Stats

1. Model: 20000 paths,60sessions, aligned historical open/previous-close, high/open, low/open,close/open ratios, contiguous5day blocks;10day sensitivity also saved. Seed918061, baseline history pre2024 versus recent252; fees1bp/slip15bp per side. Stops include opening gaps, openingtarget first, intraday ambiguity stop first. Fixed levels asofSep17; no live entry signal conditioning, no15min/profit ratchet. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. last252/2ATR6ATR: stop first 73.045%, target first 26.720%, timeout 0.235%, mean net -0.8900%, median -10.2254%, P01 -34.4840%. 3. Half Kelly from numerical max mean log(1+f*r) gridf0..1 thenhalf, using closedtrade distributions: holdout N4, win100.00%, average win34.836%, average loss0.000%, constrained halfKelly stockweight0.5000, cap-adjusted researchsize$101.99. forward2026 N4, win25.00%, average win34.790%, average loss14.427%, constrained halfKelly stockweight0.0000, cap-adjusted researchsize$0.00. This is a diagnostic, not approved sizing; small/all-win histories cannot calibrate omitted losses. 4. Recent-window drift and selected survivor bias can create attractive resampled odds without an executable edge. Simulated tail quantiles are not loss bounds. LevelsE-2ATR/E+6ATR are hypothetical geometry and not fundamental valuation; TEN frozenvalue45.41/55.36 remains separate. Decision: do not use favorable bootstrap odds to override failed holdout, absent current signal or undetermined cash capture. A fullfund101.99 exposure is12% account849.92; quarterfund25.4975 is3% account, stockloss can exceedstop on gap.

1. Model: 20000 paths,60sessions, aligned historical open/previous-close, high/open, low/open,close/open ratios, contiguous5day blocks;10day sensitivity also saved. Seed918061, baseline history pre2024 versus recent252; fees1bp/slip15bp per side. Stops include opening gaps, openingtarget first, intraday ambiguity stop first. Fixed levels asofSep17; no live entry signal conditioning, no15min/profit ratchet. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. pre2024/2ATR6ATR: stop first 66.770%, target first 31.210%, timeout 2.020%, mean net 1.4969%, median -7.0079%, P01 -9.6989%. last252/2ATR6ATR: stop first 50.780%, target first 46.420%, timeout 2.800%, mean net 5.7770%, median -7.0079%, P01 -8.3937%. 3. Half Kelly from numerical max mean log(1+f*r) gridf0..1 thenhalf, using closedtrade distributions: train N49, win30.61%, average win22.921%, average loss7.997%, constrained halfKelly stockweight0.3845, cap-adjusted researchsize$101.99. holdout N10, win20.00%, average win19.976%, average loss7.913%, constrained halfKelly stockweight0.0000, cap-adjusted researchsize$0.00. forward2026 N7, win57.14%, average win21.282%, average loss6.479%, constrained halfKelly stockweight0.5000, cap-adjusted researchsize$101.99. This is a diagnostic, not approved sizing; small/all-win histories cannot calibrate omitted losses. 4. Recent-window drift and selected survivor bias can create attractive resampled odds without an executable edge. Simulated tail quantiles are not loss bounds. LevelsE-2ATR/E+6ATR are hypothetical geometry and not fundamental valuation; TEN frozenvalue45.41/55.36 remains separate. Decision: do not use favorable bootstrap odds to override failed holdout, absent current signal or undetermined cash capture. A fullfund101.99 exposure is12% account849.92; quarterfund25.4975 is3% account, stockloss can exceedstop on gap.

