10858.154793970483},{"time":"2026-10-11T01:27:54+00:00","agent":"a21","n":"008","status":"ok","final":2538.41419610661},{"time":"2026-10-11T01:28:09+00:00","agent":"a38","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:28:26+00:00","agent":"a37","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:29:07+00:00","agent":"a28","n":"002","status":"ok","final":7398.258876963639},{"time":"2026-10-11T01:29:27+00:00","agent":"a33","n":"002","status":"ok","final":515.9040233968162},{"time":"2026-10-11T01:29:59+00:00","agent":"a39","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:30:21+00:00","agent":"a31","n":"002","status":"ok","final":11790.89728142172},{"time":"2026-10-11T01:30:22+00:00","agent":"a40","n":"001","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:30:23+00:00","agent":"a29","n":"002","status":"ok","final":464.49829544480355},{"time":"2026-10-11T01:30:48+00:00","agent":"a08","n":"006","status":"ok","final":18329.329698965677},{"time":"2026-10-11T01:31:10+00:00","agent":"a02","n":"007","status":"ok","final":4075.3090544138195},{"time":"2026-10-11T01:31:19+00:00","agent":"a34","n":"002","status":"ok","final":673.4472962216441},{"time":"2026-10-11T01:31:56+00:00","agent":"a30","n":"002","status":"ok","final":516.076749383975},{"time":"2026-10-11T01:32:03+00:00","agent":"a32","n":"002","status":"ok","final":642.4061247585988},{"time":"2026-10-11T01:32:43+00:00","agent":"a25","n":"003","status":"ok","final":4337.866897451517},{"time":"2026-10-11T01:33:57+00:00","agent":"a35","n":"002","status":"ok","final":16104.055899067147},{"time":"2026-10-11T01:34:01+00:00","agent":"a36","n":"002","status":"ok","final":230.63853700985013},{"time":"2026-10-11T01:35:34+00:00","agent":"a38","n":"002","status":"ok","final":613.9840487750861},{"time":"2026-10-11T01:36:06+00:00","agent":"a28","n":"003","status":"ok","final":14296.396198676419},{"time":"2026-10-11T01:36:12+00:00","agent":"a37","n":"002","status":"ok","final":516.076749383975},{"time":"2026-10-11T01:36:41+00:00","agent":"a31","n":"003","status":"ok","final":72726.78367570757},{"time":"2026-10-11T01:36:56+00:00","agent":"a40","n":"002","status":"ok","final":571.6356999094256},{"time":"2026-10-11T01:37:06+00:00","agent":"a33","n":"003","status":"ok","final":2008.566423242205},{"time":"2026-10-11T01:37:10+00:00","agent":"a27","n":"003","status":"ok","final":8451.902224833477},{"time":"2026-10-11T01:37:54+00:00","agent":"a05","n":"009","status":"ok","final":178360.1322076304},{"time":"2026-10-11T01:38:18+00:00","agent":"a03","n":"007","status":"ok","final":2503.8346546546277},{"time":"2026-10-11T01:38:40+00:00","agent":"a32","n":"003","status":"ok","final":1030.5387267178141},{"time":"2026-10-11T01:38:40+00:00","agent":"a26","n":"003","status":"ok","final":1299.9503858611438},{"time":"2026-10-11T01:38:42+00:00","agent":"a39","n":"002","status":"ok","final":456.4999352842199},{"time":"2026-10-11T01:39:15+00:00","agent":"a36","n":"003","status":"ok","final":565.2724029293323},{"time":"2026-10-11T01:39:15+00:00","agent":"a25","n":"004","status":"ok","final":6951.4164191892205},{"time":"2026-10-11T01:40:02+00:00","agent":"a04","n":"005","status":"ok","final":3819.361575222837},{"time":"2026-10-11T01:40:20+00:00","agent":"a34","n":"003","status":"ok","final":1214.3545852912318},{"time":"2026-10-11T01:41:57