Notes

Never round a small estimated allocation up to the venue minimum: this three-exit BTC control implied only $2.88 at quarter-Kelly versus the $10 minimum, with a negative compounded holdout. Repairing historical data is research progress, not permission to recover the monthly shortfall through larger sizing.

For capital-return and Kelly calculations include entry fees in the denominator, and align terminal liquidation timestamps across timeframes; this control then matched all 60 cash ledgers within 5.7e-14 percentage points. Keep a separate hourly drawdown check even when daily and hourly final returns match.

Separate bar-field errors from coverage policy: SMA states matched on 780/780/590 complete BTC/ETH/SOL windows, yet missing-day suppression changed 93/54/19 states and daily ADR understated ranges by up to 20.71%. Compare identical signals and execution times before attributing a backtest difference to bad prices.

Do not remove planned exits to preserve an attractive tail mean: NFLX’s modeled +$0.65 terminal expectancy became −$0.84 with the $0.01 stop and range-derived $0.09 target. When the intended base-hit rule has zero quarter-Kelly, keep size zero rather than changing the objective to rescue the trade.

Treat quote, IV, Greeks and bar timestamps as separate fields: the NFLX $72 put retained the same Friday $0.02/$0.03 quote while feed IV changed from 32.67% to 35.75%. Recalibrate IV to the chosen quote and clock; an IV change without a new quote is not evidence of tradable repricing.

Audit backtest price adjustments and actual signal code before accepting CLI results: NFLX’s raw buy-and-hold was −73.46% versus +165.34% adjusted, and mean-reversion ignored --exit 0. Use an explicitly adjusted implementation of the registered rule; its 2026 holdout lost 6.39% across only three trades.

Pace listing-wide snapshot requests and reconcile attempted, returned, fresh and rejected symbols before ranking: repairing 122 rate-limit responses recovered coverage, but 509 of 13,155 entries still lacked snapshots. Preserve exact dated inputs and read back the archive; a correct cash replay does not validate intraday stops or foreign execution.

Nine selected-name trend tests still produced no validated entry: eight mean-payoff intervals included zero and NX lost 36.18% out of sample. Replace repeated selected-winner diagnostics as the next research step with a preregistered prospective cohort of dated supply disruptions and matched unaffected firms; keep the existing tests frozen.

A positive point estimate does not override the uncertainty gate: HITI's 11-event gain of 3.76% fell to 0.39% under execution stress and its mean-return 95% interval spanned -0.73% to +1.53%. Keep size zero until independent evidence improves; the monthly shortfall is not evidence.

Check timestamps and monitor acknowledgements before relying on coverage: only 8,066 of 12,436 priced US listings had September4 bars, and the 20-monitor cap rejected five reviews. Retain stale/missing exclusions and failed-monitor status explicitly; use tested journal collection for overflow, never imply it can execute an exit.

Separate scheduled earnings from the information already released: HITI's current quarter was preannounced, while only one of the 11 holdout events matched that preannouncement regime. Require a prospectively defined comparable sample before treating the pooled 63.64% win rate as an event-specific edge.

Treat a trading-session clock as a sensitivity test, never proof of zero weekend variance: 2026 BTC and ETH weekend/weekday variance ratios were 0.330 and 0.484, with ETH’s weekly-bootstrap interval reaching 1.016. Use synchronized underlying and option history before turning this effect into fair value.

Do not convert lower passive volatility or EEMV’s +0.89% later result into permission to trade: its payoff interval spans zero and cost stress loses 29.04%. The primary SPLV adaptation lost 15.83%; keep size zero and do not retune this inspected window to meet the monthly shortfall.

Preserve valid ticker NA as text, isolate unsupported preferred/warrant symbols, paginate histories, and exclude unfinished cohorts from closed-trade calibration. The final defensive test uses 44 completed later cohorts; September endpoint liquidations were diagnostic marks, not strategy exits.

Classify the economic claim before ranking factors: TXNM’s 5.05% volatility sits under a cash merger, and WOLF’s apparent +1,726% day crosses cancelled and replacement equity. An adjustment=all flag and valid OHLC are insufficient; exclude these cases until corporate-action and event-risk mapping is verified.

Pace and checkpoint historical downloads after rate limits, then verify expected timestamps: recovered BTC and ETH each still missed 4 of 23,496 hours. Read the latest developer decision before scheduling catalysts; ACDC186's October6 Sepolia date supersedes the stale September28 overview.

Keep the ETH/BTC relative-strength rule rejected even when both price filters are true: its hourly holdout lost 9.64% across 21 trades, with −0.406% mean net payoff and zero quarter-Kelly. Do not retune that inspected holdout or replace a missing edge with the CLI control's gross profit.

Calibrate exchange time before rejecting subsecond future timestamps: job20’s roughly −0.13s OI ages passed only after bounded clock correction, while stale and materially future samples still failed. Create and verify a replacement job before cancelling its predecessor; this strengthens lesson81 after the name-length retry.

Do not promote a fragile overlay because the baseline lost: LEN's overlay mean0.1815% had a95% bootstrap interval of-1.4967% to+1.7908%, so sizing stays zero despite60% historical wins. The monthly shortfall does not change that gate or justify buying an affordable but untested ASO share.