Stats

Observed sources: EGA August12 report gives227 cells as of MondayAugust10, https://media.ega.ae/ega-delivers-resilient-h1-2026-performance-maintaining-operational-and-supply-chain-continuity-amid-regional-disruption/ ; August26 gives315/1262, roughly one week stabilization, https://media.ega.ae/egas-al-taweelah-smelter-restoration-reaches-25-completion-milestone/ . Implied interval speed88/16=5.5cells/day;947remaining/5.5=172.18days plus7stabilization reachesFebruary21,2027. This rough extrapolation is consistent with, not independent corroboration of, issuerQ1 target. Assumptions: all315 existing cells stable, identical cell capacity, cast output proxy1.6Mt/year from https://media.ega.ae/update-on-egas-al-taweelah-site/ , no new damage, no diversion for restarting cells. NumPy50000paths seed91104; fixed speed perpath triangular3/5.5/8cells/day, arbitrary uncertainty not estimated. Completion days afterAug26 P5/50/95=138.35/178.92/256.53; model81.662% beforeMarch31 is NOT real-world forecast odds. Integrated missing cast-output proxy throughMarch31 P5/50/95=237423/304145/421743tonnes. Heterogeneous cells, hot-metal reuse, alloy content, existing stocks and exports make delivered-tonne precision lower. Conclusion: use scenarios to bound questions for the next operational update; do not derive a stock target or trade probability from them. New measured output and net shipped volumes matter more than this simulation.

Use training payoffs, not favorable holdout, from fixed tests2521–2522; source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all . CSTM train p=40%, mean win8.4956%, mean loss7.6217% gives clamped halfKelly0. KALU p=41.9355%, win10.7182%, loss7.6758% gives rough halfKelly0.176252% of equity; quarter of that is0.0440631%, or$0.374973 at equity850.99. This binary-payoff approximation is a sizing ceiling, not proof of optimal allocation; stress costs make KALU training mean negative and halfKelly0. Both currently fail entry, so actual allocation0. Whole-share KALU158.42 and stock minimum1 exceed the hypothetical probe; do not round up. NumPy50,000 paths of20 iid training roundtrips, seed91104: allocating25% per trade yields CSTM median account return-6.56%, loss probability73.34%,95th-percentile drawdown22.50%; KALU median-0.98%, loss probability53.40%,95th drawdown19.89%. These are conditional empirical simulations, not calibrated future loss bounds, and omit unseen gap/persistent regime risk. Historical training target-hit frequencies only2/30 CSTM and2/31 KALU, versus stop15/30 and18/31. Conclusion: venue maximum is not a position recommendation; no funded probe justified by these distributions. Funding shortfall does not change odds, target or size.

Premises:1 The documented reversal mechanismcompensatesliquiditysuppliers, notallbuyersoffallingprices.2 Sectoraggregation mixesstock-specificshockswithsharedcashflownews.3 ETFshare/NAVarbitrage requirescapacityandcreationunits; dailyfundamentalpricescancontinuetomove. ChainA (liquidityshock assumption): first forcedseller depressesprice; second fundedintermediariesabsorb inventory/arbitrageETFbasket; third temporarydiscountmayclose beforeour nextopen,leavingsmalleredgeforthecash-onlybuyer. ChainB (cashflowshock assumption): first sectorrevenue/marginoutlookfalls; second earningsrevisionsandfundoutflowsreinforcepricechange; third fadinglosersexposesfundtopersistentlossesandgapsthroughstop. ChainC (crowding): first highvolraisesprospectiveliquiditycompensation; second otherliquiditysupplierswithdraw orfacefundinglimits; third survivingprovidersmayearnmorebutpoorlyfundedretailpositionscanexitbeforethepayoff. Whobenefitsdependsoncause: informed sellersavoidpermanentloss; liquidmarketmakerscapturetemporaryinventorypremium; issuerfeerevenue doesnotproveholderalpha. Facts endatpapers/productstructure; thesecausalsequencesarehypotheses,notobservedSep11flows. Testedweeklyselectiondidnotoutperformmatchedsectorbasket andstressfailed,so noorder. Sources https://www.nber.org/papers/w17653 ; https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr513.pdf ; https://www.ssga.com/us/en/intermediary/etfs/state-street-industrial-select-sector-spdr-etf-xli ; backtestsREV9110400.

Premises:1 XLIissuerSep9characteristics P/B6.63,FY1P/E24.59,indexprice/cashflow19.05,83holdings; theseaggregatesdonotmeasureanindividualforcedseller orincremental cashflowrevision.2 NAV171.75vsclosing171.79,premium0.03%,median30dayspread0.01%;0.08%annualexpenseembeddedinETFbarreturns.3 Secondarysharesarenotindividuallyredeemable; creationunitmechanismavailabletospecialistparticipants, withdiscountspossibleinstress. Conclusion cheap execution andaffordability do not establish assetundervaluation, a NAVarbitrage, orreversal probability; nofundamentalprice target inferredfromP/E. Who benefits: issuer collectsexpensebase, intermediariesmaycapturecreation/redemptionpricinggaps; retailbuyer receivessectorcashflowexposureandrisk. Strongestcounterargument true liquidityshocks canbeindependentoffundamentals;needdatedflow/inventoryevidence orvalidatedresidualsignal toidentifythem. Noissuer10-KcashforecastwasinventedforanETF. Source https://www.ssga.com/us/en/intermediary/etfs/state-street-industrial-select-sector-spdr-etf-xli ; observation24106.

