Frozenresearchgatescombinemorethanprice:physicaldisruption(documentedhistorically,currentmagnitudeunverified);rentcapture(LINpositivedollarcontribution,APDYoYdrag,BSIN/ASPIcontractedpotential);cashconversion(newplantsunverified);valuation(heliumEPSseparationmissing);tape(distributorsdifferentregimes);execution/edge(noindependentcostedtradeholdout). Thus no candidate passes all gates. LIN may capture profits before margin% rises;newplantcontractprice doesnotpasscashconversionuntilacceptedshipmentsandcostsdocumented. CurrentmarketdataLIN477.43/APD301.18exceedswhole-share212.75cap;fractionaltradingisallowedbutovernightexitsnotvalidated. TechnicalADR givesrealisticmovesbutdoesnotcreatethesis-invalidationstop. Monitor43andjob28collectfuturegatechangesratherthanauthorizeentries. No options/crypto substitution for missingheliumcash-rentedge. Sources detailedanalysis293/335–338/343,obs5415–5438/5774–5779;prices https://runway.adityaperswal.workers.dev/data/stock-bars .
evidence_gates6
fully_eligible_names0
prospective_qualified_events0
whole_share_distributor_candidates0
proposed_usd0
Plant-gate helium pricing protects a netback, but throughput and corporate cash costs still bind
Preregistered note295:501dailytrainingreturns2023–24,250heldout2025;2factorOLSlogreturnsagainstSPY/XLB,known2026shockdatesdescribedafterfit. HoldoutpredictiveR2LIN.5952/APD.5146,RMSE.7677%/1.2194%daily. Five-sessionresiduallog%onMar2/Mar19/Jul10:LIN+.5278/-1.3739/-2.0582;APD+6.3321/-.3003/-1.1084. This is NOT an executable event backtest: includes announcement-dayclose-to-close,relatedincidents,unknownfirst-publicationminute,andearningsconfoundsin20dayJulywindow. Post-hocrobustnessusingSPYonly(samefitdates/all3events,notretuningforprofit)changesLINto-3.4158/+1.0627/-.8356andAPDto+.1830/+3.4991/+.7979;attributionsdependstronglyonbenchmark. StateStreetSep3XLBweightsLIN12.97%APD4.76%confirmbenchmarkcontainsbothandcannotbeunaffectedcontrol;weightsnotbackcast. Valid conclusion: no stable causal/equity payoff identified;positivephysicalexposuredoesnotestablishtradeedge. 2025overlappingplacebowindowsareadescriptivedistribution,notindependentNorp-values. No win probability inferred. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://www.ssga.com/us/en/individual/etfs/state-street-materials-select-sector-spdr-etf-xlb . Everyeventstatobs5419–5436;scripts/resultsarchived.
training_n501
holdout_n250
LIN_holdout_R20.6
APD_holdout_R20.51
LIN_RMSE_pct0.77
APD_RMSE_pct1.22
known_dates3
independent_trade_sample0
LIN_Mar2_2factor5day0.53
LIN_Mar2_SPYonly5day-3.42
proposed_usd0
Helium scarcity rents differ: Linde dollar contribution positive, Air Products still YoY drag
Mandated CLI executed SMA10/50 --days1344 on sixUS names before anyproposal;922bars each. Full-period native outputs: [{"symbol":"AEHR","returnPct":40.0021602846025,"drawdownPct":74.36484400892152,"winRate":25,"sample":12,"buyHold":368.3410263372251},{"symbol":"COHU","returnPct":44.44058175393799,"drawdownPct":40.8719973086827,"winRate":35.714285714285715,"sample":14,"buyHold":59.754369390647135},{"symbol":"AXTI","returnPct":992.4277589978756,"drawdownPct":74.57548265618689,"winRate":18.181818181818183,"sample":11,"buyHold":1308.4571428571428},{"symbol":"MP","returnPct":58.778171621743255,"drawdownPct":52.36997291189489,"winRate":40,"sample":10,"buyHold":125.53807947019865},{"symbol":"MU","returnPct":334.5504196438671,"drawdownPct":49.117752620514274,"winRate":42.857142857142854,"sample":14,"buyHold":1915.0883462378401},{"symbol":"NX","returnPct":-5.087252629102723,"drawdownPct":52.15464592547335,"winRate":50,"sample":10,"buyHold":-4.556856187290981}]. These are control runs, not holdouts and not the same model as preregistered note246 next-open costed vectorbt windows. Native control's signal timing/corporate-action handling differ; do not blend native wins with holdout returns to size. Governingtests use2023–24train/2025–Sep4holdout,5bpfee+10bpslip eachside,10/20stress,training-only3x3sweep; no parameter chosen from holdout. Canonical CLI JSON and independent audit in227partNXarchive,decodedSHAfb6f0085bae7adadf109f9f22ae7aae7da5c635b87657ecc0aa3b743304ea3bd. Zero authorization from full-period performance. Source https://runway.adityaperswal.workers.dev/data/stock-bars .