1. Model: 20000 paths,60sessions, aligned historical open/previous-close, high/open, low/open,close/open ratios, contiguous5day blocks;10day sensitivity also saved. Seed918061, baseline history pre2024 versus recent252; fees1bp/slip15bp per side. Stops include opening gaps, openingtarget first, intraday ambiguity stop first. Fixed levels asofSep17; no live entry signal conditioning, no15min/profit ratchet. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. pre2024/2ATR6ATR: stop first 72.725%, target first 26.645%, timeout 0.630%, mean net 0.0919%, median -6.2328%, P01 -7.8202%. pre2024/frozen_value: stop first 25.070%, target first 74.735%, timeout 0.195%, mean net -0.2460%, median 4.4937%, P01 -14.5717%. last252/2ATR6ATR: stop first 42.310%, target first 54.870%, timeout 2.820%, mean net 7.1037%, median 17.4205%, P01 -6.5804%. last252/frozen_value: stop first 6.900%, target first 92.640%, timeout 0.460%, mean net 3.1583%, median 4.4937%, P01 -14.2872%. 3. Half Kelly from numerical max mean log(1+f*r) gridf0..1 thenhalf, using closedtrade distributions: train N46, win19.57%, average win24.725%, average loss8.099%, constrained halfKelly stockweight0.0000, cap-adjusted researchsize$0.00. holdout N8, win37.50%, average win15.583%, average loss5.965%, constrained halfKelly stockweight0.5000, cap-adjusted researchsize$101.99. forward2026 N5, win80.00%, average win19.207%, average loss8.288%, constrained halfKelly stockweight0.5000, cap-adjusted researchsize$101.99. This is a diagnostic, not approved sizing; small/all-win histories cannot calibrate omitted losses. 4. Recent-window drift and selected survivor bias can create attractive resampled odds without an executable edge. Simulated tail quantiles are not loss bounds. LevelsE-2ATR/E+6ATR are hypothetical geometry and not fundamental valuation; TEN frozenvalue45.41/55.36 remains separate. Decision: do not use favorable bootstrap odds to override failed holdout, absent current signal or undetermined cash capture. A fullfund101.99 exposure is12% account849.92; quarterfund25.4975 is3% account, stockloss can exceedstop on gap.

1. Source https://runway.adityaperswal.workers.dev/data/stock-bars SIP/all Jan2016-Sep17. TEN has2691 daily gaps and none <=-30%; worst gap-9.5963% (Apr29 2020), worst pre2024 close return-17.58995%. That does NOT establish zero future disaster risk. Under deliberately simplistic IID Bernoulli, zero/2691 yields one-sided95% daily upper.111262%; converting that bound across60sessions gives6.46126%. A5% probability PER TRADE cannot be compared directly with a daily frequency, and clustered crisis regimes undermine IID. Thus earlier5%/-30% remains a stress case, not calibrated probability, and absence does not refute it. 2. PSIX May12 2026 actual adjusted gap-34.18735%,2025Nov7-29.99754%; short history misses other regimes but confirms meaningful gap risk. LPG worst pre2024 gap-22.27513%. 3. AA feed Nov1 2016 apparent-74.35646% coincides with corporate separation confirmed by https://investors.alcoa.com/resources/investor-faqs/default.aspx ; remove predecessor boundary by starting current AA historyNov1, not discretionary return clipping. Clean worst gap-13.2465%approximately; exact values saved in observation packets. 4. Last252 TEN/FRO correlation.819413,TEN/STNG.708182,TEN/LPG.600194; holdings would share a shipping factor, not independent bets. Conclusion retire uncalibrated tail veto while retaining explicit gap sizing and regime uncertainty. Closed-trade sample2 operator-liquidated wins cannot estimate edge. Hard account risk and fund101.99 cap remain; no forced entry from removing an invalid inference.

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"MU","bars":1003,"returnPct":33.273060525293154,"buyAndHoldPct":1775.4463428681127,"maxDrawdownPct":36.24458031131956,"winRate":41.66666666666667,"avgTradePct":0.47232312281293787}

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"ALM","bars":297,"returnPct":124.51031088911115,"buyAndHoldPct":190.14373716632446,"maxDrawdownPct":18.106251492680077,"winRate":50,"avgTradePct":4.21481208976961}

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"CENX","bars":1003,"returnPct":30.783852368281472,"buyAndHoldPct":433.07189542483655,"maxDrawdownPct":34.92360735923026,"winRate":32.5,"avgTradePct":0.525451745267276}

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"AA","bars":1003,"returnPct":-25.392344990122197,"buyAndHoldPct":5.347533632286994,"maxDrawdownPct":42.471162084949704,"winRate":32.857142857142854,"avgTradePct":-0.34057498169084205}

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"PSIX","bars":432,"returnPct":42.32466690454897,"buyAndHoldPct":51.500000000000014,"maxDrawdownPct":15.0166294688044,"winRate":41.935483870967744,"avgTradePct":1.3974287736690292}

Required CLI: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback20 --days1460 --stop-pct8 --target-pct24 (actual invocation uses spaced --lookback 20). Source https://runway.adityaperswal.workers.dev/data/stock-bars . Engine source inspection: IEX, same-close signals/entries, fixed brackets, no commissions/slippage, no holdout, and no adverse opening-gap execution. Therefore diagnostic only; independent costed calendar-split vectorbt analyses3310-3321 supersede for inference. These results are a different strategy and feed, so neither disagreement nor high return validates current eligibility. Full metrics:{"symbol":"TEN","bars":1001,"returnPct":23.779307240557834,"buyAndHoldPct":178.24367088607596,"maxDrawdownPct":19.727161495833574,"winRate":37.333333333333336,"avgTradePct":0.39570689620069754}

