Stats

Premises:1 Underlying trend evidence may contextualize options but cannot establish their expectancy.2 Parameter choice must precede holdout.3 Charge costs and lag signals. vectorbt adjustedIEX closes2022–Sep4 2026 (IBIT startsJan2024). Six rules fast/slow10/50,20/50,20/100 crossed with calmfilter true/false. Long condition C>SMAfast>SMAslow; calmfilter additionally mean20r²<mean60r². Signals shifted1session; enter/exit at nextclose, so no sameclose lookahead. Select highest training totalreturn beforeJan1 2025, then freeze '20/100/calm=True'. HoldoutJan2025–Sep4 2026 startsflat; historical indicators warmed from training. All-in hypothetical $10000 underlying account, no options,multiplier,1%size or15min exits represented. Base fee5bp+slippage5bp per side;stress10+10bp per side. Dividends approximated through adjusted totalreturn prices, actual cash timing omitted. Training sweep and all results recorded in observations. Base {"return":-13.589548017360157,"drawdown":-23.38525707549124,"win_rate":35.714285714285715,"closed_trades":14,"open_trades":0,"avg_win_pct":2.52410959505145,"avg_loss_pct":-2.825700183252224}; doubled {"return":-15.97548309444459,"drawdown":-24.827139852440517,"win_rate":35.714285714285715,"closed_trades":14,"open_trades":0,"avg_win_pct":2.320325271584788,"avg_loss_pct":-3.021463799732673}; buyhold 54.908123791102504%. Return includes terminal open mark where present; winrate and average wins/losses use CLOSED trades only. Positive results are exploratory across seven markets, trade dependence and post-selection confidence not established; no new fund or stock mandate inferred. These are distinct state-based rules from CLI's crossover baseline. Actualoption payoffs/history N0, cannot transfer stock p/W/L to cheap-leg options or quarterKelly. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=XLE&timeframe=1Day&feed=iex&adjustment=all ; research_vol/analyze.py andbacktest.json.

Premises:1 Underlying trend evidence may contextualize options but cannot establish their expectancy.2 Parameter choice must precede holdout.3 Charge costs and lag signals. vectorbt adjustedIEX closes2022–Sep4 2026 (IBIT startsJan2024). Six rules fast/slow10/50,20/50,20/100 crossed with calmfilter true/false. Long condition C>SMAfast>SMAslow; calmfilter additionally mean20r²<mean60r². Signals shifted1session; enter/exit at nextclose, so no sameclose lookahead. Select highest training totalreturn beforeJan1 2025, then freeze '10/50/calm=True'. HoldoutJan2025–Sep4 2026 startsflat; historical indicators warmed from training. All-in hypothetical $10000 underlying account, no options,multiplier,1%size or15min exits represented. Base fee5bp+slippage5bp per side;stress10+10bp per side. Dividends approximated through adjusted totalreturn prices, actual cash timing omitted. Training sweep and all results recorded in observations. Base {"return":-5.966110831841542,"drawdown":-8.185524840189718,"win_rate":30.76923076923077,"closed_trades":13,"open_trades":0,"avg_win_pct":1.1972838355167497,"avg_loss_pct":-1.1997958809542595}; doubled {"return":-8.37948356066183,"drawdown":-9.73311178352725,"win_rate":30.76923076923077,"closed_trades":13,"open_trades":0,"avg_win_pct":0.9954887567298139,"avg_loss_pct":-1.3979971358810759}; buyhold 1.1067388096409392%. Return includes terminal open mark where present; winrate and average wins/losses use CLOSED trades only. Positive results are exploratory across seven markets, trade dependence and post-selection confidence not established; no new fund or stock mandate inferred. These are distinct state-based rules from CLI's crossover baseline. Actualoption payoffs/history N0, cannot transfer stock p/W/L to cheap-leg options or quarterKelly. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=TLT&timeframe=1Day&feed=iex&adjustment=all ; research_vol/analyze.py andbacktest.json.

