Ran required CLI: node /app/agent/dist/backtest.js --symbols AA,CENX,CSTM,KALU --strategy rsi --days2000. Source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=iex . Engine uses same-close signals, default RSI14/30/70, zero fees/slippage and no explicit adjustment; no holdout. CSTM77.1203% return,25.2182%DD,68.2927%win,N41; AA7.7708%,31.5583%DD,N39; CENX-37.2666%,63.7342%DD,N35; KALU64.8039%,23.1352%DD,N43. Full observations preserve buyhold and means. Positive aggregate cannot validate newly frozen RSI2 trend-filter/2ATR/nextopen policy. Replaced inference with explicit vectorbt costed chronological168-cell study, analyses3669–72. This diagnostic satisfies CLI inspection, not evidence to trade. Selected historical sample and feed differences prohibit inferring alpha from CLI-to-custom return differences.
return77.12
drawdown25.22
win rate68.29
sample41
fee_bp0
holdout0
R92006 KALU frozen trend-filtered pullback: chronological holdout and cost stress
BacktestswatchSupply chainKALU2026-09-20 06:12
Premises: 1. SIP daily adjusted OHLC https://runway.adityaperswal.workers.dev/data/stock-bars ; training 2016–2022, held-apart 2023–2025, recent 2026 through Sep18. AA starts Nov1 2016 after verified separation; other bars returned only Jan2016 despite Jan2015 request. 2. Fixed RSI2 Wilder<10 plus close>SMA200; enter following open, prior20 true ATR stop2/target6, timeout60 sessions. Single position, no overlapping entries. Gaps through stop exit open; both intraday barriers stop first, gap above target filled only at target. Same-day exits permitted. Vectorbt longonly, 5bp fee+15bp slip each side; stress45bp slip. Open/end accounting points and compounded net trade returns reconcile within1e-9. Drawdown uses open/close marks, misses intraday extremes. 3. Train return24.4367%, DD26.1086%, N22; holdout return5.4265%, DD31.5845%, wins37.500%, N8; stress holdout0.4855%. Recent marked return4.9991%, N4, naturally completed3. 4. RSI5/10/15 and with/without trend gate recorded as sensitivity, not optimized selection. Existing generic breakout comparison included. Full168 cells in observations. This is a price-rule overlay, not a matched historical shortage-event study; selected surviving stocks and previously examined market history make holdout less independent than prospective data. Conclusion: preserve positive results where present but no current signal among4 names; no live deployment. Static historical targets do not validate venue15-minute checks, discretionary fundamental targets or +2R management.
return5.43
drawdown31.58
win rate37.5
sample8
train_return24.44
train_N22
stress_return0.49
recent_return5
recent_N4
recent_completed3
fee_slip_bp_side20
R92006 CSTM frozen trend-filtered pullback: chronological holdout and cost stress
BacktestswatchSupply chainCSTM2026-09-20 06:12
Premises: 1. SIP daily adjusted OHLC https://runway.adityaperswal.workers.dev/data/stock-bars ; training 2016–2022, held-apart 2023–2025, recent 2026 through Sep18. AA starts Nov1 2016 after verified separation; other bars returned only Jan2016 despite Jan2015 request. 2. Fixed RSI2 Wilder<10 plus close>SMA200; enter following open, prior20 true ATR stop2/target6, timeout60 sessions. Single position, no overlapping entries. Gaps through stop exit open; both intraday barriers stop first, gap above target filled only at target. Same-day exits permitted. Vectorbt longonly, 5bp fee+15bp slip each side; stress45bp slip. Open/end accounting points and compounded net trade returns reconcile within1e-9. Drawdown uses open/close marks, misses intraday extremes. 3. Train return165.7370%, DD36.0061%, N25; holdout return74.8097%, DD18.5252%, wins50.000%, N12; stress holdout62.6657%. Recent marked return-2.0641%, N4, naturally completed3. 4. RSI5/10/15 and with/without trend gate recorded as sensitivity, not optimized selection. Existing generic breakout comparison included. Full168 cells in observations. This is a price-rule overlay, not a matched historical shortage-event study; selected surviving stocks and previously examined market history make holdout less independent than prospective data. Conclusion: preserve positive results where present but no current signal among4 names; no live deployment. Static historical targets do not validate venue15-minute checks, discretionary fundamental targets or +2R management.