1. Daily SIP/all source https://runway.adityaperswal.workers.dev/data/stock-bars . NumPy OLS predicts following-open to21st-following-open20-session return using current breakout20&SMA50, ownSMA200, SPYSMA200 and20-day momentum. Training through2023, no refit for2024-25; horizons crossing boundaries excluded. Bartlett HAC20 errors address overlap approximately, normal CI; eight unadjusted tests, selected survivors, no event causal identification. 2. TEN303/323 breakout days also aboveSMA50=93.808%; FRO97.090%,STNG95.798%,LPG95.165%,AA91.558%,CENX92.453%,FTK89.844%,MU95.498%. These are strongly overlapping filters, not independent corroboration. 3. TEN signal coefficient-.69255pp over20sessions,95%CI[-4.19916,2.81407],p.69869; out-of-sample R2 versus training mean-.080415 (N481). LPG+1.46304pp,CI[-1.58348,4.50956],p.34657,R2-.042958. FTK coefficient-6.70832pp,p.003574 but OOSR2-.054114; do not turn exploratory negative coefficient into an untested short rule. 4. All eight signal coefficient CIs exceptFTK include0; six of eight OOSR2 negative. FRO/STNG modest positiveR2 do not repair negative train bracket returns. PSIX/ALM/BWLP lack long training sample. Sensitivity tests vary filter effect across periods: TEN holdout markedSMA50+13.38%,none-6.58%,own200-4.32%; LPG SMA50-25.66%,own200-.75%. No filter chosen from this holdout. Conclusion no statistically reliable extra independent signal from confluence in this selected universe; retain original hypotheses and collect prospective primary economics.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked 110.5027%, closed compound 108.0478%, DD 15.8699%, win 50.00%, N=42, open=1, targets=12. holdout: marked 25.6958%, closed compound 24.8288%, DD 6.4636%, win 45.45%, N=11, open=1, targets=3. forward2026: marked 0.5309%, closed compound 0.3321%, DD 8.4508%, win 20.00%, N=5, open=1, targets=1. Stress holdout: holdout: marked 22.8380%, closed compound 22.1126%, DD 6.8370%, win 45.45%, N=11, open=1, targets=3. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Benchmark only; outside Supply entry mandate.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked 14.8198%, closed compound 15.8252%, DD 69.0629%, win 33.33%, N=51, open=1, targets=14. holdout: marked 32.4452%, closed compound 28.6361%, DD 56.5251%, win 41.18%, N=17, open=1, targets=6. forward2026: marked 57.9976%, closed compound 54.7930%, DD 36.9163%, win 50.00%, N=6, open=1, targets=3. Stress holdout: holdout: marked 27.8900%, closed compound 24.3360%, DD 57.0437%, win 41.18%, N=17, open=1, targets=6. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Retain as research candidate; historical positivity alone does not establish current signal, valuation or executable payoff.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked -89.0755%, closed compound -89.0755%, DD 94.3601%, win 16.67%, N=36, open=0, targets=4. holdout: marked -27.3629%, closed compound -30.9049%, DD 45.5460%, win 26.32%, N=19, open=1, targets=5. forward2026: marked -29.2466%, closed compound -29.2466%, DD 32.3656%, win 16.67%, N=6, open=0, targets=1. Stress holdout: holdout: marked -30.1412%, closed compound -33.4812%, DD 46.7841%, win 26.32%, N=19, open=1, targets=5. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Baseline lacks consistent train/holdout support; reject as standalone live rule.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked 106.4773%, closed compound 118.1736%, DD 69.9238%, win 33.33%, N=42, open=1, targets=13. holdout: marked 55.3394%, closed compound 29.6065%, DD 51.1477%, win 38.46%, N=13, open=1, targets=5. forward2026: marked -28.4114%, closed compound -28.4114%, DD 45.7865%, win 20.00%, N=5, open=0, targets=1. Stress holdout: holdout: marked 51.2015%, closed compound 26.2802%, DD 51.9231%, win 38.46%, N=13, open=1, targets=5. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Retain as research candidate; historical positivity alone does not establish current signal, valuation or executable payoff.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked 132.3927%, closed compound 104.7072%, DD 59.6778%, win 35.14%, N=37, open=1, targets=12. holdout: marked 48.8489%, closed compound 32.0076%, DD 44.8167%, win 36.36%, N=11, open=1, targets=4. forward2026: marked -4.7619%, closed compound -4.7619%, DD 21.9454%, win 33.33%, N=3, open=0, targets=0. Stress holdout: holdout: marked 45.4646%, closed compound 29.1352%, DD 45.4750%, win 36.36%, N=11, open=1, targets=4. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Retain as research candidate; historical positivity alone does not establish current signal, valuation or executable payoff.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. No pre-2024 training history. holdout: marked 84.6740%, closed compound 53.3232%, DD 17.3679%, win 66.67%, N=3, open=1, targets=2. forward2026: marked -7.4838%, closed compound -7.4838%, DD 50.8007%, win 33.33%, N=6, open=0, targets=2. Stress holdout: holdout: marked 83.3860%, closed compound 52.4060%, DD 17.6154%, win 66.67%, N=3, open=1, targets=2. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Short history remains unvalidated for allocation.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. No pre-2024 training history. holdout: marked 220.1909%, closed compound 220.1909%, DD 28.9480%, win 100.00%, N=4, open=0, targets=3. forward2026: marked -10.2456%, closed compound -15.5546%, DD 42.6214%, win 25.00%, N=4, open=1, targets=1. Stress holdout: holdout: marked 217.6396%, closed compound 217.6396%, DD 28.9480%, win 100.00%, N=4, open=0, targets=3. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Short history remains unvalidated for allocation.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. No pre-2024 training history. holdout: marked -16.8897%, closed compound -20.6682%, DD 32.2133%, win 14.29%, N=7, open=1, targets=1. forward2026: marked 122.8193%, closed compound 112.7180%, DD 10.9551%, win 83.33%, N=6, open=1, targets=5. Stress holdout: holdout: marked -18.1270%, closed compound -21.7711%, DD 32.9548%, win 14.29%, N=7, open=1, targets=1. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Short history remains unvalidated for allocation.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked 19.9143%, closed compound 27.7017%, DD 78.3639%, win 30.61%, N=49, open=1, targets=13. holdout: marked -25.6595%, closed compound -25.6595%, DD 40.4370%, win 20.00%, N=10, open=0, targets=2. forward2026: marked 90.8954%, closed compound 76.5638%, DD 11.9943%, win 57.14%, N=7, open=1, targets=4. Stress holdout: holdout: marked -27.1315%, closed compound -27.1315%, DD 41.0297%, win 20.00%, N=10, open=0, targets=2. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Baseline lacks consistent train/holdout support; reject as standalone live rule.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked -34.8539%, closed compound -40.0037%, DD 77.5581%, win 28.57%, N=42, open=1, targets=9. holdout: marked 7.0304%, closed compound 7.0304%, DD 26.6478%, win 33.33%, N=9, open=0, targets=1. forward2026: marked -5.7904%, closed compound -13.0842%, DD 25.8586%, win 16.67%, N=6, open=1, targets=1. Stress holdout: holdout: marked 5.1211%, closed compound 5.1211%, DD 27.0866%, win 33.33%, N=9, open=0, targets=1. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Baseline lacks consistent train/holdout support; reject as standalone live rule.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked -77.2771%, closed compound -77.2771%, DD 84.5876%, win 21.57%, N=51, open=0, targets=10. holdout: marked -10.7601%, closed compound -10.7601%, DD 41.5061%, win 36.36%, N=11, open=0, targets=1. forward2026: marked 85.9215%, closed compound 65.4321%, DD 31.1500%, win 66.67%, N=6, open=1, targets=3. Stress holdout: holdout: marked -12.7019%, closed compound -12.7019%, DD 42.0881%, win 36.36%, N=11, open=0, targets=1. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Baseline lacks consistent train/holdout support; reject as standalone live rule.