Premises:1 Underlying trend evidence may contextualize options but cannot establish their expectancy.2 Parameter choice must precede holdout.3 Charge costs and lag signals. vectorbt adjustedIEX closes2022–Sep4 2026 (IBIT startsJan2024). Six rules fast/slow10/50,20/50,20/100 crossed with calmfilter true/false. Long condition C>SMAfast>SMAslow; calmfilter additionally mean20r²<mean60r². Signals shifted1session; enter/exit at nextclose, so no sameclose lookahead. Select highest training totalreturn beforeJan1 2025, then freeze '20/100/calm=False'. HoldoutJan2025–Sep4 2026 startsflat; historical indicators warmed from training. All-in hypothetical $10000 underlying account, no options,multiplier,1%size or15min exits represented. Base fee5bp+slippage5bp per side;stress10+10bp per side. Dividends approximated through adjusted totalreturn prices, actual cash timing omitted. Training sweep and all results recorded in observations. Base {"return":5.550505053277229,"drawdown":-9.346161060235026,"win_rate":41.66666666666667,"closed_trades":24,"open_trades":1,"avg_win_pct":1.8198183228977298,"avg_loss_pct":-0.8748826533533464}; doubled {"return":0.5032716466316742,"drawdown":-12.25225586861165,"win_rate":41.66666666666667,"closed_trades":24,"open_trades":1,"avg_win_pct":1.6170899087364539,"avg_loss_pct":-1.0735710347206369}; buyhold 34.00260983036101%. Return includes terminal open mark where present; winrate and average wins/losses use CLOSED trades only. Positive results are exploratory across seven markets, trade dependence and post-selection confidence not established; no new fund or stock mandate inferred. These are distinct state-based rules from CLI's crossover baseline. Actualoption payoffs/history N0, cannot transfer stock p/W/L to cheap-leg options or quarterKelly. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=SPY&timeframe=1Day&feed=iex&adjustment=all ; research_vol/analyze.py andbacktest.json.

Premises:1 Underlying trend evidence may contextualize options but cannot establish their expectancy.2 Parameter choice must precede holdout.3 Charge costs and lag signals. vectorbt adjustedIEX closes2022–Sep4 2026 (IBIT startsJan2024). Six rules fast/slow10/50,20/50,20/100 crossed with calmfilter true/false. Long condition C>SMAfast>SMAslow; calmfilter additionally mean20r²<mean60r². Signals shifted1session; enter/exit at nextclose, so no sameclose lookahead. Select highest training totalreturn beforeJan1 2025, then freeze '10/50/calm=True'. HoldoutJan2025–Sep4 2026 startsflat; historical indicators warmed from training. All-in hypothetical $10000 underlying account, no options,multiplier,1%size or15min exits represented. Base fee5bp+slippage5bp per side;stress10+10bp per side. Dividends approximated through adjusted totalreturn prices, actual cash timing omitted. Training sweep and all results recorded in observations. Base {"return":-0.014107329798189312,"drawdown":-6.84281610523424,"win_rate":50,"closed_trades":20,"open_trades":0,"avg_win_pct":1.4236745104623536,"avg_loss_pct":-1.3956995267389634}; doubled {"return":-3.934612499779578,"drawdown":-7.960620660809836,"win_rate":45,"closed_trades":20,"open_trades":0,"avg_win_pct":1.366715748509182,"avg_loss_pct":-1.456464755776127}; buyhold 35.87695133149678%. Return includes terminal open mark where present; winrate and average wins/losses use CLOSED trades only. Positive results are exploratory across seven markets, trade dependence and post-selection confidence not established; no new fund or stock mandate inferred. These are distinct state-based rules from CLI's crossover baseline. Actualoption payoffs/history N0, cannot transfer stock p/W/L to cheap-leg options or quarterKelly. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=IWM&timeframe=1Day&feed=iex&adjustment=all ; research_vol/analyze.py andbacktest.json.