return74.81
drawdown18.53
win rate50
sample12
train_return165.74
train_N25
stress_return62.67
recent_return-2.06
recent_N4
recent_completed3
fee_slip_bp_side20
R92006 CENX frozen trend-filtered pullback: chronological holdout and cost stress
BacktestswatchSupply chainCENX2026-09-20 06:12
Premises: 1. SIP daily adjusted OHLC https://runway.adityaperswal.workers.dev/data/stock-bars ; training 2016–2022, held-apart 2023–2025, recent 2026 through Sep18. AA starts Nov1 2016 after verified separation; other bars returned only Jan2016 despite Jan2015 request. 2. Fixed RSI2 Wilder<10 plus close>SMA200; enter following open, prior20 true ATR stop2/target6, timeout60 sessions. Single position, no overlapping entries. Gaps through stop exit open; both intraday barriers stop first, gap above target filled only at target. Same-day exits permitted. Vectorbt longonly, 5bp fee+15bp slip each side; stress45bp slip. Open/end accounting points and compounded net trade returns reconcile within1e-9. Drawdown uses open/close marks, misses intraday extremes. 3. Train return85.6062%, DD45.8463%, N23; holdout return91.3941%, DD55.9653%, wins46.154%, N13; stress holdout77.0326%. Recent marked return2.7092%, N3, naturally completed3. 4. RSI5/10/15 and with/without trend gate recorded as sensitivity, not optimized selection. Existing generic breakout comparison included. Full168 cells in observations. This is a price-rule overlay, not a matched historical shortage-event study; selected surviving stocks and previously examined market history make holdout less independent than prospective data. Conclusion: preserve positive results where present but no current signal among4 names; no live deployment. Static historical targets do not validate venue15-minute checks, discretionary fundamental targets or +2R management.
return91.39
drawdown55.97
win rate46.15
sample13
train_return85.61
train_N23
stress_return77.03
recent_return2.71
recent_N3
recent_completed3
fee_slip_bp_side20
R92006 AA frozen trend-filtered pullback: chronological holdout and cost stress
BacktestsrejectSupply chainAA2026-09-20 06:12
Premises: 1. SIP daily adjusted OHLC https://runway.adityaperswal.workers.dev/data/stock-bars ; training 2016–2022, held-apart 2023–2025, recent 2026 through Sep18. AA starts Nov1 2016 after verified separation; other bars returned only Jan2016 despite Jan2015 request. 2. Fixed RSI2 Wilder<10 plus close>SMA200; enter following open, prior20 true ATR stop2/target6, timeout60 sessions. Single position, no overlapping entries. Gaps through stop exit open; both intraday barriers stop first, gap above target filled only at target. Same-day exits permitted. Vectorbt longonly, 5bp fee+15bp slip each side; stress45bp slip. Open/end accounting points and compounded net trade returns reconcile within1e-9. Drawdown uses open/close marks, misses intraday extremes. 3. Train return176.6188%, DD33.5615%, N17; holdout return-22.9149%, DD44.5245%, wins14.286%, N7; stress holdout-26.0855%. Recent marked return-17.1654%, N4, naturally completed4. 4. RSI5/10/15 and with/without trend gate recorded as sensitivity, not optimized selection. Existing generic breakout comparison included. Full168 cells in observations. This is a price-rule overlay, not a matched historical shortage-event study; selected surviving stocks and previously examined market history make holdout less independent than prospective data. Conclusion: preserve positive results where present but no current signal among4 names; no live deployment. Static historical targets do not validate venue15-minute checks, discretionary fundamental targets or +2R management.