+00:00","agent":"a33","n":"004","status":"ok","final":5077.490902643872},{"time":"2026-10-11T01:42:04+00:00","agent":"a07","n":"006","status":"ok","final":6649.225044396819},{"time":"2026-10-11T01:43:00+00:00","agent":"a27","n":"004","status":"ok","final":10137.54410714445},{"time":"2026-10-11T01:43:07+00:00","agent":"a06","n":"005","status":"ok","final":11408.150740669636},{"time":"2026-10-11T01:43:07+00:00","agent":"a30","n":"003","status":"ok","final":1145.7833836864738},{"time":"2026-10-11T01:43:09+00:00","agent":"a31","n":"004","status":"ok","final":109896.76421498042},{"time":"2026-10-11T01:43:45+00:00","agent":"a40","n":"003","status":"ok","final":704.9703251519761},{"time":"2026-10-11T01:44:15+00:00","agent":"a28","n":"004","status":"ok","final":17799.597774502115},{"time":"2026-10-11T01:44:44+00:00","agent":"a26","n":"004","status":"ok","final":1398.8001486733212},{"time":"2026-10-11T01:44:54+00:00","agent":"a32","n":"004","status":"ok","final":5764.658446754497},{"time":"2026-10-11T01:45:39+00:00","agent":"a29","n":"003","status":"ok","final":883.641584363283},{"time":"2026-10-11T01:46:08+00:00","agent":"a36","n":"004","status":"ok","final":806.1878043472235},{"time":"2026-10-11T01:46:37+00:00","agent":"a37","n":"003","status":"ok","final":1318.2053846079777},{"time":"2026-10-11T01:46:52+00:00","agent":"a03","n":"008","status":"ok","final":2819.7054375969396},{"time":"2026-10-11T01:47:14+00:00","agent":"a33","n":"005","status":"ok","final":7430.112339570055},{"time":"2026-10-11T01:48:35+00:00","agent":"a39","n":"003","status":"ok","final":804.0626502097246},{"time":"2026-10-11T01:48:36+00:00","agent":"a05","n":"010","status":"ok","final":187706.40020340178},{"time":"2026-10-11T01:48:55+00:00","agent":"a27","n":"005","status":"ok","final":18553.857978152977},{"time":"2026-10-11T01:48:59+00:00","agent":"a35","n":"003","status":"ok","final":121989.06887920001},{"time":"2026-10-11T01:49:19+00:00","agent":"a40","n":"004","status":"ok","final":869.528566581745},{"time":"2026-10-11T01:49:26+00:00","agent":"a38","n":"003","status":"ok","final":1030.010040495169},{"time":"2026-10-11T01:49:35+00:00","agent":"a30","n":"004","status":"ok","final":2970.110712185995},{"time":"2026-10-11T01:49:52+00:00","agent":"a28","n":"005","status":"ok","final":19738.223292040773},{"time":"2026-10-11T01:49:55+00:00","agent":"a31","n":"005","status":"ok","final":209440.50562513998},{"time":"2026-10-11T01:51:50+00:00","agent":"a29","n":"004","status":"ok","final":1075.096219716669},{"time":"2026-10-11T01:51:55+00:00","agent":"a36","n":"005","status":"ok","final":2678.8897421284973},{"time":"2026-10-11T01:52:15+00:00","agent":"a08","n":"007","status":"ok","final":54423.51211148211},{"time":"2026-10-11T01:53:17+00:00","agent":"a26","n":"005","status":"ok","final":1693.262902324796},{"time":"2026-10-11T01:54:03+00:00","agent":"a39","n":"004","status":"ok","final":952.6688199526158},{"time":"2026-10-11T01:54:10+00:00","agent":"a32","n":"005","status":"ok","final":8104.79680550818},{"time":"2026-10-11T01:54:31+00:00","agent":"a04","n":"006","status":"ok","final":4612.3816417501985},{"time":"2026-10-11T01:54:45+00:00","agent":"a35","n":"004","status":"ok","final":330495.1731681654},{"time":"2026-10-11T01:55:08+00:00","agent":"a28","n":"006","status":"ok","final":23607.47352806224},{"time