Stats

Ran node /app/agent/dist/backtest.js --symbols BTC/USD,ETH/USD,SOL/USD --strategy breakout --lookback 20 --days 981. Gross return/DD/win/N: BTC46.460751/14.692428/34.328358/67; ETH41.345409/24.943256/32.727273/55; SOL72.151727/18.795394/52.777778/36. Source code uses close-based same-close entry, no fees or holdout;982daily BTC/ETH and749SOL include partialSep13. It cannot implement actual manager cadence, credited-asset fees, partial exits or frozen net3R. Premise: a gross diagnostic cannot override negative executable-policy stress results. Compare 60 initial latency cells plus48 managed daily/hourly-on15m cells recorded separately; no choice from CLI ranking. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,60.81861551447694,36.1055745942662,63.63636363636363,11,2.4622409060517656],["immediate15","holdout",10,4.7448785135718765,23.68254187049399,50,6,1.2224725885354886],["immediate15","train",30,56.22943310896961,36.362924954319176,63.63636363636363,11,2.3286894776729596],["immediate15","holdout",30,-4.661790690341718,24.073027272110682,33.33333333333333,6,0],["manager","train",10,-11.319965224128794,32.1700886197668,40,10,0],["manager","holdout",10,2.23884532897776,23.01142571081386,33.33333333333333,6,0.7170145661278271],["manager","train",30,-12.913589047584736,32.44086751982418,40,10,0],["manager","holdout",30,-6.497253116745284,29.310515249898717,33.33333333333333,6,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,46.256440014381425,27.844944199800736,55.55555555555556,9,1.8893496247368675],["immediate15","holdout",10,-23.58855665335652,36.66451078761456,28.57142857142857,7,0],["immediate15","train",30,41.52147769142425,28.203289819312783,55.55555555555556,9,1.7329408431566498],["immediate15","holdout",30,-24.81196132288119,37.555842224623824,14.285714285714285,7,0],["manager","train",10,46.51209060420214,27.82868159430437,55.55555555555556,9,1.8927562089076304],["manager","holdout",10,-23.114313182756362,36.66451078761456,28.57142857142857,7,0],["manager","train",30,41.689738130383304,28.186901760039472,55.55555555555556,9,1.7353182387132127],["manager","holdout",30,-24.441814135943204,37.555842224623824,14.285714285714285,7,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,41.81225469010923,30.982952612358627,44.44444444444444,18,0.7912388822934207],["immediate15","holdout",10,14.279480652963313,22.575795206323658,57.14285714285714,7,3.2587428030398424],["immediate15","train",30,39.68391678335925,31.471634415770154,44.44444444444444,18,0.7694123471500396],["immediate15","holdout",30,11.930566976766073,22.88487479356266,57.14285714285714,7,2.6922138540248413],["manager","train",10,41.33153378262807,33.799480308373944,50,18,0.8641467590204002],["manager","holdout",10,-4.761685347058432,21.22734247320084,28.57142857142857,7,0],["manager","train",30,35.66754750309045,34.58914251732863,50,18,0.7886536080660962],["manager","holdout",30,-7.178192826427093,21.36119085850403,28.57142857142857,7,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,-8.611411416638859,40.93258539506645,37.5,16,0.11328611701546132],["immediate15","holdout",10,-12.47270212407561,25.20345082260248,20,5,0],["immediate15","train",30,-9.054935168625846,40.54646727369256,31.25,16,0.0968757084294225],["immediate15","holdout",30,-14.026669706591264,25.94128685814059,20,5,0],["manager","train",10,-7.241133793437456,40.93258539506647,37.5,16,0.1445287897907943],["manager