Question: what historical range can occur between this Sunday08UTC run and Monday00UTC next review? Prior: quiet weekends are not risk-free. Method: retrieve primary hourly bars Jan2024–Sep13 07UTC, checkpoint each page; select complete Sunday08–23UTC windows. Exclude today's incomplete window. Separate2024–25 calibration BTC/ETH N104,SOL70 (history beginsAug26 2024) and already-known2026 N36 each. No parameter sweep or strategy selected. Known2026 gross mean returns BTC−.12176%,ETH−.20451%,SOL−.30521%; positive47.22/47.22/44.44%. Endpoint P05/P95 BTC−2.0887/+2.1124%,ETH−4.5489/+3.6673%,SOL−4.4856/+3.5792%. Intrawindow low P05−3.4902/−5.7637/−6.5846%; highsP95+3.5221/+4.4851/+5.4965%. Training worst lows−7.0628/−14.4095/−12.9352%. These are realized sample excursions, not future bounds. Matched2026 correlations BTC/ETH.89293,BTC/SOL.87954,ETH/SOL.91175. All6fullrows retained in observations and localJSON. A weakly negative unconditional mean cannot reject a validated conditional momentum edge; a positive median cannot establish one. Conclusion: use shared-shock risk, preserve no-entry decision for independent signal/policy reasons. Source https://runway.adityaperswal.workers.dev/data/crypto-bars . Missing historical executablequotes and sparseSOLtraining remain limitations.
sample36
BTC_train_N104
ETH_train_N104
SOL_train_N70
BTC_mean_pct-0.12
ETH_mean_pct-0.2
SOL_mean_pct-0.31
BTC_ETH_corr0.89
BTC_SOL_corr0.88
ETH_SOL_corr0.91
horizon_hours16
size0
C91306 weekend volatility falls, but cross-coin correlation rises in the holdout
Calendar-hour logclose returns from venuebars, train2024–25 and knownholdout2026Jan1–Sep12. Do not bridge missinghours. Holdoutweekend hourlySD BTC33.0404/ETH48.8851/SOL58.7092bp versusweekday51.2783/66.7165/70.1682bp;N1752weekend,4366weekdayeach. Weekendmeans−.4173/−.6670/−.5293bp perhour are descriptive, not causal/weekendshortsignal. Worstweekend simplehours−3.5913/−4.0245/−5.0440%,so quietmedian does not boundlefttail. Synchronized6118hour holdout correlations BTCETH.887966/BTCSOL.837870/ETHSOL.848997, versus matched11801traininghours .792899/.750043/.770080. TrainingSOLcoverage partial, so composition/time differ; no formal structuralbreak claimed. Premises: 24/7risk includes weekends; correlatedcoins are one sharedriskfactor; only drawdownstress acrossjointpaths can evaluate diversification. Strongestcounterargument: correlations below1 and tail timing can differ; retain synchronized simulation ratherthan assumeperfectcorrelation or independence. Conclusion threequarterprobes must countjoint exposure, not three independent convictions. All14distribution/correlation rows recorded with source/date/units. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .
holdout_N6118
BTC_ETH_corr0.89
BTC_SOL_corr0.84
ETH_SOL_corr0.85
BTC_weekend_sd_bp33.04
ETH_weekend_sd_bp48.89
SOL_weekend_sd_bp58.71
BTC_weekend_worst-3.59
ETH_weekend_worst-4.02
SOL_weekend_worst-5.04
C91306 fund trade audit: one operator liquidation cannot calibrate payoff
Re-read query_trades fund=crypto at06:02UTC. Only trade10: BTC entry81614.8 exit81548.7 qty.001681857, opened2026-09-03T19:42:24.677Z closed20:30:20.345Z by operator liquidation. Price return−.0809902%, price PnL−.111171USD, wins0/1; stop73440,target97900 were not triggered. No contemporaneous entry reference quote or itemized fees, so slippage unavailable. Premises: externally censored N1 is not a strategy payoff sample; old breakout/media reason does not validate today's delayed-entry control. Conclusion no empirical live Kelly from this record. Current no own open/pending positions. Source query_trades fund=crypto and supplied Sep13 venue account context.