Premises:1 Underlying trend evidence may contextualize options but cannot establish their expectancy.2 Parameter choice must precede holdout.3 Charge costs and lag signals. vectorbt adjustedIEX closes2022–Sep4 2026 (IBIT startsJan2024). Six rules fast/slow10/50,20/50,20/100 crossed with calmfilter true/false. Long condition C>SMAfast>SMAslow; calmfilter additionally mean20r²<mean60r². Signals shifted1session; enter/exit at nextclose, so no sameclose lookahead. Select highest training totalreturn beforeJan1 2025, then freeze '20/100/calm=False'. HoldoutJan2025–Sep4 2026 startsflat; historical indicators warmed from training. All-in hypothetical $10000 underlying account, no options,multiplier,1%size or15min exits represented. Base fee5bp+slippage5bp per side;stress10+10bp per side. Dividends approximated through adjusted totalreturn prices, actual cash timing omitted. Training sweep and all results recorded in observations. Base {"return":-9.745647759126095,"drawdown":-15.152058406624224,"win_rate":35.714285714285715,"closed_trades":14,"open_trades":1,"avg_win_pct":1.9692888103112876,"avg_loss_pct":-2.532015586754256}; doubled {"return":-12.325371970621582,"drawdown":-17.120894011513276,"win_rate":28.57142857142857,"closed_trades":14,"open_trades":1,"avg_win_pct":2.2126865088046297,"avg_loss_pct":-2.457304011436743}; buyhold -18.32385080827238%. Return includes terminal open mark where present; winrate and average wins/losses use CLOSED trades only. Positive results are exploratory across seven markets, trade dependence and post-selection confidence not established; no new fund or stock mandate inferred. These are distinct state-based rules from CLI's crossover baseline. Actualoption payoffs/history N0, cannot transfer stock p/W/L to cheap-leg options or quarterKelly. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=IBIT&timeframe=1Day&feed=iex&adjustment=all ; research_vol/analyze.py andbacktest.json.

Ran required CLI: node /app/agent/dist/backtest.js --symbols UAMY --strategy breakout --lookback20 --days1460 --stop-pct5 --target-pct10 (actual command uses spaces between flag and20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Output1002bars,68trades,+65.4721%,34.1623%DD,39.7059%wins,mean0.9560%,buyhold1185.9267%. This all-window diagnostic has different signal exits and fixed percentage stops; it is not the subsequent preregistered ADR/five-session long-only rule, has no independently held-out execution model in this output, and cannot justify entry. The costed vectorbt train/holdout test543 supplies the decision and rejects. Do not confuse attractive gross CLI performance with validated net stock edge.

Preregistered note552 before customtest; UAMY selected after today'sfilings, so historicalholdout notpristineprospective. Datahttps://runway.adityaperswal.workers.dev/data/stock-bars, SIPadjusted daily2022-Jan3–Sep4 2026,1173bars, no pagination. Fixedclose>prior20closehigh ANDSMA50;nextopen; frozenprior20ADR,1ADRstop2ADRtarget,fifthsessionclose;stopfirstonambiguousdailybar,stopgapworseopen;flatatsplit/end,nooverlap. Fees5bp+slippage10bp eachside;stress20bp. vectorbt longonly withtwomarks/day; all35holdouttradeslong,35exits,compoundedentry-fee-inclusivePNL matchesportfolio1e-8. DailyOHLCnot15mintrigger/intradayliquidityvalidation; DD sampled twice/day mayunderstateintradayDD. Train2022–24:39trades,-41.6594%return,57.9309%DD,30.7692%wins. Held2025–Sep4 2026:35trades,-7.11606%return,58.2452%DD,40%wins,avgwin12.96375%,avgloss8.05117%,mean0.35480%;IIDbootstrap95%mean[-3.18592,4.12585]%;4-tradecircularblock[-3.98671,4.79083]%. Stress return-13.39560%,58.8669%DD,37.1429%wins,avgwin13.72262%/loss7.86336%. Primary netquarterKellypoint0.6842%capital,stress0.2811%; uncertainty-adjustedapproved0 becausemeanintervalcrosseszero,trainingfails,drawdownlargeandcurrententryfalse. Neighbors10/40 sessions testedasrobustnessonly:train-54.2387/-38.8928%;holdout+17.8054/+36.7124%,N41/25, bothmeanintervalscrosszero; notswitching tobestholdout. This rejectsfixedstockrule, notprovesnegativefundamentals. Numericaledgecannotbeimportedfromtwooperator-directed tankerwins. Internaldefaultboth-directionversiondiscarded; laterrawtrade-return denominatorcorrectedbeforeanalysis. Earlierobservationaverages supersededbycorrectedobservations; portfoliooutcomesunchanged. No trade orpaperprofitclaim.

Required backtestCLI run mean-reversion20,z2,365days;366bars includes livepartialSep7day, nofees/slippage/holdout and unusedexitflag. Figures are gross diagnostic only; do not trust forproposal. Fixed state implementation verified from /app/agent/dist/backtest/strategies.js; completed-only daily/hourly nextopen costedchronologicaltests518-523supersede decision use. Premise1 gross historical performance insufficient;premise2 netholdoutfails;conclusion allocation0 even wheregrossSOLpositive. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .

Required backtestCLI run mean-reversion20,z2,365days;366bars includes livepartialSep7day, nofees/slippage/holdout and unusedexitflag. Figures are gross diagnostic only; do not trust forproposal. Fixed state implementation verified from /app/agent/dist/backtest/strategies.js; completed-only daily/hourly nextopen costedchronologicaltests518-523supersede decision use. Premise1 gross historical performance insufficient;premise2 netholdoutfails;conclusion allocation0 even wheregrossSOLpositive. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .

Required backtestCLI run mean-reversion20,z2,365days;366bars includes livepartialSep7day, nofees/slippage/holdout and unusedexitflag. Figures are gross diagnostic only; do not trust forproposal. Fixed state implementation verified from /app/agent/dist/backtest/strategies.js; completed-only daily/hourly nextopen costedchronologicaltests518-523supersede decision use. Premise1 gross historical performance insufficient;premise2 netholdoutfails;conclusion allocation0 even wheregrossSOLpositive. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Hour results: train-87.3132%, holdout-75.1690%, DD75.2697%, p19.745%, W0.5263%, L1.2174%, N157; doubledcost-91.7263%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 0/9 holdoutspositive,range-91.9407..-43.1858%; no winner selected. Datafullsamplemissing7; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Day results: train17.5720%, holdout-12.3485%, DD22.7440%, p40.000%, W3.0222%, L6.1113%, N5; doubledcost-15.3633%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 0/9 holdoutspositive,range-37.2416..-2.4202%; no winner selected. Datafullsamplemissing0; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Hour results: train-88.1564%, holdout-72.8574%, DD72.9434%, p14.371%, W0.4742%, L0.9828%, N167; doubledcost-91.5675%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 0/9 holdoutspositive,range-89.0855..-48.1890%; no winner selected. Datafullsamplemissing2; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Day results: train3.2049%, holdout-7.5962%, DD20.2821%, p40.000%, W6.5701%, L6.5327%, N5; doubledcost-10.7744%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 1/9 holdoutspositive,range-38.1617..9.6783%; no winner selected. Datafullsamplemissing0; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Hour results: train-85.9068%, holdout-74.3840%, DD74.4637%, p9.884%, W0.4058%, L0.9172%, N172; doubledcost-92.3155%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 0/9 holdoutspositive,range-90.5265..-50.4326%; no winner selected. Datafullsamplemissing2; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Premise1 frozen note551:20bar close zscore<=-2, population SD, long while condition true; lag state onebar, fill next available open, exit when condition false. This matches CLI state semantics, whose exit flag is unused; not hold-until-mean variant. Vectorbt capital10000,fees.25%+slip.10%eachside, stress.50%+.20%; all-in unlevered per-column diagnostics, not fund-size proposal. Training2025; holdoutJan1-Sep5 2026 cash-start each window with finalclose liquidation. 1Day results: train3.8751%, holdout-15.4031%, DD18.2248%, p20.000%, W0.9024%, L4.2040%, N5; doubledcost-18.3128%. Premise2 3x3 lookback15/20/30,z1.5/2/2.5 sensitivity, 2/9 holdoutspositive,range-28.3073..2.9004%; no winner selected. Datafullsamplemissing0; hourly reindex only mark prices forward, no fills on absentbars, require complete lookback for entry; exits delayed to actualbar. All raw moments recorded observations7832-43. Premise3 no incident-conditioned cohort: this is generic bounce falsification, not Liquid-event backtest. Historicalholdout previously inspected for other rules, not pristine discovery; tiny dailyN and hourlycost churn preclude edge. Reject allocation0, clippedquarterKelly0. Counterexample some dailyneighbor profits do not validate frozenbaseline. No 15min stopfill model, stop/target orders or live trading implied. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Frozen diagnostic variants, no winner selection. Main later N29 return-1.1431% at10bp/side and-6.7137% at20bp. Next available bar-open fill after signal worsens later return to-5.6795%/-10.9945%; regular-hours-only closes yield-6.2437%/-11.5269%. Narrow0.75ATR variant admits42later cohorts, +2.84395% at10bp but-5.44207% at20bp and training-7.24753%/-11.24016%; it changes admission as well as exits, so cannot isolate a pure stop-distance effect. Wider1.25ATR admits0cohorts because target exceeds2ADR, no return estimate. All28 variant/split/cost rows recorded observations7787–7814; source https://runway.adityaperswal.workers.dev/data/stock-bars . Premises: parameters frozen before outcomes, but underlying window and winner already inspected; robustness is diagnostic, not fresh selection evidence. Conclusion reject live swing adoption. Strongest counterargument is narrow low-cost positive result, defeated for adoption by cost sensitivity, negative training and post-selection uncertainty. No claim all possible exits fail. Main sample10same-day exits also means future use must coordinate the account-wide3-in5day limit; no orders now.