","holdout",10,-12.517991817228094,25.20345082260248,20,5,0],["manager","train",30,-7.965856111354384,40.54646727369313,31.25,16,0.12006066970051443],["manager","holdout",30,-14.071155322425142,25.94128685814059,20,5,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,-22.986784517157048,44.84170257336555,33.33333333333333,21,0],["immediate15","holdout",10,5.558319556972549,15.45050611215416,50,6,2.314089532612358],["immediate15","train",30,-18.444127733175662,41.32835263345276,36.84210526315789,19,0],["immediate15","holdout",30,-0.6142790493896086,15.863899707075579,33.33333333333333,6,0.16224451770763793],["manager","train",10,3.605153035365362,34.318680373038134,38.88888888888889,18,0.456774082219499],["manager","holdout",10,0.6012353133369288,15.138951433057047,33.33333333333333,6,0.4883210743709352],["manager","train",30,-6.751293193289612,33.963077991874805,38.88888888888889,18,0.0765597050196738],["manager","holdout",30,-8.981052390595906,15.98964773474817,28.57142857142857,7,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

C91306 actual-cadence extension, 48 predetermined cells. Daily HIGH20 or hourly HIGH480 false-to-true events; prior daily ATR20 excludes signal day. Execute first eligible 15m open immediately or at actual manager UTC schedule: daily00/02/04/06/08; weekdays12:30/13:50/16:30/19:40/21, rounded up to14:00/19:45 where needed. First event retained if signals collide; expire after2h; delayed price must still exceed frozen high. Stop2ATR; +2grossR sellquarter and move stop to entry only at selected cadence. Residual target solved ex ante for aggregate net3R including25bp credited-asset fees and10/30bp slippage. Stops/targets checked every15m open, overshoots charged;20day time exit waits for selected cadence. Each train2024–25 and known2026Jan1–Sep12 holdout starts flat; final available close liquidation costed. Vectorbt from_orders ledger cash agrees<1e−5. All8cells=[mode,window,slipbp,return%,DD%,win%,N,approx halfKelly]: [["immediate15","train",10,-23.668860167219393,30.328471266594047,36.84210526315789,19,0],["immediate15","holdout",10,-8.554393363816658,24.818515274881538,33.33333333333333,6,0],["immediate15","train",30,-24.836402055925234,31.58067707025294,31.57894736842105,19,0],["immediate15","holdout",30,-8.924188024561658,25.31098266214705,33.33333333333333,6,0],["manager","train",10,-23.197542299712197,30.32847126659445,36.84210526315789,19,0],["manager","holdout",10,-8.440499931558236,24.818515274881538,33.33333333333333,6,0],["manager","train",30,-24.56075778660467,31.579696620238728,31.57894736842105,19,0],["manager","holdout",30,-8.81194220961109,25.31098266214705,33.33333333333333,6,0]]. Increasing slippage also changes fill-anchored exits and subsequent occupancy, so stress loss is not merely subtracting fees. Missing15m BTC151/ETH149 of94656slots; SOL97 of71744 fromAug26 2024, incomplete training. Missing quote skips action, no assumed fill. Fullsample missed-event counter includes pre2024 warmup; not a withinwindow rate. This is an idealized bar-open execution model, not archived bids/asks or fill proof; exact run-duration delays and network outages remain unmodeled. Conclusion reject promoting this rule now: all daily manager holdouts negative; hourly BTC/SOL turn negative at30bpslip, ETH manager already negative. No fresh entry exists. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees ; user UTC schedule.