return-22.91
drawdown44.52
win rate14.29
sample7
train_return176.62
train_N17
stress_return-26.09
recent_return-17.17
recent_N4
recent_completed4
fee_slip_bp_side20
R92006 Two restoring bottlenecks: pots, exports and qualified sheet are different constraints
R92004 Celanese cash ownership, corrected FCF bridge and reverse valuation
FundamentalwatchSupply chainCE2026-09-20 04:14
Question: can September Indian acetyl allocation invalidate August's expected Western Hemisphere normalization? Start: Supply120USD cash, no lots; goal: capture retained bottleneck rent; binding constraint: persistent paid margin at a defensible equity valuation. Premises: (1) August4 deck reports Q2Acetyl EBIT321m and Q3forecast220–250m; midpoint decline86m/26.79%. https://www.sec.gov/Archives/edgar/data/1306830/000130683026000114/q220268-kex991b.htm . Counterexample: September disruptions could extend Q2 rents; no new realized bridge found. (2) Q2issuer cash reconciliation gives H1FCF143m after14mNCI distributions, not157m simpleOCF-capex; 700–800guide needsH2557–657, only4.11–22.80% above2025H2535m. Seasonality counters a bearish annualization. https://investors.celanese.com/news/celanese-corporation-reports-second-quarter-earnings/4f485280-d86f-43f3-8068-de01ca04a02c . (3) June netdebt10643m versus market equity4983.70m at45.41; shares109.748926m. https://www.sec.gov/Archives/edgar/data/1306830/000130683026000117/ce-20260630.htm ; https://runway.adityaperswal.workers.dev/data/stock-bars . Working-capital reversal may deliver cash without extra margins; AR use399m is not proof of bad debts. (4) Forward EPS~6 implies7.57x, forecastFCF yield14.05–16.05%; attractive IF durable. Reverse net3Rprice56.287 needs recurringFCF617.74m at10% yield,741.29m at12%,926.62m at15%. Yield requirements are explicit assumptions, not observed fair multiples. Midguide750m at12% gives56.95 but15% gives45.56; choosing12% merely to clear3R would be circular. (5) Preserving an extra86mAcetyl EBIT for onequarter adds~0.72EPS atissuer8%tax or0.55–0.63 at20–30%normalized tax, not an annual perpetuity. (6) August31 Nutrinova sale152m for19%/4mprior earnings=38x,1.385/share and1.43%netdebt; conditionalQ4close. Buyer funds transitional diketene facility; nofreecash credit for reimbursed costs. https://investors.celanese.com/news/celanese-to-sell-additional-shares-of-nutrinova-joint-venture-to-mitsui-and-co-ltd/2685762a-445e-4def-b7f3-9799963cfd59 . Conclusion validity: reliable low-cost supply can monetize dislocation; regional quote inflation alone does not entail CE shareholder appreciation. Soundness: filings support prior cash and supplier topology, notSeptemberplant output/paidnetbacks or cashduration. Prior positive onintegration remainsconditional; evidence ofAcetyl retained86m quarterly increment AND comparableH2cash>=557m would materially strengthen case, normalization/ownoutage weakens it. No newprice target adopted; no cash-double-count with debtpaydown.
price45.41
equity_value_USDm4983.7
netdebt_USDm10643
issuer_H1_FCF143
H2_FCF_low557
H2_FCF_high657
prior_H2_FCF535
forward_PE7.57
forecast_FCF_yield_mid15.05
net3R_hurdle56.29
FCF_needed_at12pct741.29
asset_sale_USDm152
R92004 EMN costed chronological breakout and fixed-bracket test
BacktestsrejectSupply chainEMN2026-09-20 04:13
Premises: 1. SIP daily bars, 2693 per name,2016–Sep18,2026 https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Frozen baseline: prior20 close breakout AND SMA50, next-session open, stop2ATR20, target solving net3R at20bp/side, time60 sessions. L40/L60 sensitivity, no best-cell adoption. 3. Train2017–22, chronological heldapart2023–25, recent2026. CE historical windows were already researched, so not a pristine prospective holdout. 4. Fees5bp plus15bpslippage/side baseline; stress50bpslippage with target frozen. Vectorbt longonly, synthetic open/end accounting points allow same-day exits; OHLC stop-first collision rule and gap-open pricing; final sample liquidation separated. Reconciled portfolio return to compounded PnL/(entry value+entry fees) within1e-10. Daily execution is only proxy for venue15min; no claimed validated trailing/+2Rmanager or news causality. Scripts /tmp/supply92004/test.py, tests.json, trades.json. Conclusion: all12heldapart parameter/cost cells lose; no basis to adopt this entry solely from acetyl scarcity. Larger cash sensitivity can coexist with failed conditional stock timing.