Premises:1 Sample primary5 stressholdout dailycashportfolio, notbestneighbor, asrecorded.2 Seed9110400 draws10000paths fromoverlapping21-sessionblocks,12blocks=252sessions. Preserveswithinblockdependence, missesnewregimes,longmemoryandunseenfat-tailgaps; notforecast.3 Applyingfull187.22USDfund gives21session P5/median/P95PnL −13.13308/−1.058883/+14.38771USD,55.84%lossprobability; zero empiricalpaths gain349.01USD, whichisnotproofimpossibility.252session return P5/median/P95−25.72362/−3.98388/+24.46864%,DDmedian/P95 17.16606/30.95414%.4 HalfKelly=0.5*(p−(1−p)/(avgwin/avgloss)),clippedat0. Primary stress p44.3137%,W/L1.22364 =>negativeunclippedKelly,live0; trainingstressalso0. Baselineplugin3.146%doesnotoverridecostfailure. Accountnormal3%risk25.5297USD,positionceiling212.7475,fund187.22; alladditionalcaps, notentitlementtomaximumsize. Afullfundgaplosswouldbe22%account, sofundcapaloneisnotagapbudget. Conclusion sizezero; fundingneeddoesnotdeterminebarriers. Counterargumentresamplingmissesfuturebonanza accepted,butnoindependentpositiveedgejustifiespayingforitwiththisrule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; research/inference.py/json.

Premises:1 Primary stress holdout255 trades wins44.3137%,meanwin2.30310%,meanloss1.88218% => realized payoff1.22364, despite nominal3Rbarrier. Only9target/gaptarget exits;123time exits,106stops,17gapstops;35quarterpartials. Meanholding3.6784sessions,30same-dayfinalexits and31firstdaypartial-or-finalexits. Mediantrade−0.91748%,worst−5.56926%,best+8.43914%,maxlossstreak7.2 Signals use priorcompletebars. Currentweek modelcandidateXLY atSep8open114.47,stressentry114.641705,priorATR1.486295,stop113.155410,target119.100590; these are hypothetical historicallevels, notneworders orprospective success. Currentincompleteweekexcludedfromscoredbacktests.3 Sep10 ranks XLV−4.21509%,XLB−4.13598%,XLY−2.52481%; closes165.66/50.76/111.96. SPY20vol8.53648%below20%gate. No midweekcatchup; nextweekrank requiresSep11close.4 $187.22 buys atmost1XLV/1XLY or3XLB beforefees; quarterlyfractionalpartial cannotpreservewhole-shareafterhourlot. ADV20USD allnine>571m; smallnotionalimpactisnotbinding,but spreads,15minpolls,quarterexits,andaccountPDT sharedwithotherfundsare. Conclusion no liveentry: frequencylimitedweekly stillhasentrydayexit risk; exact15minexitpolicy andinteger sizing untested. Strongestcounterargument9tailwinsmaypay; recorded costsandmeanexpectancyrejectthatclaimfor thisrule. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; research/inference.json.

Frozen2177 applies five separately normalized lag1..5 negative cross-sectional-return weights to nine ETFs and averages them. Individual overlays are long1/short1; offsets make aggregate gross<2, correcting shorthand in2177, with no renormalization. Cash-only control clips aggregate negatives and normalizes positives to1. Next-open returns use weights known previousclose; same-close reference is explicitly non-executable. Findings: long-short nextopen holdout gross+3.38%, mean0.381bp/day; one-way turnover break-even0.6897bp, versus6bp baseline yielding−35.60% and16bp−70.75%. Training gross+22.55%,6bp−18.91%,break-even3.164bp. Long-only baseline holdout+20.48% but stress−51.74%; this positive baseline retains marketrisk and doesnotvalidate short-termselection. Highvol condition doesnot rescue holdout: long-short nextopen6bp mean−4.61bp/day, longonly−5.29. These are identification calculations with linear turnover costs, no borrow/financing/terminal-liquidation charge and no actual short order; do not rank as executable backtests. They are not the weekly ATR/3R portfolio, which separately reconciles with vectorbt. Dailyindustry results neednotreplicate monthlywithin-stockresults. All18 descriptivecells recorded; no parameter promoted. Source research/diagnostic.py/json and https://runway.adityaperswal.workers.dev/data/stock-bars ; methodological primary https://www.nber.org/papers/w17653 .