Predeclared HIGH480 false-to-true events,2prior-dailyATR20 stop,gross3R target,20dexit. Each train2024–25/knownholdout2026Jan1–Sep12 startsflat;2023warmup; full calendarreindex and no missingbar signal. Freeze ATR excluding signal UTCday; delayedopen must exceed frozenhigh. Exits at hourlyopens withovershoot, terminalclosecosted. Exact25bpcreditedcoinbuyfee,25bpsell+10/30bpslip; vectorbt from_orders finalcash reconciles<1e−5. Summary arrays=[window,delayhours,slipbp,return%,DD%,win%,N,halfKelly]: [["train",0,10,25.80951170226426,38.75805309385727,54.54545454545454,11,1.1074234688921152],["holdout",0,10,4.696690429715815,22.47771688729755,50,6,1.2266567262412873],["train",0,30,24.128878339405176,39.27212760725306,54.54545454545454,11,1.1035837157870332],["holdout",0,30,2.213872095176569,23.40242892296066,50,6,0.7948438348463944],["train",2,10,44.26454674413809,30.34690065215855,66.66666666666666,9,2.5778743326517204],["holdout",2,10,10.745037978099148,16.212938884416406,60,5,3.748471646301409],["train",2,30,39.16337300086592,30.901904875038653,66.66666666666666,9,2.3406796367351257],["holdout",2,30,8.552129893534755,16.54741917174499,60,5,3.044340137669421],["train",4,10,36.69165039103146,26.838415518202975,55.55555555555556,9,2.478271835536935],["holdout",4,10,-5.625189184198987,20.550740655707767,16.666666666666664,6,0],["train",4,30,13.784420505082947,33.73180647313123,44.44444444444444,9,0.9567173990314317],["holdout",4,30,-7.863230421705353,21.498438316177893,16.666666666666664,6,0]]. All metrics and trade returns separately in60observations; fulltimestamp ledgers results.json. This target is gross3R, NOTnet3R/quartermanagement. Notliveauthorization; separate15m managerextension uses samepolicy withincomparison. SOL missingtraining history, smallN and inspectedholdout limitinference; no winningdelay adopted. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Predeclared HIGH480 false-to-true events,2prior-dailyATR20 stop,gross3R target,20dexit. Each train2024–25/knownholdout2026Jan1–Sep12 startsflat;2023warmup; full calendarreindex and no missingbar signal. Freeze ATR excluding signal UTCday; delayedopen must exceed frozenhigh. Exits at hourlyopens withovershoot, terminalclosecosted. Exact25bpcreditedcoinbuyfee,25bpsell+10/30bpslip; vectorbt from_orders finalcash reconciles<1e−5. Summary arrays=[window,delayhours,slipbp,return%,DD%,win%,N,halfKelly]: [["train",0,10,30.213027025341177,46.15312671371922,47.368421052631575,19,0.6655620017244688],["holdout",0,10,13.733733534189312,22.33626371317833,57.14285714285714,7,3.2111305573046494],["train",0,30,17.323180684257437,47.430072562762525,47.368421052631575,19,0.5315881447167456],["holdout",0,30,10.59334598278349,22.646299516825607,57.14285714285714,7,2.5415440053830203],["train",2,10,65.13182818830985,29.70900019235322,58.82352941176471,17,1.3087744922116473],["holdout",2,10,-5.058183811965861,22.270999613474316,33.33333333333333,6,0],["train",2,30,58.22971245385128,30.998762313050896,58.82352941176471,17,1.2421399986722799],["holdout",2,30,-7.3096712370086445,22.58129595322985,33.33333333333333,6,0],["train",4,10,90.02254429139484,36.88848596186753,58.82352941176471,17,1.5944071109014066],["holdout",4,10,2.4861778723921333,19.11969765402035,42.857142857142854,7,1.2278289411408494],["train",4,30,80.86009144118312,38.138183030829694,52.94117647058824,17,1.5580817012878787],["holdout",4,30,-9.16236274477933,22.91093510051304,42.857142857142854,7,0]]. All metrics and trade returns separately in60observations; fulltimestamp ledgers results.json. This target is gross3R, NOTnet3R/quartermanagement. Notliveauthorization; separate15m managerextension uses samepolicy withincomparison. SOL missingtraining history, smallN and inspectedholdout limitinference; no winningdelay adopted. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Predeclared HIGH480 false-to-true events,2prior-dailyATR20 stop,gross3R target,20dexit. Each train2024–25/knownholdout2026Jan1–Sep12 startsflat;2023warmup; full calendarreindex and no missingbar signal. Freeze ATR excluding signal UTCday; delayedopen must exceed frozenhigh. Exits at hourlyopens withovershoot, terminalclosecosted. Exact25bpcreditedcoinbuyfee,25bpsell+10/30bpslip; vectorbt from_orders finalcash reconciles<1e−5. Summary arrays=[window,delayhours,slipbp,return%,DD%,win%,N,halfKelly]: [["train",0,10,-34.10207940996422,51.223500264434584,31.818181818181817,22,0],["holdout",0,10,3.865085512335696,14.969439154901975,50,6,1.9140143661182405],["train",0,30,-34.282329645730755,51.08963483343785,31.818181818181817,22,0],["holdout",0,30,1.8313690835902605,15.627950319962947,50,6,1.1118445833222594],["train",2,10,-5.6929659613824555,36.152899252152224,35,20,0.13411144355404447],["holdout",2,10,-0.37151101460021607,15.905558141881693,33.33333333333333,6,0.19618486599530183],["train",2,30,-11.752650092185462,38.00140267095728,35,20,0],["holdout",2,30,-2.4671337768238715,17.01318308981469,33.33333333333333,6,0],["train",4,10,-10.086221598371736,38.0388548608827,42.10526315789473,19,0],["holdout",4,10,2.5817921205252503,13.96480952579604,50,6,1.0229440715258162],["train",4,30,-16.22131508368463,41.63473902675122,36.84210526315789,19,0],["holdout",4,30,0.665531823142814,14.89383800138655,50,6,0.5123497529498771]]. All metrics and trade returns separately in60observations; fulltimestamp ledgers results.json. This target is gross3R, NOTnet3R/quartermanagement. Notliveauthorization; separate15m managerextension uses samepolicy withincomparison. SOL missingtraining history, smallN and inspectedholdout limitinference; no winningdelay adopted. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Diagnostic C91306, predeclared60 cells. Fresh false-to-true close>prior20calendar highs daily or480hour highs hourly. Train2024–25; known2026Jan1–Sep12holdout, each startsflat;2023warmup. Freeze ATR20 excluding signal UTCday, entry nextopen or delay; cancel delayedopen<=frozenhigh. Stop entry−2ATR,targetentry+6ATR,time20days; check at each sampledopen, gapovershootcharged, terminalclosecosted. Gross3R target notnet3R or mandatedquarter/BE policy; NOT live authorization. Exact25bp creditedcoin buyfee (vectorbt cash-equivalent1/.9975−1),25bpsell,10/30bpslip; customorderledger vectorbt finalcash reconciles<1e−5. No fill-forward missingprices forsignals;calendar reindex. SOL training incomplete. All cells=[{"return":34.458878026702806,"drawdown":31.588470493420452,"win_rate":55.55555555555556,"sample":9,"mean_trade_pct":4.142065346208124,"half_kelly_fraction":1.51244063285213,"skipped_entries":0,"days":731,"symbol":"SOL","tf":"1Day","delay":0,"slippage_bp":10,"window":"train"},{"return":-29.803963815426744,"drawdown":39.02815504192447,"win_rate":28.57142857142857,"sample":7,"mean_trade_pct":-4.466468344010586,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"SOL","tf":"1Day","delay":0,"slippage_bp":10,"window":"holdout"},{"return":29.704431327084198,"drawdown":32.13358173937147,"win_rate":55.55555555555556,"sample":9,"mean_trade_pct":3.726327313674249,"half_kelly_fraction":1.350967339190396,"skipped_entries":0,"days":731,"symbol":"SOL","tf":"1Day","delay":0,"slippage_bp":30,"window":"train"},{"return":-26.398607832625544,"drawdown":35.29850243033073,"win_rate":14.285714285714285,"sample":7,"mean_trade_pct":-3.8927