":"2026-10-11T01:55:28+00:00","agent":"a06","n":"006","status":"ok","final":17262.549663896196},{"time":"2026-10-11T01:55:50+00:00","agent":"a40","n":"005","status":"ok","final":908.643159872858},{"time":"2026-10-11T01:57:14+00:00","agent":"a29","n":"005","status":"ok","final":1167.8439028587516},{"time":"2026-10-11T01:58:07+00:00","agent":"a08","n":"008","status":"ok","final":74300.28753099772},{"time":"2026-10-11T01:58:25+00:00","agent":"a37","n":"004","status":"ok","final":1653.9529213996989},{"time":"2026-10-11T01:58:40+00:00","agent":"a07","n":"007","status":"ok","final":7846.994033870101},{"time":"2026-10-11T01:58:47+00:00","agent":"a38","n":"004","status":"ok","final":1389.200825340662},{"time":"2026-10-11T01:58:55+00:00","agent":"a36","n":"006","status":"ok","final":3447.412638589928},{"time":"2026-10-11T02:01:07+00:00","agent":"a40","n":"006","status":"ok","final":973.2554414709997},{"time":"2026-10-11T02:01:44+00:00","agent":"a05","n":"011","status":"ok","final":233687.6236998251},{"time":"2026-10-11T02:02:49+00:00","agent":"a28","n":"007","status":"ok","final":25955.502705915744},{"time":"2026-10-11T02:04:23+00:00","agent":"a27","n":"006","status":"ok","final":1286.9267848923214},{"time":"2026-10-11T02:04:25+00:00","agent":"a39","n":"005","status":"ok","final":1035.2916163628959},{"time":"2026-10-11T02:05:03+00:00","agent":"a01","n":"009","status":"ok","final":16764.378497076203},{"time":"2026-10-11T02:07:53+00:00","agent":"a35","n":"005","status":"ok","final":515893.0639199248},{"time":"2026-10-11T02:08:01+00:00","agent":"a28","n":"008","status":"ok","final":29099.282045191063},{"time":"2026-10-11T02:08:53+00:00","agent":"a32","n":"006","status":"ok","final":45164.45441877173},{"time":"2026-10-11T02:09:04+00:00","agent":"a04","n":"007","status":"ok","final":5042.569726719474},{"time":"2026-10-11T02:09:29+00:00","agent":"a30","n":"005","status":"ok","final":4257.935818019393},{"time":"2026-10-11T02:11:00+00:00","agent":"a07","n":"008","status":"ok","final":8199.132056298431},{"time":"2026-10-11T02:11:43+00:00","agent":"a37","n":"005","status":"ok","final":1881.748916786542},{"time":"2026-10-11T02:12:19+00:00","agent":"a39","n":"006","status":"ok","final":1332.286103285517},{"time":"2026-10-11T02:14:27+00:00","agent":"a34","n":"004","status":"ok","final":2796.477369224968},{"time":"2026-10-11T02:15:04+00:00","agent":"a08","n":"009","status":"ok","final":80297.64812651242},{"time":"2026-10-11T02:15:37+00:00","agent":"a38","n":"005","status":"ok","final":1389.200825340662},{"time":"2026-10-11T02:16:31+00:00","agent":"a05","n":"012","status":"ok","final":236718.27715500284},{"time":"2026-10-11T02:17:02+00:00","agent":"a37","n":"006","status":"ok","final":1942.0761425079395},{"time":"2026-10-11T02:17:34+00:00","agent":"a36","n":"007","status":"ok","final":4270.961723857283},{"time":"2026-10-11T02:21:59+00:00","agent":"a08","n":"010","status":"ok","final":104398.72370252454}],"market":{"btc":[["2021-10-10",1.0],["2021-10-17",1.1257],["2021-10-24",1.1133],["2021-10-31",1.1215],["2021-11-07",1.1576],["2021-11-14",1.1987],["2021-11-21",1.0725],["2021-11-28",1.0479],["2021-12-05",0.9037],["2021-12-12",0.9157],["2021-12-19",0.854],["2021-12-26",0.929],["2022-01-02",0.8651],["2022-01-09",0.7659],["2022-01-16",0.788],["2022-01-23",0.6631],["2022-01-30",0.6931],["2