sample1
win rate0
raw_return-0.08
price_pnl-0.11
open0
pending0
C91304 live trade record cannot estimate the stablecoin rule's edge
query_trades fund=crypto returns one closed BTC lot: entry81614.80, exit81548.70, quantity.001681857, openedSep3 19:42:24.677UTC, closed20:30:20.345UTC by operator liquidation. Raw price return−.0809902%, price-PnL−.111171USD, observed0/1wins; neither stop nor target determined exit. No entry quote/reference price or separate fee ledger supports a slippage estimate. Its old price-breakout thesis was not tonight's supply rule. Premises: one externally terminated trade is not a strategy distribution; do not estimate win/loss odds or half-Kelly from this lot. Current own book has no open or pending crypto orders; EEM belongs to trend and is outside this review. Source query_trades fund=crypto; user venue context asSep13 04:01UTC. Conclusion use recorded research controls, not0/1pseudo-confidence; actualallocation0.
closed_N1
wins0
raw_return_pct-0.08
price_PnL_USD-0.11
own_open0
own_pending0
C91304 one-day forecast check also fails to establish incremental supply information
Frozen question: does seven-day USDC unit growth, available after two calendar days, predict the next seven-day log return after controlling for preceding seven-day price momentum? OLS with intercept and Newey-West/HAC7 covariance; training2024Jan1–2025Dec24 (purged final7days), holdout2026Jan1–Sep5 with outcomes throughSep12. Current-vintage Llama aggregate history, publication vintages unknown; no causal identification. BTC/ETH trainN724, SOLN479 due missing history; each holdoutN248. USDC training betas BTC-.000083, ETH-.229591, SOL-.504461, all95%CIs include0. Holdout betas .938418/1.453205/1.425510, CIs[-.322597,2.199434]/[-.563767,3.470178]/[-.388353,3.239373], p .14469/.15791/.12348. CombinedUSDC+USDT holdout betas2.69896/3.34271/3.51804 also cross0. Lagged future correlations~.25–.28 look encouraging in2026, but training future correlations nearzero/negative; contemporaneous trailing-price correlations generally larger. Counterexample: SOL2026USDC future corr .27246 exceeds past .24848, so it is false to claim supply always follows price. Validity: the models do not establish incremental stable predictive content; they do NOT prove no effect or direction of causality. Priors only weakly updated toward context rather than standalone signal. Need prospective publication-stamped samples and repeatable net holdout improvement to upgrade. Sources https://stablecoins.llama.fi/stablecoin/2 ; https://stablecoins.llama.fi/stablecoin/1 ; https://runway.adityaperswal.workers.dev/data/crypto-bars .
BTC_train_n724
SOL_train_n479
holdout_n248
BTC_hold_beta0.94
ETH_hold_beta1.45
SOL_hold_beta1.43
BTC_p0.14
ETH_p0.16
SOL_p0.12
specifications12
C91302 dependence and uncertain Kelly estimates across seven volatile majors
Premise1:100OKX8hour settlements Aug11–Sep13 00UTC; latestBTC.497035bp percentile35.35,SOL.276108bp55.56. ETH1bp equals priorP95, with23of99prior also1bp; strictpercentile76.77 versusweak100. Premise2: relative Sep12 08UTC, currentcoinOI BTC27484.289 +.120866%,ETH639751.978 +.919325%,SOL2741317.07 -1.54566%; quantities remove direct price valuation but do not reveal long/shortdirection. Premise3: periodfunding sums61.9642/62.3439/22.0552bp; positivecarry is a cost toswaplongs, not earned by cashspot. Strong bearcase forETH: higherOI pluspositivefunding could mean leverage crowding that amplifies downturns. Counterexample: arbitragehedges can generate sameobservations; tied1bp is common, no globalextreme or squeeze probability. Conclusion reject funding-onlyentry; report tieinterval and preserve actualprice trigger. Sources https://www.okx.com/api/v5/public/funding-rate-history ; https://www.okx.com/api/v5/public/open-interest . Raw observations26531–35.