1. Preregistered note3573, detailed methods3577. Daily SIP/all through Sep17; 2016-2023 train (AA starts Nov1 2016), 2024-25 calendar holdout, 2026 separate known-period diagnostic. Source https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Next-open prior20-close breakout AND SMA50, prior20 arithmetic ATR, E-2ATR/E+6ATR brackets, 60-session max; entry-day stops and adverse opening gaps included, target opening priority, ambiguous intraday stop first. Vectorbt $10000 reinvested fractional proxy, 1bp fee+5bp slip per side; stress15bp slip. Fixed brackets omit +2R management and actual 15min/overnight execution. 3. train: marked -68.7795%, closed compound -68.9770%, DD 86.1382%, win 19.57%, N=46, open=1, targets=8. holdout: marked 13.3817%, closed compound 13.3817%, DD 29.8290%, win 37.50%, N=8, open=0, targets=1. forward2026: marked 96.7260%, closed compound 85.0282%, DD 22.0055%, win 80.00%, N=5, open=1, targets=4. Stress holdout: holdout: marked 11.5820%, closed compound 11.5820%, DD 30.2487%, win 37.50%, N=8, open=0, targets=1. 4. L10/40 and other trend filters recorded as sensitivities, not selected by holdout. Current survivor selection, retrospective adjustments, small N, prior viewing of history and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. Marked return includes open paper; closed statistics do not. Conclusion: Baseline lacks consistent train/holdout support; reject as standalone live rule.