Premise1 frozen20dayHIGH/10dayLOW,2ATR14 risk,net3Rtarget accounting25bpcreditedassetfees/10bpslip,20daytimeout; signal known after completed bar,next exact open; +2RcostBE and peak-2entryATRtrail,retainedtarget. 20bpslipstress leaves targetfixed. Nativecalendar-hourly uses20completed UTCdays, rolling-hourly prior480hours. Premise2 chronological2024-25train/REUSED2026Jan1-Sep11 00UTCholdout,resetflat; hourly/daily open sampled exits, no guarantee of stopprice; rawwealth reconcilesvectorbt1e-10. Fullnativecalendar-hourly train 21.988671% N12; holdout-10.135134% DD35.033237% win28.571429% N7; same-targetstress-11.384479%. Premise3 attribution ladder fullnative daily/daily-8.216357%,daily/hourly-27.081318%,calendar/hourly-10.135134%; matched-full24h daily/daily7.041002%,rolling480/hourly1.172455%. Matching drops March12 incompletehourday and suppresseslaterentries, a selectioneffect not pricealpha. Native vs aggregate matched2026returns coincide; source construction not explanation for that ladder. Positive matched variants cannot replace previously frozen rejected policy; no switching based on this reused sample. Conclusion watch only; current20dayentryabsent. Validity: controlled changes attribute differences within implementation; soundness: tinyreusedsamples, historicalbar anomalies and sampledquotes limit predictive/executionconfidence. 80cells perasset (168matched+72fullnative total240), not80independenttrials. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; methodnotes2175/2180.

Premise1 frozen20dayHIGH/10dayLOW,2ATR14 risk,net3Rtarget accounting25bpcreditedassetfees/10bpslip,20daytimeout; signal known after completed bar,next exact open; +2RcostBE and peak-2entryATRtrail,retainedtarget. 20bpslipstress leaves targetfixed. Nativecalendar-hourly uses20completed UTCdays, rolling-hourly prior480hours. Premise2 chronological2024-25train/REUSED2026Jan1-Sep11 00UTCholdout,resetflat; hourly/daily open sampled exits, no guarantee of stopprice; rawwealth reconcilesvectorbt1e-10. Fullnativecalendar-hourly train 41.228668% N20; holdout7.324178% DD19.830197% win42.857143% N7; same-targetstress5.832105%. Premise3 attribution ladder fullnative daily/daily-17.632466%,daily/hourly-16.866373%,calendar/hourly7.324178%; matched-full24h daily/daily-6.406080%,rolling480/hourly17.240439%. Matching drops March12 incompletehourday and suppresseslaterentries, a selectioneffect not pricealpha. Native vs aggregate matched2026returns coincide; source construction not explanation for that ladder. Daily original-17.632466 and hourly original17.240439 reproduced in their respective policies. A failed daily policy is not by itself a veto of a distinct hourly policy; retain frozenhourlywatch, not parameterhunt. Conclusion watch only; current20dayentryabsent. Validity: controlled changes attribute differences within implementation; soundness: tinyreusedsamples, historicalbar anomalies and sampledquotes limit predictive/executionconfidence. 80cells perasset (168matched+72fullnative total240), not80independenttrials. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; methodnotes2175/2180.

Premise1 frozen20dayHIGH/10dayLOW,2ATR14 risk,net3Rtarget accounting25bpcreditedassetfees/10bpslip,20daytimeout; signal known after completed bar,next exact open; +2RcostBE and peak-2entryATRtrail,retainedtarget. 20bpslipstress leaves targetfixed. Nativecalendar-hourly uses20completed UTCdays, rolling-hourly prior480hours. Premise2 chronological2024-25train/REUSED2026Jan1-Sep11 00UTCholdout,resetflat; hourly/daily open sampled exits, no guarantee of stopprice; rawwealth reconcilesvectorbt1e-10. Fullnativecalendar-hourly train 30.436744% N20; holdout2.078879% DD19.776351% win42.857143% N7; same-targetstress0.659729%. Premise3 attribution ladder fullnative daily/daily-3.355267%,daily/hourly-7.958797%,calendar/hourly2.078879%; matched-full24h daily/daily7.421737%,rolling480/hourly7.635753%. Matching drops March12 incompletehourday and suppresseslaterentries, a selectioneffect not pricealpha. Native vs aggregate matched2026returns coincide; source construction not explanation for that ladder. Positive matched variants cannot replace previously frozen rejected policy; no switching based on this reused sample. Conclusion watch only; current20dayentryabsent. Validity: controlled changes attribute differences within implementation; soundness: tinyreusedsamples, historicalbar anomalies and sampledquotes limit predictive/executionconfidence. 80cells perasset (168matched+72fullnative total240), not80independenttrials. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; methodnotes2175/2180.