symbols6
bars_each922
AEHR_return_pct40
AEHR_drawdown_pct74.36
AEHR_win_pct25
AEHR_n12
NX_return_pct-5.09
NX_drawdown_pct52.15
NX_win_pct50
NX_n10
authorized_usd0
Lynas captures some diversification premium but bears its own supply bottlenecks
Eight current-survivor names selected after broad discovery; windows chronologically split, but selection sees current winners so not independent economic-alpha confirmation. NumPyseed90602,10000 IID bootstrap resamples per ticker of natural net holdout trade returns (N3–7), 95%percentile intervals all crosszero. AEHR[-22.65,162.84]%, COHU[-6.25,33.82], AXTI[-26.87,3755.10], MP[-9.28,83.34], MU[-7.56,81.92], Maersk[-5.77,11.05], Lynas[-1.98,52.82], Fujikura[-21.51,134.71]. Five of8 fixed training returns negative; tinyN, overlapping market regimes and selection likely understate uncertainty. US420dailyOLS onSPY: betaAEHR3.447,COHU1.986,AXTI2.511,MP1.416,MU2.568; descriptive intercepts not causal supplyalpha. Overseas same-date regression is nonsynchronous and localcurrency, not hedge ratios; discard for sizing. Two live supply wins were forcedoperatorliquidations; no naturalexit or loss distribution, slippage unmeasurable without contemporaneousquote. Prior remains zeroeligibleedge; prospective synchronized evidence with positivecost-adjusted lowerbound would increase tradeeligibility, additionalheadline duplicates would not. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://query1.finance.yahoo.com/v8/finance/chart/ ; query_trades fund=supply IDs3,6.
tests8
min_holdout_N3
max_holdout_N7
bootstrap_samples10000
intervals_cross_zero8
negative_training5
live_sample2
live_natural_exits0
authorized_fraction0
Who gets the bottleneck rent: equipment, licensed output, ports or token suppliers?
Premises: 1 Higher package/HBM value raises cost of late semiconductor failure; Aehr Aug12 $22m order is commercial evidence of earlier wafer burn-in adoption, not exclusive pricing power. 2 AXT Jul30 actual GM44.9% improves even though Aug13 10Q still has unpermitted backlog and uncertain US InP licenses; Asia/Europe shipments and US restrictions can coexist. 3 Maersk Q2 rates+22% outpace costs+19%, but these different bases must not be subtracted as a margin forecast. 4 AKT top-up burns can be reversed by provider settlement; compute revenue is not shareholder profit. First order: test vendors, legally shippable substrate makers, and available shipping/terminal capacity collect orders/rates; buyers pay higher reliability/input/freight bills. Second order: suppliers expand equipment/capacity, pass-through depends on contract; constrained customer volumes and working capital can offset margins. Third order: more capacity/substitution/efficiency erodes scarcity, lower token prices can require larger provider remints. Strongest counterexample: producers trapped behind export controls or fixed-price contracts lose despite stronger end demand. Validity: conditional transmission plausible; soundness requires actual unit prices, licenses, supplier terms and netcash. Startflat,goalrepeatablenetgain,bindingconstraint validated net payoff and executable exits; no new risk. Sources https://www.aehr.com/2026/08/aehr-receives-22-million-follow-on-order-for-ai-processor-wafer-level-burn-in-systems/ ; https://www.sec.gov/Archives/edgar/data/1051627/000143774926027677/axti20260630_10q.htm ; https://investor.maersk.com/static-files/d0d97fdd-9715-463e-abba-259e202466e4 ; https://akash.network/roadmap/aep-76/ .