022-02-06",0.7754],["2022-02-13",0.7694],["2022-02-20",0.7023],["2022-02-27",0.6897],["2022-03-06",0.7029],["2022-03-13",0.6911],["2022-03-20",0.7549],["2022-03-27",0.8567],["2022-04-03",0.849],["2022-04-10",0.7713],["2022-04-17",0.7259],["2022-04-24",0.7217],["2022-05-01",0.7038],["2022-05-08",0.6227],["2022-05-15",0.5732],["2022-05-22",0.5542],["2022-05-29",0.5391],["2022-06-05",0.5474],["2022-06-12",0.4862],["2022-06-19",0.3764],["2022-06-26",0.3849],["2022-07-03",0.3534],["2022-07-10",0.3817],["2022-07-17",0.3805],["2022-07-24",0.4131],["2022-07-31",0.4262],["2022-08-07",0.424],["2022-08-14",0.4447],["2022-08-21",0.3936],["2022-08-28",0.3578],["2022-09-04",0.3659],["2022-09-11",0.3993],["2022-09-18",0.3552],["2022-09-25",0.3441],["2022-10-02",0.3486],["2022-10-09",0.3556],["2022-10-16",0.3524],["2022-10-23",0.358],["2022-10-30",0.3774],["2022-11-06",0.3825],["2022-11-13",0.2988],["2022-11-20",0.2979],["2022-11-27",0.3006],["2022-12-04",0.313],["2022-12-11",0.3126],["2022-12-18",0.3062],["2022-12-25",0.3079],["2023-01-01",0.304],["2023-01-08",0.3134],["2023-01-15",0.3818],["2023-01-22",0.4154],["2023-01-29",0.4344],["2023-02-05",0.4196],["2023-02-12",0.3985],["2023-02-19",0.4441],["2023-02-26",0.4309],["2023-03-05",0.4104],["2023-03-12",0.4024],["2023-03-19",0.5118],["2023-03-26",0.5117],["2023-04-02",0.5154],["2023-04-09",0.5182],["2023-04-16",0.5544],["2023-04-23",0.5048],["2023-04-30",0.5348],["2023-05-07",0.5201],["2023-05-14",0.4925],["2023-05-21",0.4894],["2023-05-28",0.5135],["2023-06-04",0.4961],["2023-06-11",0.4743],["2023-06-18",0.4819],["2023-06-25",0.5573],["2023-07-02",0.5601],["2023-07-09",0.5518],["2023-07-16",0.5531],["2023-07-23",0.5504],["2023-07-30",0.5357],["2023-08-06",0.5322],["2023-08-13",0.5361],["2023-08-20",0.4792],["2023-08-27",0.4775],["2023-09-03",0.4751],["2023-09-10",0.4728],["2023-09-17",0.4853],["2023-09-24",0.4802],["2023-10-01",0.5121],["2023-10-08",0.5107],["2023-10-15",0.4968],["2023-10-22",0.5487],["2023-10-29",0.6317],["2023-11-05",0.6406],["2023-11-12",0.6781],["2023-11-19",0.6835],["2023-11-26",0.6851],["2023-12-03",0.7313],["2023-12-10",0.8011],["2023-12-17",0.757],["2023-12-24",0.7865],["2023-12-31",0.7736],["2024-01-07",0.8037],["2024-01-14",0.7635],["2024-01-21",0.7607],["2024-01-28",0.769],["2024-02-04",0.7791],["2024-02-11",0.8837],["2024-02-18",0.9539],["2024-02-25",0.9464],["2024-03-03",1.1547],["2024-03-10",1.2616],["2024-03-17",1.2513],["2024-03-24",1.2296],["2024-03-31",1.3041],["2024-04-07",1.269],["2024-04-14",1.2013],["2024-04-21",1.1881],["2024-04-28",1.1548],["2024-05-05",1.1711],["2024-05-12",1.1249],["2024-05-19",1.2125],["2024-05-26",1.2534],["2024-06-02",1.2398],["2024-06-09",1.2742],["2024-06-16",1.2199],["2024-06-23",1.1564],["2024-06-30",1.1484],["2024-07-07",1.0219],["2024-07-14",1.1123],["2024-07-21",1.2471],["2024-07-28",1.2486],["2024-08-04",1.0641],["2024-08-11",1.0742],["2024-08-18",1.0689],["2024-08-25",1.1749],["2024-09-01",1.0484],["2024-09-08",1.0039],["2024-09-15",1.0818],["2024-09-22",1.1632],["2024-09-29",1.2002],["2024-10-06",1.1493],["2024-10-13",1.1502],["2024-10-20",1.263],["2024-10-27",1.2445],["2024-11-03",1.2583],["2024-11-10",1.4704],["2024-11-17",1.6439],["2024-11-24",1.7911],["2024-12-01",1.778],["2024-12-08",1.8498],["2024-12-15",1.9112],["2024-12-22",1.7415],["2024-12-29",