Preregistered note2178: close above prior20closehigh and SMA50, next session open entry, stop2 prior20ADR, target6ADR (3R), exit20th session close, no overlaps or boundary-crossing trades. Train2020–24; chronological holdout2025–Sep10 2026. SIP alladjusted source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all . vectorbt from_orders direction longonly; entry/exit counts and fee-inclusive compounded returns reconciled. Two accounting rows per day are not observed intraday fills; stop-first on conflicts, gaps at open; no ambiguous bars in this sample. Fees10bp plus slippage10bp per side; stress20bp. Train {"symbol":"KALU","lookback":20,"period":"train","slippage":0.001,"return_pct":-14.048808275721857,"drawdown_pct":55.15054140326862,"n":31,"win_pct":41.935483870967744,"mean_pct":0.037782042531024046,"avg_win_pct":10.718160567175962,"avg_loss_pct":7.675824669712543,"half_kelly":0.0017625245625975394,"targets":2,"stops":18,"time_exits":11,"ambiguous":0,"block_mean_CI_pct":[-3.7034761041582973,3.7334277526978497]}. Holdout {"symbol":"KALU","lookback":20,"period":"holdout","slippage":0.001,"return_pct":47.757195511096015,"drawdown_pct":28.952594454141124,"n":17,"win_pct":47.05882352941176,"mean_pct":2.9397191657662316,"avg_win_pct":13.818911217710431,"avg_loss_pct":6.7306737692952785,"half_kelly":0.10636580261108641,"targets":4,"stops":8,"time_exits":5,"ambiguous":0,"block_mean_CI_pct":[-2.181648799186779,7.518684316145904]}. Stress holdout return 42.817883455928765%. Four-trade circular block bootstrap, 10,000 samples seed91104, mean confidence interval crosses zero. L10/L40 neighbors are diagnostics only; all rows recorded, no winning parameter chosen. Test is a generic technical benchmark, not an event-causal test, and the family/regime has already been inspected. It omits live15min exit sampling, +2R gain management, exact dividend cash and executable quotes. Conclusion: research watch only, no current signal and no defensible new allocation. Positive holdout alone cannot fix negative compounded training or support retrospective sizing.

Preregistered note2178: close above prior20closehigh and SMA50, next session open entry, stop2 prior20ADR, target6ADR (3R), exit20th session close, no overlaps or boundary-crossing trades. Train2020–24; chronological holdout2025–Sep10 2026. SIP alladjusted source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all . vectorbt from_orders direction longonly; entry/exit counts and fee-inclusive compounded returns reconciled. Two accounting rows per day are not observed intraday fills; stop-first on conflicts, gaps at open; no ambiguous bars in this sample. Fees10bp plus slippage10bp per side; stress20bp. Train {"symbol":"CSTM","lookback":20,"period":"train","slippage":0.001,"return_pct":-38.61151000307898,"drawdown_pct":47.11561891853783,"n":30,"win_pct":40,"mean_pct":-1.1747752262880142,"avg_win_pct":8.495596940707726,"avg_loss_pct":7.621690004285176,"half_kelly":0,"targets":2,"stops":15,"time_exits":13,"ambiguous":0,"block_mean_CI_pct":[-5.367075983564136,3.0106106059726225]}. Holdout {"symbol":"CSTM","lookback":20,"period":"holdout","slippage":0.001,"return_pct":31.783908071503248,"drawdown_pct":19.56254607203325,"n":15,"win_pct":46.666666666666664,"mean_pct":2.320655535863315,"avg_win_pct":11.991223743047435,"avg_loss_pct":6.14109164542279,"half_kelly":0.09676475002014875,"targets":2,"stops":5,"time_exits":8,"ambiguous":0,"block_mean_CI_pct":[-2.497770448188134,7.632229134682402]}. Stress holdout return 27.889096029296635%. Four-trade circular block bootstrap, 10,000 samples seed91104, mean confidence interval crosses zero. L10/L40 neighbors are diagnostics only; all rows recorded, no winning parameter chosen. Test is a generic technical benchmark, not an event-causal test, and the family/regime has already been inspected. It omits live15min exit sampling, +2R gain management, exact dividend cash and executable quotes. Conclusion: research watch only, no current signal and no defensible new allocation. Positive holdout alone cannot fix negative compounded training or support retrospective sizing.

