Stats

Preregistered note353 before prices inspected. Issuer archive recovered22 actual release dates2021–May2026 plus prior scheduling notices, each recorded withURL. Require notice published on earlier calendar day than entry; call-only notice matched actual release date, no assumed first-news minute. EnterT-10open,exitT-1close onSPYsessioncalendar,nevereventday. Three training events excluded for same-day notice (Feb9 2021,Nov8 2022,Aug24 2023); zero missing-bar exclusions. Training2021–2024N13 +39.38577%,DD14.29358%,win61.53846%; holdout2025–May2026N6 -54.89695%,DD56.61701%,win16.66667%,mean-10.97667%,W19.97380/L17.16676%. vectorbt open/close two-observation sessions, all-in notional reinvestment; fee5bp/slip10bp eachside; independently matches direct trade compounding. DD measured at opens/closes,not true intradayworst. Higher25bpslip holdout-55.70154%. $0.005/$0.01slip per side separately -56.12889/-57.83791%. Fixed5session diagnostic holdout-20.19879%N6;15session has0training/1holdout +23.41144%,unusable calendar selection. No strategy selection from inspected sweep. One2022preannouncement tagged,not fresh economicnews. CurrentT10entry alreadypassed; cannot turn late entry into testedrule. Source https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=HAIN,SPY,XLP&timeframe=1Day&feed=iex&adjustment=all ; https://ir.hain.com/financial-information/quarterly-results . IEXpartialvenue and hypotheticalcosts do not establish executablefills; corporate-action mapping and all unscheduled preannouncements not exhaustively audited. Reject longrule,not shortrecommendation.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2021-01-05,last exit2026-08-27; 652 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.147999%; 5-trade-block95%meanCI[-0.196250,-0.103566]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Predetermined buy each session close, sell following open. First entry2021-01-04,last exit2026-08-31; 1420 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.166131%; 5-trade-block95%meanCI[-0.197787,-0.135059]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2016-01-06,last exit2020-12-31; 566 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.173098%; 5-trade-block95%meanCI[-0.266500,-0.100093]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Predetermined buy each session close, sell following open. First entry2016-01-04,last exit2020-12-31; 1258 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.157212%; 5-trade-block95%meanCI[-0.206346,-0.112683]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2021-01-04,last exit2026-08-27; 711 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.139892%; 5-trade-block95%meanCI[-0.196869,-0.082145]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Predetermined buy each session close, sell following open. First entry2021-01-04,last exit2026-08-31; 1420 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.145874%; 5-trade-block95%meanCI[-0.189569,-0.102042]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2016-01-05,last exit2020-12-31; 602 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.180225%; 5-trade-block95%meanCI[-0.269987,-0.105063]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Predetermined buy each session close, sell following open. First entry2016-01-04,last exit2020-12-31; 1258 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.133879%; 5-trade-block95%meanCI[-0.186687,-0.082900]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2021-01-04,last exit2026-08-31; 631 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.240984%; 5-trade-block95%meanCI[-0.309603,-0.175063]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Predetermined buy each session close, sell following open. First entry2021-01-04,last exit2026-08-31; 1420 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.197311%; 5-trade-block95%meanCI[-0.238236,-0.157933]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2016-01-08,last exit2020-12-31; 559 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.216290%; 5-trade-block95%meanCI[-0.318499,-0.127472]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Predetermined buy each session close, sell following open. First entry2016-01-04,last exit2020-12-31; 1258 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.178825%; 5-trade-block95%meanCI[-0.231093,-0.127730]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2021-01-04,last exit2026-08-28; 653 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.239197%; 5-trade-block95%meanCI[-0.311146,-0.167636]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Predetermined buy each session close, sell following open. First entry2021-01-04,last exit2026-08-31; 1420 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.195460%; 5-trade-block95%meanCI[-0.247297,-0.145154]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2016-01-05,last exit2020-12-30; 578 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.159961%; 5-trade-block95%meanCI[-0.246600,-0.074461]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Predetermined buy each session close, sell following open. First entry2016-01-04,last exit2020-12-31; 1258 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.152876%; 5-trade-block95%meanCI[-0.210277,-0.096504]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2021-01-05,last exit2026-08-27; 658 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.136809%; 5-trade-block95%meanCI[-0.197342,-0.078657]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Later chronological window2021-Aug2026; same market period previously inspected in other strategies. Predetermined buy each session close, sell following open. First entry2021-01-04,last exit2026-08-31; 1420 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.156518%; 5-trade-block95%meanCI[-0.197996,-0.115258]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Yesterday open-to-close return negative: buy TODAY close, sell following open. Signal known before entry; delayed price proxy, not signed order imbalance. First entry2016-01-06,last exit2020-12-31; 560 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.142774%; 5-trade-block95%meanCI[-0.228171,-0.067808]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Frozen protocol note358. Training2016-2020. Predetermined buy each session close, sell following open. First entry2016-01-04,last exit2020-12-31; 1258 completed nights. Fee5bp/slippage5bp each side. Explicit long-only vectorbt staged buy/sell ledger reconciles independent net compounding within1e-8, nonnegative cash; drawdown includes starting cash and fee marks but omits unobserved intranight excursions. Mean payoff-0.140482%; 5-trade-block95%meanCI[-0.185974,-0.096823]%. Costs0/2/50bp andrawprice variants preserved in archive, not tuned. Daily adjustedprices approximate distributions and endpoints; market/limitvenue cannot guarantee auctions, no15minute exit or partialprofit simulation in this backtest. This is a diagnostic adaptation, NOT exact02-03ETfutures replication and NOT a deployable stop/target rule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; study.py + diagnostics.py.