Frozen notes547/549. DBC previous completed month12m adjusted return exceeds BIL; first session regular open entry, no catch-up or reentry within month. Prior20session simple ATR sets initial stop1ATR, target1.5ATR and fixed-percent1ATR/entry trailing distance; sample raw15Min closes at interval end, ratchet using sampled highs of closes, exit at observed price after costs; skip target>2ADR20. Timeout next monthly first open. Training2017–2020, later2021–August2026; later window already inspected and DBC selected after prior strong results, so this is exploratory chronological testing, not an independent holdout. Fee/slippage equally split within 20bp each side. Raw-price triggers, adjusted total-return reinvestment valuation proxy; actual dividends and executable quotes unavailable. Vectorbt explicitly long-only replay matches independently compounded cash to floating-point tolerance, no negative cash. Results: {"label":"trail_main","split":"later","mult":1,"cost_bp":20,"session":"all","lag":false,"gate":true,"return_pct":-6.713681659033122,"drawdown_pct":12.656278526234054,"win_rate_pct":37.93103448275862,"sample":29,"avg_win_pct":2.386409567822121,"avg_loss_pct":1.806399784928038,"mean_pct":-0.21602382354004657,"quarter_kelly":0,"median_hold_days":1.03125,"target_exits":11,"stop_exits":18,"timeout_exits":0,"extended_exits":13,"replay_difference":-6.661338147750939e-16,"min_cash":0}. Drawdown marked at available15Min closes including extended hours; original monthly later29.7078% here versus saved27.5443%daily, different resolution rather than return disagreement. Fixed monthly total +104.6964% and training -11.6024% reproduce saved study. Premises: same entry population and costs permit descriptive exit comparison; executable fills and fresh independent evidence remain absent. Conclusion reject this adaptation for trading; no positive robust edge and actual size0. Counterexample: some months profit; this does not validate the average. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; economic hypothesis https://pages.stern.nyu.edu/~lpederse/papers/TimeSeriesMomentum.pdf .

Frozen notes547/549. DBC previous completed month12m adjusted return exceeds BIL; first session regular open entry, no catch-up or reentry within month. Prior20session simple ATR sets initial stop1ATR, target1.5ATR and fixed-percent1ATR/entry trailing distance; sample raw15Min closes at interval end, ratchet using sampled highs of closes, exit at observed price after costs; skip target>2ADR20. Timeout next monthly first open. Training2017–2020, later2021–August2026; later window already inspected and DBC selected after prior strong results, so this is exploratory chronological testing, not an independent holdout. Fee/slippage equally split within 10bp each side. Raw-price triggers, adjusted total-return reinvestment valuation proxy; actual dividends and executable quotes unavailable. Vectorbt explicitly long-only replay matches independently compounded cash to floating-point tolerance, no negative cash. Results: {"label":"trail_main","split":"later","mult":1,"cost_bp":10,"session":"all","lag":false,"gate":true,"return_pct":-1.1430862148126497,"drawdown_pct":10.075759941838047,"win_rate_pct":37.93103448275862,"sample":29,"avg_win_pct":2.5913874160504338,"avg_loss_pct":1.6098159515185475,"mean_pct":-0.016256053475140725,"quarter_kelly":0,"median_hold_days":1.03125,"target_exits":11,"stop_exits":18,"timeout_exits":0,"extended_exits":13,"replay_difference":-9.992007221626409e-16,"min_cash":0}. Drawdown marked at available15Min closes including extended hours; original monthly later29.7078% here versus saved27.5443%daily, different resolution rather than return disagreement. Fixed monthly total +104.6964% and training -11.6024% reproduce saved study. Premises: same entry population and costs permit descriptive exit comparison; executable fills and fresh independent evidence remain absent. Conclusion reject this adaptation for trading; no positive robust edge and actual size0. Counterexample: some months profit; this does not validate the average. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; economic hypothesis https://pages.stern.nyu.edu/~lpederse/papers/TimeSeriesMomentum.pdf .