Edge test frozen Sep11: BB S7.54, Oct2C8 close.40 requires+11.405836%, covers3/12 (25%); Oct2P7 close.32 requires-11.405836%, covers2/12 (16.67%). Twelve consecutive quarterly releases Sep2023-Jun2026, pre-newsclose to firstpost-newsclose, medianabsolute9.987505%; first8 C2/P1, last4 C1/P1. Neither40%gate nor documented3xexpectedoptionpayout passes. Oct2expiry8daysafter Sep24event; defaultSep23RTHexit, hurdleexceedsmedian1.418331pp so noholdthrough exception. Sources: issuer-dated event observations24113–24126, https://investors.blackberry.com/ ; https://runway.adityaperswal.workers.dev/data/stock-bars and /data/option-bars. 1. Start state: profitable software supplier with uneven cash generation and a stock price below20/50day averages. Goal: an affordable, primary-dated option with multi-premium payoff. Binding constraint: current directional coverage, not conviction in the turnaround. 2. First order, stronger paid development licenses and SecuSUITE deliveries can improve reported revenue; second order, automaker production and secure communications renewals determine follow-on royalties and cash; third order, sustained receipts support R&D and capital returns, benefiting shareholders and the software ecosystem. 3. Adverse chain: delayed production or license timing weakens receipts; then fixed engineering costs compress cash margins; then buybacks and discretionary projects compete for funding, reducing upside capture. These are conditional mechanisms, not forecasts for other issuers. 4. Counterargument: certification and embedded customer relationships can support durable economics even while quarterly licenses are lumpy; current filings support both statements. Primary mechanism https://www.sec.gov/Archives/edgar/data/1070235/000107023526000076/bbry-20260531.htm . No independent customer-volume confirmation or measured options expectancy; no outside-mandate FUND proposal.

Edge test frozen Sep11: BB S7.54, Oct2C8 close.40 requires+11.405836%, covers3/12 (25%); Oct2P7 close.32 requires-11.405836%, covers2/12 (16.67%). Twelve consecutive quarterly releases Sep2023-Jun2026, pre-newsclose to firstpost-newsclose, medianabsolute9.987505%; first8 C2/P1, last4 C1/P1. Neither40%gate nor documented3xexpectedoptionpayout passes. Oct2expiry8daysafter Sep24event; defaultSep23RTHexit, hurdleexceedsmedian1.418331pp so noholdthrough exception. Sources: issuer-dated event observations24113–24126, https://investors.blackberry.com/ ; https://runway.adityaperswal.workers.dev/data/stock-bars and /data/option-bars. 1. Issuer Q1 baseline and Q2 guidance are recorded in observations24176/24178; sources https://www.sec.gov/Archives/edgar/data/1070235/000107023526000071/q1fy27ex-991.htm and https://www.sec.gov/Archives/edgar/data/1070235/000107023526000076/bbry-20260531.htm . Current quarterly revenue growth is real, but much of secure-communications growth came from SecuSUITE licenses while retention remains below100%; stronger sales and weaker recurring retention can both be true. 2. Q1 cash generation after property/intangible spending was approximately0.1m; achieving100m annual operatingcash guidance requires95.4m over three quarters, averaging31.8m. Seasonality is a counterargument: FY26Q4 delivered45.6mOCF of50.3mfull-year. 3. Using May31 share/cash balances with Sep10stock gives4.4189bn equity,4.2070bn EV excluding restrictedcash;6.925x revenue-guide midpoint and32.612x adjustedEBITDA midpoint. This is not a current reconciled valuation or an earnings surprise forecast. 4. QNX950m royaltybacklog is not comparable with133.9m contractedRPO: usage-royalty exclusions differ. Customers benefit from certified software; shareholders need conversion into collectedcash. Full new-quarter receipts and incremental paid volumes would strengthen business confidence, but no arbitrary probability upgrade overrides the measured option gate.

Edge test frozen Sep11: BB S7.54, Oct2C8 close.40 requires+11.405836%, covers3/12 (25%); Oct2P7 close.32 requires-11.405836%, covers2/12 (16.67%). Twelve consecutive quarterly releases Sep2023-Jun2026, pre-newsclose to firstpost-newsclose, medianabsolute9.987505%; first8 C2/P1, last4 C1/P1. Neither40%gate nor documented3xexpectedoptionpayout passes. Oct2expiry8daysafter Sep24event; defaultSep23RTHexit, hurdleexceedsmedian1.418331pp so noholdthrough exception. Sources: issuer-dated event observations24113–24126, https://investors.blackberry.com/ ; https://runway.adityaperswal.workers.dev/data/stock-bars and /data/option-bars. 1. Fixed historical close moves are applied to7.54; value Oct2 contracts with8days remaining using European Black-Scholes, rate/dividend0 and assumed post-event annual IV30/50/80/120%. This is sensitivity, not observed IV, American executable value, or a calibrated forecast; the pre-event stock path is omitted. 2. Call mean gross payout .7999/.9036/1.1211/1.4841 times .40 premium; put .2878/.3628/.5480/.8956 times .32. No scenario gives3xmean or a4xsingle-event gross target. 3. NumPy20k paths,max20 whole-contract tickets,178.71fundcash and672.28otherbooks fixed,1.30assumed roundtrip fee, each new premium<=5% current equity. Intrinsic-only liquidation yields halfKelly0 both. Call93.815% cannot complete20tickets due to risk/funding gates; median finalcash106.6357 and funddrawdown46.2201%. Put100% cannotcomplete20; medianfinal16.6977/drawdown90.6565%. These are deliberately zero-residual stress results, not ruin probabilities or actual expectancy. 4. Strongest countercase is persistent high post-event volatility and pre-event rally; neither is measured with historical option executions. Cheap-threshold scenario is recorded separately, not permission to substitute today's larger premium.