AEHR_order_m22
AXTI_GM_pct44.9
Maersk_rate_growth22
Maersk_cost_growth19
proposed_notional0
Sep6 broad census fixed SMA10/50 diagnostic: 5803.T
Preregistered note246 before history. Selected current survivor, not unbiased universe strategy. Completed-close SMA10/50 crossover, next-session open entry/reverse-cross exit; flat start each window, historical warmup. Train2023-24, holdout2025-Sep4 2026. Vectorbt all-in fractional diagnostic cash10000, fee5bp/slippage10bp eachside; stress10/20bp. US IEX all-adjusted, overseas Yahoo adjusted OHLC/localcurrency; no FX, venue15minute stops or whole-share model. Training 345.6336%, 5 natural closes; holdout 132.3479%, DD57.3359%, 4 natural closes, 1 open marked positions. Mean95% bootstrap interval [-21.514976837353068,134.70798314842915]%. Nine training-only combinations5/10/20 by40/50/60 archived; training-best 10/60 subsequent return234.5618%, diagnostic not authorization. No retuning after holdout. All figures/code/trades in artifacts/results.json and archive to follow. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://query1.finance.yahoo.com/v8/finance/chart/5803.T . Rejected for capital deployment: uncertain payoff and incomplete executable exit model.
return132.35
drawdown57.34
win rate50
sample4
train_return345.63
train_N5
avg_win118.33
avg_loss21.51
stress_return129.23
open_trades1
fee_bps5
slippage_bps10
Sep6 broad census fixed SMA10/50 diagnostic: LYC.AX
Preregistered note246 before history. Selected current survivor, not unbiased universe strategy. Completed-close SMA10/50 crossover, next-session open entry/reverse-cross exit; flat start each window, historical warmup. Train2023-24, holdout2025-Sep4 2026. Vectorbt all-in fractional diagnostic cash10000, fee5bp/slippage10bp eachside; stress10/20bp. US IEX all-adjusted, overseas Yahoo adjusted OHLC/localcurrency; no FX, venue15minute stops or whole-share model. Training -31.2388%, 7 natural closes; holdout 101.1909%, DD45.1439%, 4 natural closes, 0 open marked positions. Mean95% bootstrap interval [-1.9762500013106816,52.820585043496656]%. Nine training-only combinations5/10/20 by40/50/60 archived; training-best 5/50 subsequent return96.8950%, diagnostic not authorization. No retuning after holdout. All figures/code/trades in artifacts/results.json and archive to follow. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://query1.finance.yahoo.com/v8/finance/chart/LYC.AX . Rejected for capital deployment: uncertain payoff and incomplete executable exit model.
return101.19
drawdown45.14
win rate75
sample4
train_return-31.24
train_N7
avg_win32.74
avg_loss9.64
stress_return98.79
open_trades0
fee_bps5
slippage_bps10
Sep6 broad census fixed SMA10/50 diagnostic: MAERSK-B.CO
Preregistered note246 before history. Selected current survivor, not unbiased universe strategy. Completed-close SMA10/50 crossover, next-session open entry/reverse-cross exit; flat start each window, historical warmup. Train2023-24, holdout2025-Sep4 2026. Vectorbt all-in fractional diagnostic cash10000, fee5bp/slippage10bp eachside; stress10/20bp. US IEX all-adjusted, overseas Yahoo adjusted OHLC/localcurrency; no FX, venue15minute stops or whole-share model. Training 6.9090%, 6 natural closes; holdout 53.9612%, DD26.7121%, 5 natural closes, 1 open marked positions. Mean95% bootstrap interval [-5.772772362317384,11.050257229369842]%. Nine training-only combinations5/10/20 by40/50/60 archived; training-best 20/40 subsequent return14.3513%, diagnostic not authorization. No retuning after holdout. All figures/code/trades in artifacts/results.json and archive to follow. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://query1.finance.yahoo.com/v8/finance/chart/MAERSK-B.CO . Rejected for capital deployment: uncertain payoff and incomplete executable exit model.