1.715],["2025-01-05",1.7996],["2025-01-12",1.7297],["2025-01-19",1.8539],["2025-01-26",1.8775],["2025-02-02",1.7875],["2025-02-09",1.7648],["2025-02-16",1.7585],["2025-02-23",1.7611],["2025-03-02",1.7247],["2025-03-09",1.4771],["2025-03-16",1.5107],["2025-03-23",1.5749],["2025-03-30",1.5073],["2025-04-06",1.4349],["2025-04-13",1.5324],["2025-04-20",1.5584],["2025-04-27",1.7152],["2025-05-04",1.7248],["2025-05-11",1.9049],["2025-05-18",1.9476],["2025-05-25",1.9943],["2025-06-01",1.9328],["2025-06-08",1.9344],["2025-06-15",1.9319],["2025-06-22",1.8472],["2025-06-29",1.9824],["2025-07-06",1.9979],["2025-07-13",2.1787],["2025-07-20",2.1454],["2025-07-27",2.1847],["2025-08-03",2.0895],["2025-08-10",2.1825],["2025-08-17",2.148],["2025-08-24",2.0764],["2025-08-31",1.9804],["2025-09-07",2.0333],["2025-09-14",2.1089],["2025-09-21",2.1082],["2025-09-28",2.0521],["2025-10-05",2.2591],["2025-10-12",2.1032],["2025-10-19",1.9876],["2025-10-26",2.0959],["2025-11-02",2.0224],["2025-11-09",1.9159],["2025-11-16",1.7245],["2025-11-23",1.5886],["2025-11-30",1.6532],["2025-12-07",1.6538],["2025-12-14",1.6131],["2025-12-21",1.622],["2025-12-28",1.6091],["2026-01-04",1.6746],["2026-01-11",1.6651],["2026-01-18",1.7138],["2026-01-25",1.5857],["2026-02-01",1.4082],["2026-02-08",1.2867],["2026-02-15",1.2593],["2026-02-22",1.2376],["2026-03-01",1.2034],["2026-03-08",1.207],["2026-03-15",1.3322],["2026-03-22",1.2415],["2026-03-29",1.2077],["2026-04-05",1.263],["2026-04-12",1.2942],["2026-04-19",1.3502],["2026-04-26",1.4391],["2026-05-03",1.4374],["2026-05-10",1.5041],["2026-05-17",1.4171],["2026-05-24",1.4099],["2026-05-31",1.3479],["2026-06-07",1.1587],["2026-06-14",1.2028],["2026-06-21",1.1583],["2026-06-28",1.09],["2026-07-05",1.1645],["2026-07-12",1.1669],["2026-07-19",1.1841],["2026-07-26",1.1965],["2026-08-02",1.163],["2026-08-09",1.1874],["2026-08-16",1.1508],["2026-08-23",1.4222],["2026-08-30",1.4212],["2026-09-06",1.4699],["2026-09-13",1.4058],["2026-09-20",1.4852],["2026-09-27",1.5454],["2026-10-04",1.5831],["2026-10-11",1.5118]],"alt":[["2021-10-10",1.0],["2021-10-17",1.0135],["2021-10-24",1.0584],["2021-10-31",1.1389],["2021-11-07",1.2128],["2021-11-14",1.1616],["2021-11-21",1.2185],["2021-11-28",1.345],["2021-12-05",1.0944],["2021-12-12",1.1453],["2021-12-19",1.0426],["2021-12-26",1.2241],["2022-01-02",1.1527],["2022-01-09",0.9629],["2022-01-16",1.0643],["2022-01-23",0.7649],["2022-01-30",0.7145],["2022-02-06",0.8786],["2022-02-13",0.8139],["2022-02-20",0.7552],["2022-02-27",0.7483],["2022-03-06",0.7369],["2022-03-13",0.7202],["2022-03-20",0.7881],["2022-03-27",0.8889],["2022-04-03",0.9453],["2022-04-10",0.7925],["2022-04-17",0.7293],["2022-04-24",0.7278],["2022-05-01",0.6324],["2022-05-08",0.5738],["2022-05-15",0.3818],["2022-05-22",0.3578],["2022-05-29",0.331],["2022-06-05",0.3265],["2022-06-12",0.272],["2022-06-19",0.2518],["2022-06-26",0.2791],["2022-07-03",0.2581],["2022-07-10",0.2698],["2022-07-17",0.2726],["2022-07-24",0.2931],["2022-07-31",0.3093],["2022-08-07",0.3257],["2022-08-14",0.3458],["2022-08-21",0.2952],["2022-08-28",0.2678],["2022-09-04",0.2856],["2022-09-11",0.2995],["2022-09-18",0.263],["2022-09-25",0.2732],["2022-10-02",0.265],["2022-10-09",0.2751],["2022-10-16",0.2587],["2022-10-23",0.2562],["2022-10-30",0.2953],["2022-11-06",0.2947],["2022-11-13",0.2124],