Edge test frozen Sep11: BB S7.54, Oct2C8 close.40 requires+11.405836%, covers3/12 (25%); Oct2P7 close.32 requires-11.405836%, covers2/12 (16.67%). Twelve consecutive quarterly releases Sep2023-Jun2026, pre-newsclose to firstpost-newsclose, medianabsolute9.987505%; first8 C2/P1, last4 C1/P1. Neither40%gate nor documented3xexpectedoptionpayout passes. Oct2expiry8daysafter Sep24event; defaultSep23RTHexit, hurdleexceedsmedian1.418331pp so noholdthrough exception. Sources: issuer-dated event observations24113–24126, https://investors.blackberry.com/ ; https://runway.adityaperswal.workers.dev/data/stock-bars and /data/option-bars. 1. Twelve dates were frozen before response measurement; eight earlier releases train, four later releases hold out. Vectorbt uses T-3/T-5/T-10 open entries and T-1 close exits, 5bp fees plus 10bp slippage each side; stress raises slippage to30bp. Full $100 stock allocation is a diagnostic, outside this fund's instrument mandate. 2. Training returns respectively -10.3639/-2.73664/-20.6952%; select T-5 by training return only. Selected holdout +2.23248%, daily drawdown9.40210%, wins2/4. Stress train-5.79981%, holdout+.609767%, drawdown9.76377%. Unselected T-10 holdout+23.2792% must not replace negative training. 3. Separate required CLI momentum20 diagnostic:684bars,N27,return+261.0066%,drawdown36.1282%,win40.7407%; no chronological holdout in that CLI output, so no adoption. 4. Validity: an unsplit momentum winner cannot establish an earnings option edge. Soundness: daily time-only stock proxy omits option theta, volatility crush, spreads, and mandated option stops/targets/15minute checks. All training losses plus failed event coverage leave sizezero.

Premises:1 All20 weekly costed cells reconcile to vectorbt; holdout2021–Sep10 2026 with full-current-week excluded, historical window not globally untouched.2 Primary baseline holdout +34.9556%, CAGR5.43259%, market beta0.453798 and annual simple OLS intercept−0.051691%, HAC5 t−0.00823; this is not evidence of alpha. Stress beta0.453772, annual intercept−8.00789%, t−1.27233. Fixed-beta OLS cannot eliminate all conditional exposures.3 On the same220 training weekly opportunities, time-only loser mean0.038393% vs equal9 basket0.192879%, selection delta−0.154486pp (month-cluster5000 CI[−0.392151,+0.073100]). On255 holdout weeks loser0.128452% vs equal9 0.172813% vs SPY0.212909%; delta−0.044361pp CI[−0.305723,+0.216633]. Costs identical6bp/side; no claim significant negative delta, just absent demonstrated incremental selection.4 Full-window uncosted SPY buyhold +103.0918% training/+121.2061% holdout, equal9 +82.8131/+113.6551%; these have different exposure and cannot substitute for matched control. Conclusion: positive baseline total is compatible with equity exposure; signal-selection evidence fails. Counterexample: nonlinear crash recovery might escape fixed beta; pre-frozen highvol model still stressed−7.2856% holdout,N41,CIcross0. Uncertainty, nine-sector scope and changing index composition prevent universal anti-reversal claim. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; research/results.json,inference.json and matched_train/test.csv.

Question tree2171, frozen experiments2174/2177. Start state: Quant $187.22 cash, account $850.99, no open/pending lots verified. Goal: repeatable net return; binding constraint is retaining the economic mechanism in an affordable executable portfolio. Premises:1 Nagel2011 uses five lag-specific daily stock/48industry long-short portfolios and removes conditional market exposure; transaction-price liquidity provision is not a market-order backtest.2 Da/Liu/Schaumburg2011 separate stock-specific reversal from continuing industry/cash-flow news.3 AQR2012 execution research rejects institutional reversal after costs, but its capacity conclusion does not by itself reject a tiny ETF account.4 Issuer XLI states creation-unit redemption and potential stress discounts: retail shares are not a free NAV redemption option. Inference: the long-only sector adaptation removes diversification/hedges and changes the source of expected compensation, so it must be tested independently. Strongest counterargument: low ETF spreads and tiny size could make liquidity provision viable; tested with baseline6bp per side, stress16bp and matched opportunity control, not assumed false. Contradiction resolved by different populations, horizons, cost definitions and conditioning: high-volatility industry reversal can coexist with monthly industry momentum. No additional source was changing this implementation distinction; remaining gaps are point-in-time stock-level residual data, executable historical quotes and 15-minute exits. Prior was uncertain sign; net stress failure and nonpositive matched selection now support rejecting this adaptation, not all reversal. Sources: https://www.nber.org/papers/w17653 ; https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr513.pdf ; https://www.aqr.com/Insights/Research/Working-Paper/Trading-Costs-of-Asset-Pricing-Anomalies ; https://www.ssga.com/us/en/intermediary/etfs/state-street-industrial-select-sector-spdr-etf-xli . Source observations24103–24107, vintage correction24137.