Frozen notes547/549. DBC previous completed month12m adjusted return exceeds BIL; first session regular open entry, no catch-up or reentry within month. Prior20session simple ATR sets initial stop1ATR, target1.5ATR and fixed-percent1ATR/entry trailing distance; sample raw15Min closes at interval end, ratchet using sampled highs of closes, exit at observed price after costs; skip target>2ADR20. Timeout next monthly first open. Training2017–2020, later2021–August2026; later window already inspected and DBC selected after prior strong results, so this is exploratory chronological testing, not an independent holdout. Fee/slippage equally split within 10bp each side. Raw-price triggers, adjusted total-return reinvestment valuation proxy; actual dividends and executable quotes unavailable. Vectorbt explicitly long-only replay matches independently compounded cash to floating-point tolerance, no negative cash. Results: {"label":"trail_main","split":"train","mult":1,"cost_bp":10,"session":"all","lag":false,"gate":true,"return_pct":-1.8176291636592623,"drawdown_pct":7.049404188092845,"win_rate_pct":45.45454545454545,"sample":22,"avg_win_pct":1.237877806449379,"avg_loss_pct":1.168113067267603,"mean_pct":-0.07448085194170204,"quarter_kelly":0,"median_hold_days":1.2552083333333335,"target_exits":8,"stop_exits":14,"timeout_exits":0,"extended_exits":5,"replay_difference":-8.881784197001252e-16,"min_cash":0}. Drawdown marked at available15Min closes including extended hours; original monthly later29.7078% here versus saved27.5443%daily, different resolution rather than return disagreement. Fixed monthly total +104.6964% and training -11.6024% reproduce saved study. Premises: same entry population and costs permit descriptive exit comparison; executable fills and fresh independent evidence remain absent. Conclusion reject this adaptation for trading; no positive robust edge and actual size0. Counterexample: some months profit; this does not validate the average. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; economic hypothesis https://pages.stern.nyu.edu/~lpederse/papers/TimeSeriesMomentum.pdf .

Frozen notes547/549. DBC previous completed month12m adjusted return exceeds BIL; first session regular open entry, no catch-up or reentry within month. Full-month hold on only the same entries that satisfy the swing target<=2ADR20 gate. Training2017–2020, later2021–August2026; later window already inspected and DBC selected after prior strong results, so this is exploratory chronological testing, not an independent holdout. Fee/slippage equally split within 10bp each side. Raw-price triggers, adjusted total-return reinvestment valuation proxy; actual dividends and executable quotes unavailable. Vectorbt explicitly long-only replay matches independently compounded cash to floating-point tolerance, no negative cash. Results: {"label":"monthly_matched","split":"later","mult":1,"cost_bp":10,"session":"all","lag":false,"gate":true,"return_pct":37.36428440482524,"drawdown_pct":26.470485269856113,"win_rate_pct":51.724137931034484,"sample":29,"avg_win_pct":4.907200275789534,"avg_loss_pct":2.7464593861411704,"mean_pct":1.2123300941678148,"quarter_kelly":0.061762819226527266,"median_hold_days":30,"target_exits":0,"stop_exits":0,"timeout_exits":29,"extended_exits":0,"replay_difference":-3.774758283725532e-15,"min_cash":0}. Drawdown marked at available15Min closes including extended hours; original monthly later29.7078% here versus saved27.5443%daily, different resolution rather than return disagreement. Fixed monthly total +104.6964% and training -11.6024% reproduce saved study. Premises: same entry population and costs permit descriptive exit comparison; executable fills and fresh independent evidence remain absent. Conclusion retain research WATCH only; monthly profitability does not license swing exits. Counterexample: some months profit; this does not validate the average. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; economic hypothesis https://pages.stern.nyu.edu/~lpederse/papers/TimeSeriesMomentum.pdf .