["2022-11-20",0.2015],["2022-11-27",0.2197],["2022-12-04",0.2236],["2022-12-11",0.2143],["2022-12-18",0.1924],["2022-12-25",0.1881],["2023-01-01",0.1847],["2023-01-08",0.2079],["2023-01-15",0.2501],["2023-01-22",0.2605],["2023-01-29",0.2812],["2023-02-05",0.2866],["2023-02-12",0.2617],["2023-02-19",0.2847],["2023-02-26",0.269],["2023-03-05",0.2451],["2023-03-12",0.2375],["2023-03-19",0.2657],["2023-03-26",0.2555],["2023-04-02",0.2585],["2023-04-09",0.26],["2023-04-16",0.283],["2023-04-23",0.2534],["2023-04-30",0.2543],["2023-05-07",0.2425],["2023-05-14",0.2322],["2023-05-21",0.2322],["2023-05-28",0.2398],["2023-06-04",0.2399],["2023-06-11",0.1941],["2023-06-18",0.1965],["2023-06-25",0.2127],["2023-07-02",0.2149],["2023-07-09",0.2086],["2023-07-16",0.2279],["2023-07-23",0.233],["2023-07-30",0.227],["2023-08-06",0.22],["2023-08-13",0.2219],["2023-08-20",0.1983],["2023-08-27",0.1955],["2023-09-03",0.1896],["2023-09-10",0.186],["2023-09-17",0.1888],["2023-09-24",0.189],["2023-10-01",0.2029],["2023-10-08",0.196],["2023-10-15",0.1893],["2023-10-22",0.2031],["2023-10-29",0.2236],["2023-11-05",0.2398],["2023-11-12",0.2738],["2023-11-19",0.2702],["2023-11-26",0.2688],["2023-12-03",0.2717],["2023-12-10",0.3157],["2023-12-17",0.303],["2023-12-24",0.3272],["2023-12-31",0.3214],["2024-01-07",0.286],["2024-01-14",0.3009],["2024-01-21",0.2968],["2024-01-28",0.292],["2024-02-04",0.2978],["2024-02-11",0.3207],["2024-02-18",0.3428],["2024-02-25",0.3462],["2024-03-03",0.4318],["2024-03-10",0.4933],["2024-03-17",0.4805],["2024-03-24",0.4775],["2024-03-31",0.5032],["2024-04-07",0.4658],["2024-04-14",0.3872],["2024-04-21",0.4035],["2024-04-28",0.3926],["2024-05-05",0.4043],["2024-05-12",0.3845],["2024-05-19",0.397],["2024-05-26",0.4058],["2024-06-02",0.4024],["2024-06-09",0.3946],["2024-06-16",0.3659],["2024-06-23",0.3347],["2024-06-30",0.3466],["2024-07-07",0.3049],["2024-07-14",0.3392],["2024-07-21",0.3686],["2024-07-28",0.3477],["2024-08-04",0.292],["2024-08-11",0.289],["2024-08-18",0.2992],["2024-08-25",0.347],["2024-09-01",0.3055],["2024-09-08",0.3045],["2024-09-15",0.3163],["2024-09-22",0.3364],["2024-09-29",0.3657],["2024-10-06",0.3411],["2024-10-13",0.3674],["2024-10-20",0.3823],["2024-10-27",0.3291],["2024-11-03",0.2959],["2024-11-10",0.4259],["2024-11-17",0.4917],["2024-11-24",0.5327],["2024-12-01",0.6006],["2024-12-08",0.6644],["2024-12-15",0.5948],["2024-12-22",0.4609],["2024-12-29",0.4503],["2025-01-05",0.5204],["2025-01-12",0.4547],["2025-01-19",0.4544],["2025-01-26",0.433],["2025-02-02",0.3506],["2025-02-09",0.3218],["2025-02-16",0.3418],["2025-02-23",0.3374],["2025-03-02",0.3442],["2025-03-09",0.2576],["2025-03-16",0.266],["2025-03-23",0.28],["2025-03-30",0.2571],["2025-04-06",0.2251],["2025-04-13",0.2478],["2025-04-20",0.251],["2025-04-27",0.294],["2025-05-04",0.2776],["2025-05-11",0.3464],["2025-05-18",0.3425],["2025-05-25",0.3364],["2025-06-01",0.3038],["2025-06-08",0.2994],["2025-06-15",0.2803],["2025-06-22",0.2474],["2025-06-29",0.2817],["2025-07-06",0.2859],["2025-07-13",0.3343],["2025-07-20",0.394],["2025-07-27",0.3821],["2025-08-03",0.3263],["2025-08-10",0.3607],["2025-08-17",0.3498],["2025-08-24",0.3433],["2025-08-31",0.3168],["2025-09-07",0.3279],["2025-09-14",0.3632],["2025-09-21",0.3553],["2025-09-28",0.3317],["2025-10-05",0.3467],["2025-10-12",0.2986],["2