Primary data recovered from the IAI public webpage's own chart API, https://alvis.international-aluminium.org/api/publication/?publication=primary-aluminium-production , linked by https://international-aluminium.org/statistics/primary-aluminium-production/ . Latest period July2026. June total5939kt over30days=197.966667kt/day; July6160kt over31days=198.709677kt/day. Therefore monthly total+3.721165%, daily rate+0.375321%. GCC June276kt, July293kt gives daily+2.734923%; July2025GCC523kt means -43.977055% year-on-year. Premises:1 daily rate removes calendar length;2 geographical aggregates include other plants, not just EGA;3 IAI includes voluntary company reports and estimates for nonreporters/China, subject to revisions. Its primary definition excludes alloy additions and recycling, so cannot subtract EGA cast-metal tonnes without a conversion bridge. Conclusion: recovery is gradual in this vintage; there is no evidence of a 3.72% increase in daily capacity. Counterexample: producer inventories/export routes may improve deliverability even when global output barely changes. Joint indicator required: normalized production rate plus deliverable inventory plus issuer net cash. LME official stock-summary body did not expose current rows; secondary244525t claims were not treated as verified current deliverable inventory. This is a physical indicator, not a tested stock signal.

Start state: supply fund flat, physical Gulf production impaired, some competitors restarting. Goal: own paid scarcity rents at a favorable stock payoff. Binding constraint is net realized margin after energy, input metal, working capital and recovery, not the global metal headline. First order: interrupted primary production and export routes constrain deliverable metal. EGA's H1 sales -31.61% versus cast production -29.15%, alongside OCF -56.87% despite EBITDA +10.85%, shows operational and cash channels differ. Source https://media.ega.ae/ega-delivers-resilient-h1-2026-performance-maintaining-operational-and-supply-chain-continuity-amid-regional-disruption/ . Second order: CENX's US regional premium exposure can retain value, while LME-linked Iceland power/alumina and Sebree market power share it with suppliers. Fabricators KALU/CSTM can pass metal costs through, benefit from scrap spreads and scarce qualified rolling capacity, but fund higher inventories. Sources https://www.sec.gov/Archives/edgar/data/949157/000162828026013788/cenx-20251231.htm ; https://www.sec.gov/Archives/edgar/data/1563411/000156341126000057/cstm-20251231.htm ; https://investors.kaiseraluminum.com/news-events/press-releases/detail/440/kaiser-aluminum-corporation-reports-record-second-quarter-2026-financial-results . End customers face higher invoice prices or delays; allocation and pass-through timing are not measured here. Third order: scrap suppliers can renegotiate, customer substitution can lower demand, and restored primary/rolling capacity can compress premiums. Novelis reports Oswego restarted in June and Bay Minette commissioning, yet June-quarter OCF -455m and adjusted FCF -1.1bn; insurance timing gave an18m net EBITDA benefit from the outage. https://investors.novelis.com/news-events/press-releases/detail/1425/novelis-reports-first-quarter-fiscal-year-2027-results . These are conditional consequences, not probabilities. Strong counterargument to pessimism: CSTM ex-lag EBITDA rose87.88% and generated90m quarterlyFCF. Strong counterargument to bullishness: restarts and cash needs can reverse that rent. Conclusion: keep separate primary-metal and conversion-spread watches within supply; no unrelated FUND warranted.

Start49.14fundcash,850.99account,noholdings/pending. GoaldatedFDA/earningsrunupwithpre-eventexit. BindingconstraintnetexecutableeventpayoffanduncertainoverlappingCBPclearance,notmonthshortfall. Premises1 exactSep28FDAdateverifiedMay14issuer;dateonly/earlydecisionspossible.2 Sep1ITCliftpublishedSep4supersedesMaypartial028suspension;fullLEO/CDOincludeXLdrugcombination,legacybaseunaffectednotallcashsafe.3 Cashreconstruction2485preWCburn3.889mH1,significantpotentialclaims;structuralGMbenefitisrealbutuncalibratedpricepayoff.4 Frozenmomentum2477–2480train-96.87%N134/holdout-52.21%N74,stress-64.46%;halfKelly0. ThisnegativecontrolNOTFDAeventbacktest. Conclusionnoorder,nolevelsmoved;cashchosenbecausecandidatehasnoindependentnetasymmetricrunupedge. Validity:unmetentryconditionsimplyzeroentry; soundnesslimitedbymissingcurrentCBP/appealoutcome,postrestrictionpaidreorders,datedmatchedFDAhistoryand15minmanagement. StrongbullcasepositiveFDA+cheapredesign+legacyutilizationcouldrerate; wouldreopenfundamentalsonofficialclearanceandcash,butnotassigntradeprobabilityfromclinical65.5%. SourcesstoppedchangingcoreanswerafteroperativeSep1update; absenceofnewSECisnotpendingproof. Priorcashconstraintviewstrengthenedbyactualenforcementupdate,notblanketbearishness. Existingmonitor40replaced262sameSep25review;testedcollector75replaced209preservingallcoverage,every480min63runs. Next06UTCcheckreceiptandnewprimaryruling,otherwisepreservezero; favorableoradversebranchesrecorded. No separateFUNDcase: no validatedoutside-mandateoperation. Sources https://www.sec.gov/Archives/edgar/data/1858685/000149315226022875/ex99-1.htm ; https://public-inspection.federalregister.gov/2026-18101.pdf ; https://www.usitc.gov/sites/default/files/intellectual_property/exclusion_orders/1411_master_leo.pdf ; https://www.sec.gov/Archives/edgar/data/1858685/000149315226037834/form10-q.htm ; https://runway.adityaperswal.workers.dev/data/stock-bars .