Premises: 1. PSIX H1 OCF75.724m less capex2.736m gives72.988m simple FCF; deduct the22.687m received tariff refunds owed or potentially owed to customers and50.301m remains, before other working-capital normalization. Latest10-Q explicitly recognizes no refund gain https://www.sec.gov/Archives/edgar/data/1137091/000162828026054328/psix-20260630.htm . 2. Sep17 SIP45.55 times June30 23.050m shares approximates1049.9275m equity value; debt72.6m less cash70.062m implies1052.4655m EV excluding leases. H1 cash yields6.9517% reported/4.7909% ex-refund, NOT annual yields; annualizing H1 net income gives21.7277x PE, a sensitivity not forecast. Source https://www.sec.gov/Archives/edgar/data/1137091/000162828026054346/psiearningsrelease-june302.htm and https://runway.adityaperswal.workers.dev/data/stock-bars . 3. Annual supplier21% concentration fell below10% in H12026; customer B/C still40% H1. Different periods resolve apparent contradiction. Expired broad Weichai collaboration does not cancel separate290.108m purchase commitments. Conclusion: cash improved, but meaningful customer claims/ramp risk prevent capitalizing gross cash as permanent rent. Strong counterargument backlog normalization could restore margins; need paid shipments and steady throughput to defend higher earnings/target. 4. ALM new minimum1720t total and conditional3mUSD upfront improve route-to-market, but undisclosed floor, annual tonnes, capex and recovery prevent calculating incremental NAV https://www.sec.gov/Archives/edgar/data/1670061/000149315226043044/ex99-1.htm . Legal/procurement claim in issuer release is not independently verified here. No speculative target from contract headline. Binding constraints are profit conversion and defensible valuation; facts end at disclosed periods.

1. ALM Sep17 https://www.sec.gov/Archives/edgar/data/1670061/000149315226043044/ex99-1.htm describes WBH/Sandvik minimum1720tWO3, conditional3mUSD upfront, indexprice/floor and possible termination. First order customerfinancing enables recycledsupply; second both ALM andWBH secure route/feed, but margin allocation unknown withoutfloor/recovery/capex; third replicatedtailings plants could erode scarcity rents. Totalvolume notannualrate, so noannuityvaluation. 2. PSIX Aug6 https://www.sec.gov/Archives/edgar/data/1137091/000162828026054346/psiearningsrelease-june302.htm: powerorders faceproductionconstraints, while FY25 concentration https://www.sec.gov/Archives/edgar/data/1137091/000162828026013207/psix-20251231.htm shows supplier21%, fourcustomers53%. First demandhelps; second component/rampcosts andbuyerconcentration affectcapture; thirdverticalintegration canrelieveconstraint butworkingcapital consumes cash. 3. EGA Aug26 https://media.ega.ae/egas-al-taweelah-smelter-restoration-reaches-25-completion-milestone/ 315/1262cells restarted, Q12027fulloutputforecast. Firstrestoration increasespotentialmetal; secondrestartedcells usemetal internally; thirdnetexports eventually reducecompetitorpremium—notoneforonecellcounttonnage. 4. MUmedia labor threat isnotconfirmedoutputloss; nextcapacityforecast2027 fromJune24issuer https://investors.micron.com/static-files/631b1a32-5537-46ae-8f40-82e42fc79dfe . Counterargumenthigherpricesbenefitproducers failswhen ownvolume fallsorcontractsfixed. Facts endatdatedoperations/contracts; returncausalityandfuturemargins remain assumptions. Noordersfromheadlinealone.