025-10-19",0.2774],["2025-10-26",0.292],["2025-11-02",0.2754],["2025-11-09",0.2615],["2025-11-16",0.2311],["2025-11-23",0.2092],["2025-11-30",0.2165],["2025-12-07",0.2105],["2025-12-14",0.205],["2025-12-21",0.1952],["2025-12-28",0.1931],["2026-01-04",0.2235],["2026-01-11",0.2177],["2026-01-18",0.2096],["2026-01-25",0.1905],["2026-02-01",0.1656],["2026-02-08",0.1503],["2026-02-15",0.1547],["2026-02-22",0.1475],["2026-03-01",0.1435],["2026-03-08",0.1411],["2026-03-15",0.1558],["2026-03-22",0.1469],["2026-03-29",0.1451],["2026-04-05",0.1463],["2026-04-12",0.1448],["2026-04-19",0.1484],["2026-04-26",0.1534],["2026-05-03",0.1534],["2026-05-10",0.1704],["2026-05-17",0.1552],["2026-05-24",0.1569],["2026-05-31",0.1562],["2026-06-07",0.1373],["2026-06-14",0.1442],["2026-06-21",0.1374],["2026-06-28",0.127],["2026-07-05",0.136],["2026-07-12",0.1289],["2026-07-19",0.1273],["2026-07-26",0.1291],["2026-08-02",0.1255],["2026-08-09",0.1236],["2026-08-16",0.1211],["2026-08-23",0.1567],["2026-08-30",0.1431],["2026-09-06",0.1565],["2026-09-13",0.1468],["2026-09-20",0.1641],["2026-09-27",0.1842],["2026-10-04",0.1823],["2026-10-11",0.1688]],"breadth":[["2021-10-10",0.62],["2021-10-17",0.637],["2021-10-24",0.646],["2021-10-31",0.689],["2021-11-07",0.766],["2021-11-14",0.8],["2021-11-21",0.754],["2021-11-28",0.743],["2021-12-05",0.677],["2021-12-12",0.574],["2021-12-19",0.497],["2021-12-26",0.557],["2022-01-02",0.549],["2022-01-09",0.511],["2022-01-16",0.449],["2022-01-23",0.317],["2022-01-30",0.194],["2022-02-06",0.229],["2022-02-13",0.249],["2022-02-20",0.171],["2022-02-27",0.06],["2022-03-06",0.097],["2022-03-13",0.071],["2022-03-20",0.137],["2022-03-27",0.194],["2022-04-03",0.263],["2022-04-10",0.177],["2022-04-17",0.106],["2022-04-24",0.094],["2022-05-01",0.077],["2022-05-08",0.071],["2022-05-15",0.003],["2022-05-22",0.014],["2022-05-29",0.023],["2022-06-05",0.029],["2022-06-12",0.051],["2022-06-19",0.009],["2022-06-26",0.006],["2022-07-03",0.014],["2022-07-10",0.043],["2022-07-17",0.049],["2022-07-24",0.071],["2022-07-31",0.134],["2022-08-07",0.197],["2022-08-14",0.186],["2022-08-21",0.157],["2022-08-28",0.154],["2022-09-04",0.126],["2022-09-11",0.171],["2022-09-18",0.166],["2022-09-25",0.134],["2022-10-02",0.117],["2022-10-09",0.12],["2022-10-16",0.163],["2022-10-23",0.206],["2022-10-30",0.286],["2022-11-06",0.423],["2022-11-13",0.229],["2022-11-20",0.229],["2022-11-27",0.274],["2022-12-04",0.326],["2022-12-11",0.269],["2022-12-18",0.174],["2022-12-25",0.134],["2023-01-01",0.137],["2023-01-08",0.174],["2023-01-15",0.36],["2023-01-22",0.497],["2023-01-29",0.634],["2023-02-05",0.671],["2023-02-12",0.654],["2023-02-19",0.683],["2023-02-26",0.737],["2023-03-05",0.651],["2023-03-12",0.571],["2023-03-19",0.694],["2023-03-26",0.654],["2023-04-02",0.637],["2023-04-09",0.694],["2023-04-16",0.797],["2023-04-23",0.729],["2023-04-30",0.646],["2023-05-07",0.606],["2023-05-14",0.443],["2023-05-21",0.477],["2023-05-28",0.483],["2023-06-04",0.451],["2023-06-11",0.331],["2023-06-18",0.209],["2023-06-25",0.286],["2023-07-02",0.314],["2023-07-09",0.357],["2023-07-16",0.411],["2023-07-23",0.38],["2023-07-30",0.369],["2023-08-06",0.346],["2023-08-13",0.377],["2023-08-20",0.289],["2023-08-27",0.254],["2023-09-03",0.243],["2023-09-10",0.231],["2023-09-17",0.249],["2023-09-24",0.311],["2023-10-01",0.329