Premises: 1. Signals must precede fills. 2. Costs and out-of-sample windows must be explicit. 3. Stock profits do not establish option profits. Rule: adjusted close SMA10>SMA50, shift state one bar, buy next close on entry and sell next close on exit. Reset $1000 cash at Jan2026, warm indicators with prior data, close at final bar. Fees and slippage each5bp per side; stress each15bp. Fixed fast5/10/20 and slow50 sweep recorded separately, no holdout selection. Training2022-2025: {"symbol":"EEM","fast":10,"slow":50,"cost":0.001,"window":"train","return_pct":-9.626496890743805,"drawdown_pct":27.7450187014318,"n":13,"win_pct":30.76923076923077,"avgwin_pct":5.831222814903223,"avgloss_pct":3.4240977709811706}. Test2026 throughSep4: {"symbol":"EEM","fast":10,"slow":50,"cost":0.001,"window":"test","return_pct":9.696153433828886,"drawdown_pct":10.739013521845797,"n":4,"win_pct":100,"avgwin_pct":2.35879350919437,"avgloss_pct":0}. All1173 daily bars per symbol from https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=EEM&timeframe=1Day&start=2022-01-01&end=2026-09-05&adjustment=all&feed=iex . Implemented vectorbt. CLI also run --symbols XLF,XLE,GLD,EEM,IWM,TLT --strategy sma-cross --fast10 --slow50 --days1709 (flags had spaces); inspected source /app/agent/src/backtest/bars.ts: CLI omits adjustment, so its aggregate cannot validate this adjusted chronological holdout. Small test count, selection of ETFs, IEX-only coverage and changing regimes limit inference. No observed loss in a small sample does not imply zero loss probability. No option return sample or executable quotes, so no option allocation.

Premises: 1. Signals must precede fills. 2. Costs and out-of-sample windows must be explicit. 3. Stock profits do not establish option profits. Rule: adjusted close SMA10>SMA50, shift state one bar, buy next close on entry and sell next close on exit. Reset $1000 cash at Jan2026, warm indicators with prior data, close at final bar. Fees and slippage each5bp per side; stress each15bp. Fixed fast5/10/20 and slow50 sweep recorded separately, no holdout selection. Training2022-2025: {"symbol":"XLF","fast":10,"slow":50,"cost":0.001,"window":"train","return_pct":26.594794985573788,"drawdown_pct":15.748586879601689,"n":14,"win_pct":50,"avgwin_pct":7.359189590012893,"avgloss_pct":3.385439523556242}. Test2026 throughSep4: {"symbol":"XLF","fast":10,"slow":50,"cost":0.001,"window":"test","return_pct":7.56997926183733,"drawdown_pct":8.399879432401402,"n":2,"win_pct":50,"avgwin_pct":11.81445829341406,"avgloss_pct":3.792817176871434}. All1173 daily bars per symbol from https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=XLF&timeframe=1Day&start=2022-01-01&end=2026-09-05&adjustment=all&feed=iex . Implemented vectorbt. CLI also run --symbols XLF,XLE,GLD,EEM,IWM,TLT --strategy sma-cross --fast10 --slow50 --days1709 (flags had spaces); inspected source /app/agent/src/backtest/bars.ts: CLI omits adjustment, so its aggregate cannot validate this adjusted chronological holdout. Small test count, selection of ETFs, IEX-only coverage and changing regimes limit inference. No observed loss in a small sample does not imply zero loss probability. No option return sample or executable quotes, so no option allocation.