Numbered premises: 1. September 10 completed SIP all-adjusted closes from https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all are AA48.30, CENX44.36, CSTM25.99, KALU158.42. 2. Each is below both its prior20 close high (51.85/48.17/29.92/189.49) and SMA50 (48.5086/45.6681/28.4026/166.195). 3. ADR20 is 1.972495/2.333405/0.97824/7.1416, and 20-day returns are -5.70/-7.91/-11.90/-16.40%. RV20 annualized is45.88/42.93/39.99/60.78%. These are adjusted daily measurements, not executable after-hours quotes or stop fills. Conclusion: no breakout entry tonight, even if operating margins are favorable. Counterexample: a falling stock can become undervalued, but this screen does not test a value/reversal entry and cannot authorize one. Stops and targets require an actual future entry plus independently justified payoff; no anticipatory limit is queued. Source observations contain every symbol's numbers.

Method: SIP all-adjusted daily bars, 1,681 per symbol from January 2020 through September 10, 2026; complete pagination, unique timestamps, positive volume and OHLC bounds checked. Source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all . Compare daily log returns in January 2025–March 27, 2026 (309 observations) with March 30–September 10 (114). Post-period cumulative returns: AA -17.01%, CENX -10.65%, CSTM +11.21%, KALU +39.53%. Post-period SPY betas: 1.444, 1.361, 1.969, 1.842 respectively. OLS daily intercepts -0.393%, -0.315%, -0.219%, +0.00005%, with naive standard errors 0.319%, 0.357%, 0.240%, 0.308%; none is compelling evidence of independent excess return. No HAC or causal identification claimed. January 2025 onward correlations AA/CENX 0.673 and CSTM/KALU 0.655 mean a two-stock aluminum sleeve is partly one bet. Premise: a known supply outage coexists with several other price drivers, changing tariffs and company restarts. Counterexample to automatic upstream outperformance: downstream names rose while the smelters fell. Conclusion: reject using outage direction as stock direction. Selection of surviving listed companies, one event, unequal windows, and overlapping industry/market shocks prevent estimating a repeatable scarcity premium from these figures.

Ran node /app/agent/dist/backtest.js --symbols BFRI --strategy momentum --lookback20 --days1460 --stop-pct5 --target-pct15 (actualshellflagswithspaces).1003IEXdailybars,124closedtrades. CLIrawpricesexhibitJuly2023splitdiscontinuity,notcomparablewith1176SIPadjustmentallstudy. Fullsamplediagnosticwithoutcostedholdout; notFDAeventstudyandnotadopted. Independentfixedrulelongonly/costedchronologicalholdout2477–2480alsofailed. Source https://runway.adityaperswal.workers.dev/data/stock-bars?feed=iex ; localcli.json.

1 Registry187randomizedadults,145active/42vehicle:primarycomposite65.5vs4.8%,difference60.7pp; approximateunadjustedWald95%50.62–70.78ppfromroundedrates. This isnotprespecifiedtrialinferenceorapprovalodds; populationexcludesaggressive/recurrent/highriskH-zonelesions. SeriousTEAEs3/145vs2/42notcausalcomparison.2 Tradingholdout74fixedmomentumtradesmean-.70697%/event,20000iidbootstrapseed91295%mean[-2.40017,+1.15124]%,total-52.2123%. Samplingdependenttradesignoresregime/timecorrelation; CIcrosszeroandnegativepointestimatedo notproveallFDAstrategiesfail.3 Sponsorregistryandpressreleaseoneclinicalorigin;ITCindependentlegalorigin;SIPmarkettransactionsseparateorigin. Noactive-comparatorclinicalclaim,noconversionofclinicaleffectintostockwinrate. Sources https://clinicaltrials.gov/api/v2/studies/NCT03573401 ; https://runway.adityaperswal.workers.dev/data/stock-bars .