],["2023-10-08",0.374],["2023-10-15",0.38],["2023-10-22",0.446],["2023-10-29",0.534],["2023-11-05",0.574],["2023-11-12",0.803],["2023-11-19",0.803],["2023-11-26",0.8],["2023-12-03",0.794],["2023-12-10",0.826],["2023-12-17",0.82],["2023-12-24",0.817],["2023-12-31",0.811],["2024-01-07",0.834],["2024-01-14",0.806],["2024-01-21",0.763],["2024-01-28",0.76],["2024-02-04",0.757],["2024-02-11",0.766],["2024-02-18",0.771],["2024-02-25",0.774],["2024-03-03",0.803],["2024-03-10",0.791],["2024-03-17",0.8],["2024-03-24",0.78],["2024-03-31",0.786],["2024-04-07",0.771],["2024-04-14",0.706],["2024-04-21",0.583],["2024-04-28",0.569],["2024-05-05",0.517],["2024-05-12",0.511],["2024-05-19",0.537],["2024-05-26",0.569],["2024-06-02",0.526],["2024-06-09",0.494],["2024-06-16",0.377],["2024-06-23",0.283],["2024-06-30",0.297],["2024-07-07",0.203],["2024-07-14",0.197],["2024-07-21",0.346],["2024-07-28",0.32],["2024-08-04",0.231],["2024-08-11",0.089],["2024-08-18",0.071],["2024-08-25",0.163],["2024-09-01",0.126],["2024-09-08",0.1],["2024-09-15",0.109],["2024-09-22",0.169],["2024-09-29",0.263],["2024-10-06",0.16],["2024-10-13",0.24],["2024-10-20",0.417],["2024-10-27",0.386],["2024-11-03",0.36],["2024-11-10",0.506],["2024-11-17",0.709],["2024-11-24",0.74],["2024-12-01",0.831],["2024-12-08",0.911],["2024-12-15",0.857],["2024-12-22",0.783],["2024-12-29",0.786],["2025-01-05",0.777],["2025-01-12",0.737],["2025-01-19",0.706],["2025-01-26",0.634],["2025-02-02",0.543],["2025-02-09",0.406],["2025-02-16",0.446],["2025-02-23",0.4],["2025-03-02",0.22],["2025-03-09",0.183],["2025-03-16",0.134],["2025-03-23",0.163],["2025-03-30",0.157],["2025-04-06",0.077],["2025-04-13",0.071],["2025-04-20",0.074],["2025-04-27",0.111],["2025-05-04",0.126],["2025-05-11",0.18],["2025-05-18",0.234],["2025-05-25",0.211],["2025-06-01",0.163],["2025-06-08",0.094],["2025-06-15",0.126],["2025-06-22",0.077],["2025-06-29",0.114],["2025-07-06",0.186],["2025-07-13",0.349],["2025-07-20",0.594],["2025-07-27",0.677],["2025-08-03",0.56],["2025-08-10",0.646],["2025-08-17",0.743],["2025-08-24",0.737],["2025-08-31",0.674],["2025-09-07",0.694],["2025-09-14",0.814],["2025-09-21",0.8],["2025-09-28",0.646],["2025-10-05",0.691],["2025-10-12",0.509],["2025-10-19",0.277],["2025-10-26",0.231],["2025-11-02",0.231],["2025-11-09",0.197],["2025-11-16",0.151],["2025-11-23",0.114],["2025-11-30",0.137],["2025-12-07",0.097],["2025-12-14",0.106],["2025-12-21",0.063],["2025-12-28",0.06],["2026-01-04",0.071],["2026-01-11",0.089],["2026-01-18",0.126],["2026-01-25",0.126],["2026-02-01",0.106],["2026-02-08",0.054],["2026-02-15",0.063],["2026-02-22",0.026],["2026-03-01",0.006],["2026-03-08",0.02],["2026-03-15",0.029],["2026-03-22",0.066],["2026-03-29",0.086],["2026-04-05",0.1],["2026-04-12",0.123],["2026-04-19",0.1],["2026-04-26",0.106],["2026-05-03",0.149],["2026-05-10",0.329],["2026-05-17",0.303],["2026-05-24",0.303],["2026-05-31",0.303],["2026-06-07",0.257],["2026-06-14",0.223],["2026-06-21",0.251],["2026-06-28",0.203],["2026-07-05",0.22],["2026-07-12",0.263],["2026-07-19",0.237],["2026-07-26",0.263],["2026-08-02",0.229],["2026-08-09",0.257],["2026-08-16",0.291],["2026-08-23",0.483],["2026-08-30",0.649],["2026-09-06",0.691],["2026-09-13",0.677],["2026-09-20",0.686],["2026-09-27",0.874],["2026-10-04",0.874],["2026-10-11",0.844]]},"meta":{"submissions":246}}