Frozen notes547/549. DBC previous completed month12m adjusted return exceeds BIL; first session regular open entry, no catch-up or reentry within month. Original full-month hold, no ADR gate. Training2017–2020, later2021–August2026; later window already inspected and DBC selected after prior strong results, so this is exploratory chronological testing, not an independent holdout. Fee/slippage equally split within 10bp each side. Raw-price triggers, adjusted total-return reinvestment valuation proxy; actual dividends and executable quotes unavailable. Vectorbt explicitly long-only replay matches independently compounded cash to floating-point tolerance, no negative cash. Results: {"label":"monthly_ungated","split":"later","mult":1,"cost_bp":10,"session":"all","lag":false,"gate":false,"return_pct":104.69636283313055,"drawdown_pct":29.707788585251205,"win_rate_pct":59.523809523809526,"sample":42,"avg_win_pct":5.446640655425865,"avg_loss_pct":3.405966881790677,"mean_pct":1.8634423665525028,"quarter_kelly":0.08553172884171127,"median_hold_days":30,"target_exits":0,"stop_exits":0,"timeout_exits":42,"extended_exits":0,"replay_difference":-8.881784197001252e-15,"min_cash":0}. Drawdown marked at available15Min closes including extended hours; original monthly later29.7078% here versus saved27.5443%daily, different resolution rather than return disagreement. Fixed monthly total +104.6964% and training -11.6024% reproduce saved study. Premises: same entry population and costs permit descriptive exit comparison; executable fills and fresh independent evidence remain absent. Conclusion retain research WATCH only; monthly profitability does not license swing exits. Counterexample: some months profit; this does not validate the average. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; economic hypothesis https://pages.stern.nyu.edu/~lpederse/papers/TimeSeriesMomentum.pdf .

Preregistered20-sessionmomentum;10/40sensitivityreportedwithoutselectingholdoutwinner. Vectorbt laggedclose signal,nextpositive-volumeopen,persistentdesiredposition,flatstart eachsplit,lastopenforcedexit;5bpfee+10bpslippageeachside,25bpstress. TrainNov2024–Dec2025:20day-32.6185%,DD35.5869%,N10,p20%,W6.3481%,L6.0441%; rawquarterKelly-.140422,clipped0. HoldoutJan–Sep42026:+11.7464%,DD22.4831%,N7,p28.5714%,W22.1586%,L4.9969%;stress+9.58875%. Holdout-onlydescriptivequarterKelly3.11595%isNOTadopted: priortrainfails and thisstrategyholdsacrossevents,unlikeCatalystmandate.10/40daytrain-52.0329/-40.3158%;holdout+32.9979/+27.4141%,N10/4. Allsweepvaluesrecordedobservations;unselectedpositiveholdoutsacknowledged. No lossboundfromdailyexits;IEXfirstprintsnotauction/NBBO,zero-volumepricesremainindicatorstalenessrisk. Separateeventaudit500trainN0/holdoutN1; no evidencegenericmomentumtransfers tothree-sessionprePDUFAwindow. Rejectrisk,size0.

Preregisterednote548 beforeeventreturns: enteropen fifthregularsession strictlybeforePDUFA,exitclose secondsessionbefore; chronological first2training,last1holdout. Dates issuer https://ir.telixpharma.com/static-files/9959d5f3-16cc-49fc-89eb-4a4af5b57049 and https://telixpharma.com/news-views/telix-reports-204m-revenue-up-63-yoy/ . IEX452barscheckedagainstSPYsessiondates;vectorbt5bpfees+10bpslippageeachside,25bpstress. GozellixMar17–20 entry17.455/exit17.69 givespaper1.042736%,butexitvolume0 INVALID. PixclaraApr21–24 entry16.75/exit19.38 paper15.354908%,butentryvolume0 INVALID. Do notcounttheirwins or shiftfilldates. ZircaixAug20–25 entry10.96/exit12.18 validpositiveprints2275/2640; holdout+10.798492%,DD2.706111%,win100%,N1;stress+10.466594%,DD2.852198%. Thus eligibletrainN0/holdoutN1; W10.7985%,noobservedloss,calibratedp/averageL/Kelly undefined,size0. Selectedsameissuer/nonindependentproducts,earningsoverlap,incompletefirstnewsarchive and sparseIEX preventtransport. KnownGozellixearlyapproval also preventsclaimfixeddeadlineexitguaranteesnonews. CurrentT-5Sep3alreadyelapsed; donotshrinkruletofitSep8. Positive loneholdout reported transparently, notproof noedgeexists. Artifactresearch/analyze_final.py/final_results.json; observations preservefigures.

Ran node /app/agent/dist/backtest.js --symbols TLX --strategy momentum --lookback20 --days650. 444bars,18trades,return-4.061632%,DD46.722770%,win33.3333%,buyhold-22.753717%. CLI lacks the dated-event filter and this result is fullwindow/no explicit executioncosts or holdout; never use for allocation. It triggers separate vectorbt costed chronological audit recordednext. RawIEX zero-volume bars require gating; firstprint is notopeningauction. No proposedtrade.