833838910852,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"SOL","tf":"1Day","delay":0,"slippage_bp":30,"window":"holdout"},{"return":25.420885921607205,"drawdown":30.301096474759305,"win_rate":57.14285714285714,"sample":7,"mean_trade_pct":4.488646425436358,"half_kelly_fraction":1.259588472030996,"skipped_entries":3,"days":731,"symbol":"SOL","tf":"1Day","delay":1,"slippage_bp":10,"window":"train"},{"return":-12.867061844365734,"drawdown":27.75180407776515,"win_rate":16.666666666666664,"sample":6,"mean_trade_pct":-1.9201880405994836,"half_kelly_fraction":0,"skipped_entries":1,"days":255,"symbol":"SOL","tf":"1Day","delay":1,"slippage_bp":10,"window":"holdout"},{"return":21.95779562644765,"drawdown":30.718316751062513,"win_rate":57.14285714285714,"sample":7,"mean_trade_pct":4.071524831557579,"half_kelly_fraction":1.1396809107852504,"skipped_entries":3,"days":731,"symbol":"SOL","tf":"1Day","delay":1,"slippage_bp":30,"window":"train"},{"return":-14.933366476414351,"drawdown":28.613605016046417,"win_rate":16.666666666666664,"sample":6,"mean_trade_pct":-2.3117253883536195,"half_kelly_fraction":0,"skipped_entries":1,"days":255,"symbol":"SOL","tf":"1Day","delay":1,"slippage_bp":30,"window":"holdout"}]. Models ignore historical quotes/depth/vintages, bar-open execution is idealized; compare15m manager experiment separately. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Diagnostic C91306, predeclared60 cells. Fresh false-to-true close>prior20calendar highs daily or480hour highs hourly. Train2024–25; known2026Jan1–Sep12holdout, each startsflat;2023warmup. Freeze ATR20 excluding signal UTCday, entry nextopen or delay; cancel delayedopen<=frozenhigh. Stop entry−2ATR,targetentry+6ATR,time20days; check at each sampledopen, gapovershootcharged, terminalclosecosted. Gross3R target notnet3R or mandatedquarter/BE policy; NOT live authorization. Exact25bp creditedcoin buyfee (vectorbt cash-equivalent1/.9975−1),25bpsell,10/30bpslip; customorderledger vectorbt finalcash reconciles<1e−5. No fill-forward missingprices forsignals;calendar reindex. SOL training incomplete. All cells=[{"return":-14.708540624975669,"drawdown":44.64384424276863,"win_rate":37.5,"sample":16,"mean_trade_pct":0.008970593893272566,"half_kelly_fraction":0.0024873558075877433,"skipped_entries":0,"days":731,"symbol":"ETH","tf":"1Day","delay":0,"slippage_bp":10,"window":"train"},{"return":-14.267479191675875,"drawdown":26.66399395121769,"win_rate":20,"sample":5,"mean_trade_pct":-2.705731396618394,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"ETH","tf":"1Day","delay":0,"slippage_bp":10,"window":"holdout"},{"return":-19.996206876800684,"drawdown":45.63156263014141,"win_rate":31.25,"sample":16,"mean_trade_pct":-0.39026800899233693,"half_kelly_fraction":0,"skipped_entries":0,"days":731,"symbol":"ETH","tf":"1Day","delay":0,"slippage_bp":30,"window":"train"},{"return":-15.965104127502231,"drawdown":27.24834401499575,"win_rate":20,"sample":5,"mean_trade_pct":-3.0941328333627305,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"ETH","tf":"1Day","delay":0,"slippage_bp":30,"window":"holdout"},{"return":36.47267904464448,"drawdown":21.57214775719677,"win_rate":50,"sample":12,"mean_trade_pct":3.538484334478619,"half_kelly_fraction":1.230337788564526,"skipped_entries":4,"days":731,"symbol":"ETH","tf":"1Day","delay":1,"slippage_bp":10,"window":"train"},{"return":-27.458382948521432,"drawdown":31.940702695446642,"win_rate":20,"sample":5,"mean_trade_pct":-5.9858843962766635,"half_kelly_fraction":0,"skipped_entries":3,"days":255,"symbol":"ETH","tf":"1Day","delay":1,"slippage_bp":10,"window":"holdout"},{"return":0.6462735851387151,"drawdown":33.212708089850764,"win_rate":41.66666666666667,"sample":12,"mean_trade_pct":0.9303542086100828,"half_kelly_fraction":0.3220385665883565,"skipped_entries":4,"days":731,"symbol":"ETH","tf":"1Day","delay":1,"slippage_bp":30,"window":"train"},{"return":-28.89480936909855,"drawdown":32.88662968098834,"win_rate":20,"sample":5,"mean_trade_pct":-6.361191370663713,"half_kelly_fraction":0,"skipped_entries":3,"days":255,"symbol":"ETH","tf":"1Day","delay":1,"slippage_bp":30,"window":"holdout"}]. Models