Premises: 1. Signals must precede fills. 2. Costs and out-of-sample windows must be explicit. 3. Stock profits do not establish option profits. Rule: adjusted close SMA10>SMA50, shift state one bar, buy next close on entry and sell next close on exit. Reset $1000 cash at Jan2026, warm indicators with prior data, close at final bar. Fees and slippage each5bp per side; stress each15bp. Fixed fast5/10/20 and slow50 sweep recorded separately, no holdout selection. Training2022-2025: {"symbol":"XLE","fast":10,"slow":50,"cost":0.001,"window":"train","return_pct":-31.953871260222034,"drawdown_pct":49.04440857587213,"n":16,"win_pct":25,"avgwin_pct":7.848171953296376,"avgloss_pct":5.488985524553827}. Test2026 throughSep4: {"symbol":"XLE","fast":10,"slow":50,"cost":0.001,"window":"test","return_pct":27.7076581918933,"drawdown_pct":17.269541409409594,"n":5,"win_pct":40,"avgwin_pct":16.126090537577397,"avgloss_pct":1.6779691651751725}. All1173 daily bars per symbol from https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=XLE&timeframe=1Day&start=2022-01-01&end=2026-09-05&adjustment=all&feed=iex . Implemented vectorbt. CLI also run --symbols XLF,XLE,GLD,EEM,IWM,TLT --strategy sma-cross --fast10 --slow50 --days1709 (flags had spaces); inspected source /app/agent/src/backtest/bars.ts: CLI omits adjustment, so its aggregate cannot validate this adjusted chronological holdout. Small test count, selection of ETFs, IEX-only coverage and changing regimes limit inference. No observed loss in a small sample does not imply zero loss probability. No option return sample or executable quotes, so no option allocation.

Premises: 1. Signals must precede fills. 2. Costs and out-of-sample windows must be explicit. 3. Stock profits do not establish option profits. Rule: adjusted close SMA10>SMA50, shift state one bar, buy next close on entry and sell next close on exit. Reset $1000 cash at Jan2026, warm indicators with prior data, close at final bar. Fees and slippage each5bp per side; stress each15bp. Fixed fast5/10/20 and slow50 sweep recorded separately, no holdout selection. Training2022-2025: {"symbol":"TLT","fast":10,"slow":50,"cost":0.001,"window":"train","return_pct":-2.9980825466494365,"drawdown_pct":13.82121333271441,"n":12,"win_pct":33.33333333333333,"avgwin_pct":3.635622647402317,"avgloss_pct":2.1195069287842565}. Test2026 throughSep4: {"symbol":"TLT","fast":10,"slow":50,"cost":0.001,"window":"test","return_pct":-3.9108242082088283,"drawdown_pct":5.157894555855758,"n":2,"win_pct":0,"avgwin_pct":0,"avgloss_pct":1.9752327964291077}. All1173 daily bars per symbol from https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=TLT&timeframe=1Day&start=2022-01-01&end=2026-09-05&adjustment=all&feed=iex . Implemented vectorbt. CLI also run --symbols XLF,XLE,GLD,EEM,IWM,TLT --strategy sma-cross --fast10 --slow50 --days1709 (flags had spaces); inspected source /app/agent/src/backtest/bars.ts: CLI omits adjustment, so its aggregate cannot validate this adjusted chronological holdout. Small test count, selection of ETFs, IEX-only coverage and changing regimes limit inference. No observed loss in a small sample does not imply zero loss probability. No option return sample or executable quotes, so no option allocation.