N100
BTC_percentile35.35
ETH_strict76.77
ETH_weak100
SOL_percentile55.56
ETH_OI639751.98
BTC_sum_bp61.96
ETH_sum_bp62.34
SOL_sum_bp22.06
C91208 ETF boundary control preserves weak flow-return relation
DescriptiveOLS next-sessionNAVreturn on contemporaneous fractionalshareschange from primary issuerHistorical workbook; no execution rule/backtest. 2024–25trainingN497 corr-.026228,beta-.001625,positive-flow nextmean+.192300% versus nonpositive+.186261%. Known2026N173corr+.038953,beta+.328987,positiveflow nextmean-.025226% versus nonpositive-.065636%. Two missing historical shares createNA,notzeros. Premise1 small signed correlations switch sign; premise2 in-sample slope change is not causal evidence; premise3 actual publication timestamp unavailable and finaltraining observation's forwardreturn crossesJan1 byone session, so split not pristine. Counterexample data may contain nonlinear longer horizon information; this limited diagnostic cannot reject everyflowstrategy. Conclusion do not act on aggregateIBITflowalone or select threshold fromknown2026. Source https://www.ishares.com/us/products/333011/ishares-bitcoin-trust-etf DataDownload; fullURL in observations.
trainN497
holdoutN173
traincorr-0.03
holdoutcorr0.04
train_positive_mean0.19
train_nonpositive_mean0.19
holdout_positive_mean-0.03
holdout_nonpositive_mean-0.07
C91208 hourly weekend risk and dependence: lower typical volatility is not a stop guarantee
Premise1 direct OKX history100 realized8hour settlements Aug10 08UTC–Sep12 08UTC, latest ranked only against99 prior. BTC latest.855375bp/median.646500/P95 1.0=>69.69697percentile; ETH.463076/.679862/1.0=>36.363636;SOL.871952/.121511/1.0=>75.757576. Period sums62.640955/62.307298/23.329550bp (0.626410/0.623073/0.233296% of constant notional, not leveraged equity return). Counterexample latest SOL is above its median, but not aboveP95; positive is not extreme. History is one venue and one recent month; caps/ties at1bp affect percentile; no pooled-exchange inference. Premise2 contemporaneous coinOI27451.11BTC/633924.129ETH/2784353.63SOL is quantity rather than dollar-price effect; no long historical OI percentile available. Conclusion no funding-reversal signal adopted; collecting prospective readings continues. Sources https://www.okx.com/api/v5/public/funding-rate-history ; https://www.okx.com/api/v5/public/open-interest .
N100
BTCpercentile69.7
ETHpercentile36.36
SOLpercentile75.76
BTCperiod_bp62.64
ETHperiod_bp62.31
SOLperiod_bp23.33
C91206 major-coin dependence strengthens in 2026; regression is not alpha
Premise 1: pandas aligned2159 hourly log returns over90completed UTCdays; BTC/ETH correlation.8680598,BTC/SOL.7972226,ETH/SOL.8025502. Annualized volatility38.64/52.07/57.36%; hourly1% tails−.979/−1.473/−1.560%. Premise 2: OKX100 most recent8-hour funding settlements Aug10–Sep12 put latest BTC.324729bp at18thpercentile,ETH.274278bp at17th,SOL.605550bp at67th. All inside respective5–95% ranges. Premise 3: current coin OI is27112.401/622879.194/2717751.07; endpoint change versus11:45 is not signed flow or demonstrated extreme. Conclusion: three majors are correlated exposure; no funding-extreme trade. Counterexamples: correlations can jump in crashes; one-exchange funding does not represent all venues. Sampling excludes other venues and treats returns rather than causal drivers. No regression causal claim or price forecast. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://www.okx.com/api/v5/public/funding-rate-history ; https://www.okx.com/api/v5/public/open-interest .