ignore historical quotes/depth/vintages, bar-open execution is idealized; compare15m manager experiment separately. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Diagnostic C91306, predeclared60 cells. Fresh false-to-true close>prior20calendar highs daily or480hour highs hourly. Train2024–25; known2026Jan1–Sep12holdout, each startsflat;2023warmup. Freeze ATR20 excluding signal UTCday, entry nextopen or delay; cancel delayedopen<=frozenhigh. Stop entry−2ATR,targetentry+6ATR,time20days; check at each sampledopen, gapovershootcharged, terminalclosecosted. Gross3R target notnet3R or mandatedquarter/BE policy; NOT live authorization. Exact25bp creditedcoin buyfee (vectorbt cash-equivalent1/.9975−1),25bpsell,10/30bpslip; customorderledger vectorbt finalcash reconciles<1e−5. No fill-forward missingprices forsignals;calendar reindex. SOL training incomplete. All cells=[{"return":-20.246747205256327,"drawdown":30.99747645438119,"win_rate":38.88888888888889,"sample":18,"mean_trade_pct":-0.807128911609015,"half_kelly_fraction":0,"skipped_entries":0,"days":731,"symbol":"BTC","tf":"1Day","delay":0,"slippage_bp":10,"window":"train"},{"return":-10.117357948740391,"drawdown":24.74016665003479,"win_rate":33.33333333333333,"sample":6,"mean_trade_pct":-1.4488758062713305,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"BTC","tf":"1Day","delay":0,"slippage_bp":10,"window":"holdout"},{"return":-30.93986553801804,"drawdown":34.9768857956267,"win_rate":31.57894736842105,"sample":19,"mean_trade_pct":-1.5044362646029363,"half_kelly_fraction":0,"skipped_entries":0,"days":731,"symbol":"BTC","tf":"1Day","delay":0,"slippage_bp":30,"window":"train"},{"return":-12.248870136238835,"drawdown":25.93474703271117,"win_rate":33.33333333333333,"sample":6,"mean_trade_pct":-1.8422946456240648,"half_kelly_fraction":0,"skipped_entries":0,"days":255,"symbol":"BTC","tf":"1Day","delay":0,"slippage_bp":30,"window":"holdout"},{"return":-25.735904053020175,"drawdown":35.95149351306597,"win_rate":43.75,"sample":16,"mean_trade_pct":-1.315661643226236,"half_kelly_fraction":0,"skipped_entries":7,"days":731,"symbol":"BTC","tf":"1Day","delay":1,"slippage_bp":10,"window":"train"},{"return":-0.2686556274030494,"drawdown":13.776620128535388,"win_rate":40,"sample":5,"mean_trade_pct":0.21390039637057848,"half_kelly_fraction":0.2250106253509645,"skipped_entries":1,"days":255,"symbol":"BTC","tf":"1Day","delay":1,"slippage_bp":10,"window":"holdout"},{"return":-27.41915671025025,"drawdown":35.92711984837932,"win_rate":43.75,"sample":16,"mean_trade_pct":-1.4781360920575894,"half_kelly_fraction":0,"skipped_entries":7,"days":731,"symbol":"BTC","tf":"1Day","delay":1,"slippage_bp":30,"window":"train"},{"return":-0.21227381922949462,"drawdown":13.034926770540778,"win_rate":40,"sample":5,"mean_trade_pct":0.19279753982137687,"half_kelly_fraction":0.22387335315499568,"skipped_entries":1,"days":255,"symbol":"BTC","tf":"1Day","delay":1,"slippage_bp":30,"window":"holdout"}]. Models ignore historical quotes/depth/vintages, bar-open execution is idealized; compare15m manager experiment separately. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees .

Ran node /app/agent/dist/backtest.js --symbols BTC/USD,ETH/USD,SOL/USD --strategy sma-cross --fast 1 --slow 200 --days 981. Grossreturn/DD/win/closedN:BTC14.4761/31.2332/15.3846/13;ETH10.7680/37.8551/11.1111/9;SOL8.3228/32.9545/0/3. Sourceengine same-closeexecution, nofees/holdout;982barsBTCETH/749SOLincludingpartialSep13bar, openmark-to-marketexplainspositiveSOLtotalwith0closedwins. Cannotimplementstablecoinsignal; suppliesonlymandatoryCLIcontrol. Costed daily/hourly168cells andholdoutsrecordedseparately. Noorder,nochoicefromgrossoutput. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return-11.006206%, DD34.649737%, win0.000%, N3; holdout return5.433297%, DD10.758956%, win100.000%, N2. Price-only holdout18.366768% N1. Daily signal mapped to hourly opens; identical fills when00UTC exists, hourly marks expose greater intraday drawdown. This is not independent replication or an hourly supply series. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return-11.006206%, DD33.249904%, win0.000%, N3; holdout return5.433297%, DD9.663307%, win100.000%, N2. Price-only holdout18.366768% N1. Daily portfolio marks understate intraday drawdown; hourly companion uses identical information timing. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return75.031958%, DD35.057611%, win33.333%, N9; holdout return3.666842%, DD6.462538%, win100.000%, N2. Price-only holdout11.301377% N1. Daily signal mapped to hourly opens; identical fills when00UTC exists, hourly marks expose greater intraday drawdown. This is not independent replication or an hourly supply series. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return75.031958%, DD33.073196%, win33.333%, N9; holdout return3.666842%, DD4.774546%, win100.000%, N2. Price-only holdout11.301377% N1. Daily portfolio marks understate intraday drawdown; hourly companion uses identical information timing. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return30.157692%, DD36.117294%, win27.778%, N18; holdout return-2.140782%, DD6.395037%, win50.000%, N2. Price-only holdout10.679989% N1. Daily signal mapped to hourly opens; identical fills when00UTC exists, hourly marks expose greater intraday drawdown. This is not independent replication or an hourly supply series. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Frozen baseline: lag2 calendar-day USDC30-day supply growth>0 AND previous completed daily close>SMA200. Enter first available bar open when state becomes true; exit first available open when false. Split starts flat and initializes if state true, terminal liquidation at last close. Training2024–25, known holdout2026Jan1–Sep12, 2023warmup, no partialSep13bar. Vectorbt100USD all-in single-symbol independent portfolios, 25bpfee+10bpslip per side; stress30bpslip. Sweeps lag1/2/7 and combinedUSDC+USDT, controlprice-only; no parameter selection from holdout. Baseline training return30.157692%, DD33.907307%, win27.778%, N18; holdout return-2.140782%, DD5.824592%, win50.000%, N2. Price-only holdout10.679989% N1. Daily portfolio marks understate intraday drawdown; hourly companion uses identical information timing. Current-vintage supply lacks publication archive; assumed lags cannot remove revisions. Fees modeled in quote currency are a close proxy, not exact venue credited-coin accounting. No stops/targets/+2R partial/trailing policy here, so baseline cannot authorize trading; incremental filter is not adopted. SOL has418missing calendar days across2023–24; missing SMA means unavailable, not bearish; trade counts small. Half-Kelly recorded only where both wins/losses exist, uncapped fractions unstable; allocationzero. All28 cells recorded separately in observations. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://docs.alpaca.markets/us/docs/crypto-fees

Ran node /app/agent/dist/backtest.js --symbols BTC/USD,ETH/USD,SOL/USD,LTC/USD,XRP/USD,HYPE/USD,LINK/USD --strategy breakout --lookback 20 --days 981. Results (return%,drawdown%,win%,closed trades): BTC/USD 46.4608/14.6924/34.328/67; ETH/USD 41.4426/24.9433/32.727/55; SOL/USD 72.1517/18.7954/52.778/36; LTC/USD -17.5515/47.2206/27.083/48; XRP/USD 476.5253/19.9595/47.059/34; HYPE/USD -1.6918/19.2206/29.412/17; LINK/USD -22.5658/44.6927/34.545/55. Source inspection confirms prior20 CLOSE toggle and same-close execution, no fees or held-out split; current partial September13 daily bar included. Consequently these gross figures do not validate HIGH20/LOW10 or managed pullback execution. The separate 156-cell vectorbt study3077–3090 supplies costed daily/hourly controls and calendar holdouts; HYPE lacks training. No parameter selected from this diagnostic and no order. Source https://runway.adityaperswal.workers.dev/data/crypto-bars . All raw metrics recorded in observations26628–26841.