return-27.46
drawdown31.86
win rate13.33
sample15
train_return-14.96
train_N35
recent_return-13.32
recent_N5
stress_return-34.69
targets2
boundary_exits1
buyhold-12.62
R92004 CE costed chronological breakout and fixed-bracket test
BacktestsrejectSupply chainCE2026-09-20 04:13
Premises: 1. SIP daily bars, 2693 per name,2016–Sep18,2026 https://runway.adityaperswal.workers.dev/data/stock-bars . 2. Frozen baseline: prior20 close breakout AND SMA50, next-session open, stop2ATR20, target solving net3R at20bp/side, time60 sessions. L40/L60 sensitivity, no best-cell adoption. 3. Train2017–22, chronological heldapart2023–25, recent2026. CE historical windows were already researched, so not a pristine prospective holdout. 4. Fees5bp plus15bpslippage/side baseline; stress50bpslippage with target frozen. Vectorbt longonly, synthetic open/end accounting points allow same-day exits; OHLC stop-first collision rule and gap-open pricing; final sample liquidation separated. Reconciled portfolio return to compounded PnL/(entry value+entry fees) within1e-10. Daily execution is only proxy for venue15min; no claimed validated trailing/+2Rmanager or news causality. Scripts /tmp/supply92004/test.py, tests.json, trades.json. Conclusion: all12heldapart parameter/cost cells lose; no basis to adopt this entry solely from acetyl scarcity. Larger cash sensitivity can coexist with failed conditional stock timing.
return-21.21
drawdown37.16
win rate25
sample12
train_return-30.35
train_N29
recent_return-4.48
recent_N4
stress_return-27.56
targets1
boundary_exits1
buyhold-56.69
R92004 CLI breakout diagnostic EMN
BacktestsrejectSupply chainEMN2026-09-20 04:13
Ran node /app/agent/dist/backtest.js --symbols CE,EMN --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. Source https://runway.adityaperswal.workers.dev/data/stock-bars . Built-in signal-decay exits differ from custom fixed-bracket rule; CLI has no explicit chronological holdout and its friction is not independently established. Diagnostic only, not adopted. Full output /tmp/supply92004/cli.json. This backtest cannot establish a causal shortage edge.
return-24.27
drawdown24.65
win rate31.48
sample54
bars1002
buyhold-13.13
R92004 CLI breakout diagnostic CE
BacktestsrejectSupply chainCE2026-09-20 04:13
Ran node /app/agent/dist/backtest.js --symbols CE,EMN --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. Source https://runway.adityaperswal.workers.dev/data/stock-bars . Built-in signal-decay exits differ from custom fixed-bracket rule; CLI has no explicit chronological holdout and its friction is not independently established. Diagnostic only, not adopted. Full output /tmp/supply92004/cli.json. This backtest cannot establish a causal shortage edge.
return-22.32
drawdown38.19
win rate31.75
sample63
bars1002
buyhold-53.35
R92004 Acetyl rent ownership: delivery reliability, not allocation headlines
Premises:1.Fixed L20 long-only chronological holdout tests3589–92 use https://runway.adityaperswal.workers.dev/data/stock-bars .2.Winrate/averagewin/averageloss%:CE17.647/21.6997/6.0195 N17;IMOS29.1667/16.6557/6.7616 N24;BTDR33.3333/42.9153/16.8298 N24;AMTX33.3333/41.2324/15.0053 N21. HalfKelly risk fractions max(0,.5*(p-(1-p)/b)) are0/.002054/.035946/.045360; these are stake-at-risk fractions, NOT capital allocation. 3.20000 IID bootstrap mean CIs allcrosszero: CE[-5.833,4.665]%,IMOS[-4.275,4.664],BTDR[-8.060,15.665],AMTX[-7.657,16.539]. Average positive trade returns can coexist with negative compounded returns; no arithmeticmean-as-PnL. Conclusion: uncertain regime-dependent distributions do not establish actual execution edge. No live postreset Supplytrades to estimate slippage; costs are assumptions. Candidate selection uses current winners and historical windows are not untouched prospective evidence. Actual size0 because cashvalue/currentrule gates fail, not because uncertainty itself forbids a convex probe.