Premise1 fully paginated Aug19 08UTC-Sep11 07UTC hourly OHLC,552 bars each/no missing hours,551 aligned logreturns. BTC/ETH correlation .844566,BTC/SOL .732571,ETH/SOL .726796; hourly SD .480154/.650954/.740780%. Premise2 latest15-note returned sample has quote6 receipts maxgap30.0107minutes, entry5 maxgap29.9372minutes on15min cadence; sample is not exhaustive and does not prove stop outages. Premise3 one closed crypto trade10 operator liquidation after47m56s entry81614.8 exit81548.7 cannot estimate strategy exit odds or order-type slippage. Conclusion BTC/ETH positions would be one correlated risk; do not size independently. Fresh complete bar window supports current signal measurement, not historical execution. Prior fear-of-missing-tail remains separate from the correlation fact. Source https://runway.adityaperswal.workers.dev/data/crypto-bars ; query_trades and query_research fund=crypto.
joint_N551
BTC_ETH0.84
BTC_SOL0.73
ETH_SOL0.73
quote_maxgap_min30.01
entry_maxgap_min29.94
live_trade_N1
missing_hours0
C91106 measured hourly tails and limitations of the live track record
Question: is ETH daily/hourly performance disagreement real timing or inconsistent bars? Prior favored possible measurement artifact. 1 FullypaginatedBTC/ETH984daily23612hourly,SOL746/17881 throughSep11 00UTC; 981/981/742 full24h days. ETH2026 252matched days have0OHLCdiscrepancies; raw2024/25 smallclose mismatches common and2025Nov-Dec open/high/low anomalies larger. RequeryDec2 ETH confirms nativeopen3020.1375 vs00UTC hourly2798.84 (+7.906758%); cause unresolved, source construction documentation does not prove cause. 2 Identical2026 priceinputs give identicalmatchedpolicyresults. FullnativeETHdaily/daily-17.632466%; merelygatingonfullhourlydays excludesMar17->22 -11.994780%trade afterMar12missinghour, liftingresultto-6.406080%:11.226386pp selectioneffect. 3 Withfullnativeinputs,dailyentry/hourlyexit-16.866373%;hourlycalendarentry/exit7.324178%;rolling480hour17.240439%. This ladder changes entry AND channel-exit decision cadence, not only earlier entry; rollingwindowchanges level definition. Daily->hourlyriskcheck improvement.766093pp, decisioncadence24.190551pp, rollingdefinition9.916260pp. Validity: within frozen conditional experiment and statedcoveragepolicies, constructions cannot explain identical2026prices. Soundness: historicalbar revisions, no executablehistoricalquotes, tinyreused2026samples mean timing advantage is observed not proven enduring. Strongestcounterexample is omittedMarchloss: a dataqualityfilter can manufacture performance; retainfull-nativecontrol. Conclusion stop using distinct dailystrategy failure as standalone veto of hourlyedge; freezehourlywatch while requiring its actualsignal and executablepolicy. No newbacktestparameter selected. Sources https://runway.adityaperswal.workers.dev/data/crypto-bars ; https://docs.alpaca.markets/us/docs/real-time-crypto-pricing-data ; notes2175/2180 and backtests2523-25.
ETH_matched_days_2026252
ETH_OHLC_mismatches_20260
ETH_daily_full-17.63
ETH_daily_matched-6.41
coverage_selection_pp11.23
ETH_daily_hourly-16.87
ETH_calendar_hourly7.32
ETH_rolling_hourly17.24
daily_hourly_check_pp0.77
decision_cadence_pp24.19
rolling_definition_pp9.92
ETH_Dec2_open_difference_pct7.91
C91102 matched-close regressions distinguish coin exposure from corporate and oil risk
Premises:1 Observ23064complete24hourdays show large historical O/H/L/V discrepancies despite smallclose differences. 2 Additionalcontrol masks supplied daily on exactlysame complete-day calendar as reconstructed data and recomputes indicators, rather than cherry-picking prices. 3 Fixed20daystrategy matched versus reconstructed returns identical to floatingpoint:BTCtrain+.569980%,holdout+4.831708%;ETHtrain+31.615468%,holdout-6.406080%;SOLtrain+21.135460%,holdout-7.586258%. Conclusion the full-versus-reconstructed sign/magnitude change here is explained by coverage/warmupmask, not demonstrated OHLCcorrection alpha. Full suppliedBTC-5.685453% versus matched+4.831708% highlights missing-data fragility, not reason to choose winner. BTCmatchedtrain stress-2.791931% and holdoutN5 small. Validity of exact control supports attribution for this experiment only; other rules can remain price-sensitive. No gap filling or missingbar execution allowed. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .
BTC_full_daily_holdout-5.69
BTC_matched_holdout4.83
BTC_matched_train0.57
BTC_matched_train_stress-2.79
ETH_matched_holdout-6.41
SOL_matched_holdout-7.59
matched_daily_holdout_bars252
full_daily_holdout_bars253
C91100 funding history rejects a current extreme narrative
Premises:1 Collected100 consecutive8h OKX USDT-swap settled rates per asset, Aug9 00UTC–Sep11 00UTC, observations23045–47. 2 Latest settled percentile BTC70/ETH18/SOL34; 5th/95th bounds BTC.005875/1, ETH.129385/1,SOL-.727958/1bp. 3 Current indicative rates are not final settlements and one venue is not global leverage. Conclusion none is an outer5% extreme in this sample; no contrarian funding entry. Valid inference descriptive ranking, not prediction. Counterexample longer regime or other venue can be extreme; price effect requires time-aligned costed predictive test. Mean rates .620243/.619086/.248752bp per8h;100 normalized payments total .620243/.619086/.248752% of constant notional over33days inclusive timestamps, before hedging/costs; Alpaca cash spot cannot collect short-perpetual carry here. Non-independent adjacent observations, changing OI and no exchange-flow attribution limit inference. Source https://www.okx.com/api/v5/public/funding-rate-history .
Premises: 1. Re-read2183–2185 costed vectorbt tests:2025train, REUSED2026Jan1-Sep9holdout,25bpfees/side10bpslippage20stress; all36hourlycellslose and supplied dailynegative. 2. Python recomputation22952 pW-(1-p)L and .5*(p-(1-p)/(W/L)) produces negative expectancy and allocation for all3. 3. Sole actualBTC10 operator liquidation does not calibrate setup odds or order slippage. Conclusion zero allocation to recorded compression rule. Validity: positive-edge requirement fails; soundness limited by smallN, reusedholdout, unresolved historical aggregation errors. Counterexample: an independent different rule may profit, so rejection is not a bearish price forecast. Current>=3Rtarget/+2Ratmostquarter exits differ from frozen1.5R rule and need independentdaily/hourly plus15minexecution evidence. No new experiment, simulation, refit or CLI rerun this tradingrun; recorded CLI2186 grossdiagnostic has no fees or holdout and cannot rescue rule. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .
Premises:1. Re-read costed vectorbt backtests2183–85: 2025training, REUSED2026Jan1-Sep9holdout,25bpfees/side10bpslippage20stress; all36hourlycells lose, supplied daily also negative. 2. Python audit observation22912 recomputes pW-(1-p)L and .5*(p-(1-p)/(W/L)); all allocations negative before clamp. 3. Sole liveBTC10 forced liquidation cannot calibrate strategy or slippage. Conclusion allocation0; this is arithmetic audit, not new experiment/backtest. Validity conditional on positiveedge gate; soundness limited by smallN, reusedwindow and historicalaggregation defects. Counterexample independent newstrategy could profit; a rally cannot validate rejected rule. Existing1.5Rtest and +1Rreferences fail current>=3R/+2Rquarter policy: newversion needs independent daily/hourly15minexecution evidence, no retuning past holdout. Source https://runway.adityaperswal.workers.dev/data/crypto-bars .