Fixed Donchian prior20 calendar-day HIGH entry / prior10day LOW exit at next exact OPEN; 480/240hours. Unfilled missing calendar hours; prices beforeSep13 02UTC only, incompleteSep13day excluded. Vectorbt creditedcoinbuy fee equivalent .0025/.9975, sell25bp; slip10bp/30bpstress. Training2024–25 return0.296718%,DD52.438495%,closedN12,win33.333%,openN0. Reset2026 known/reusedholdout return-1.201527%,DD24.551662%,closedN4,win50.000%,openN1; stressreturn-2.963609%. All15/20/30day sensitivities recorded, no bestparameter selected. Calendarholdout is not newly untouched data. Returnincludesopenmarkedposition;winrateclosedonly, no assumedfinalfill. No required2ATRstop/3Rtarget/quarterat2R/15minexitmanagement in this CONTROL, so notlivepolicyvalidation. FreshHIGHentryabsent across7, no late entry. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees . Experiment156cells observations26910–27075; currenttop10survivorselection limits generalization.

Fixed Donchian prior20 calendar-day HIGH entry / prior10day LOW exit at next exact OPEN; 20/10days. OHLCrequires24hours/day; prices beforeSep13 02UTC only, incompleteSep13day excluded. Vectorbt creditedcoinbuy fee equivalent .0025/.9975, sell25bp; slip10bp/30bpstress. Training2024–25 return-39.654890%,DD56.876500%,closedN11,win36.364%,openN0. Reset2026 known/reusedholdout return27.044492%,DD14.403432%,closedN2,win50.000%,openN1; stressreturn25.780875%. All15/20/30day sensitivities recorded, no bestparameter selected. Calendarholdout is not newly untouched data. Returnincludesopenmarkedposition;winrateclosedonly, no assumedfinalfill. No required2ATRstop/3Rtarget/quarterat2R/15minexitmanagement in this CONTROL, so notlivepolicyvalidation. FreshHIGHentryabsent across7, no late entry. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees . Experiment156cells observations26910–27075; currenttop10survivorselection limits generalization.

Fixed Donchian prior20 calendar-day HIGH entry / prior10day LOW exit at next exact OPEN; 480/240hours. Unfilled missing calendar hours; prices beforeSep13 02UTC only, incompleteSep13day excluded. Vectorbt creditedcoinbuy fee equivalent .0025/.9975, sell25bp; slip10bp/30bpstress. Training2024–25 unavailable. Reset2026 known/reusedholdout return56.037399%,DD31.948567%,closedN4,win75.000%,openN0; stressreturn53.560657%. All15/20/30day sensitivities recorded, no bestparameter selected. HYPEstartsFeb17 2026: positive2026results are IN-SAMPLE descriptive, no pre2026training or valid OOS. Returnincludesopenmarkedposition;winrateclosedonly, no assumedfinalfill. No required2ATRstop/3Rtarget/quarterat2R/15minexitmanagement in this CONTROL, so notlivepolicyvalidation. FreshHIGHentryabsent across7, no late entry. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees . Experiment156cells observations26910–27075; currenttop10survivorselection limits generalization.

Fixed Donchian prior20 calendar-day HIGH entry / prior10day LOW exit at next exact OPEN; 20/10days. OHLCrequires24hours/day; prices beforeSep13 02UTC only, incompleteSep13day excluded. Vectorbt creditedcoinbuy fee equivalent .0025/.9975, sell25bp; slip10bp/30bpstress. Training2024–25 unavailable. Reset2026 known/reusedholdout return37.837008%,DD30.053656%,closedN3,win66.667%,openN0; stressreturn36.192841%. All15/20/30day sensitivities recorded, no bestparameter selected. HYPEstartsFeb17 2026: positive2026results are IN-SAMPLE descriptive, no pre2026training or valid OOS. Returnincludesopenmarkedposition;winrateclosedonly, no assumedfinalfill. No required2ATRstop/3Rtarget/quarterat2R/15minexitmanagement in this CONTROL, so notlivepolicyvalidation. FreshHIGHentryabsent across7, no late entry. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/crypto-fees . Experiment156cells observations26910–27075; currenttop10survivorselection limits generalization.