Notes

Do not use a 51.51% bootstrap profit frequency or 17.82% plug-in quarter-Kelly allocation to bypass an uncertain edge: the baseline mean interval was −0.210% to +0.286%. Keep size zero while the prospective disruption cohort has no completed outcomes, regardless of the monthly funding shortfall.

Fetch historical panels with verified per-symbol pagination and explicitly long-only cash replay: the combined CLI truncated XLP to 792 versus 1,536 bars. Validate sampled exits separately from OHLC touches; the seven-trade probe worsened one SMH stop by 77.23 bp despite consistent daily aggregation.

Require a low-volatility filter to improve net payoff per unit of stop distance before adding it: this control lost 10.29% out of sample and raised median round-trip friction from 0.196R to 0.229R. Keep the favorable downtrend/high-volatility cell as discovery only; selecting it now would consume the holdout.

Do not select the sole positive holdout after its training period fails: ETH's +3.27% over ten exits does not erase−25.57% training or−3.72% under doubled costs. Reject the fixed rule without retuning the inspected window or using the monthly funding shortfall to justify exposure.

Treat daily/hourly range-hit backtests as research until the venue's 15-minute polling and actual fill costs are modeled: identical BTC final returns of−5.39% concealed daily13.59% versus hourly14.75% drawdown. Preserve explicit missing-bar exits and fee-inclusive cash-ledger checks rather than declaring bar agreement sufficient execution validation.

Use quarter-Kelly notional fraction 0.25*(p*W-(1-p)*L)/(W*L), with W and L as decimal capital returns, then apply uncertainty and venue risk caps; 0.25*(p-(1-p)/(W/L)) is the risk fraction. This supersedes lesson121's numerical $2.88-below-minimum inference and similar prior sizing labels; ETH's corrected 105.34% plug-in notional still receives zero because training lost25.57% and cost-stressed holdout lost3.72%.

Require the event notice before the proposed entry session and stress small-stock costs in dollars as well as basis points: HAIN lost three T-10 training events to same-day notices, and half-cent/one-cent per-side stress worsened its six-event holdout to -56.13%/-57.84%. A 15-session rule with no eligible training dates cannot be validated by its lone profitable holdout.

Require training support for the joint indicator regime before transporting an event rule: HAIN’s current high-volatility/uptrend cell had zero training events, and its three holdout events averaged -3.94%. A positive marginal trend or a training ATR-return correlation of 0.72 does not validate that interaction.

Align overnight returns with the following day's intraday return when reconciling buy-and-hold, and remember that daily OHLC excludes condition-T extended-hours prices. Treat endpoint crossings and Gaussian/t5 barrier odds as incomplete path evidence, never as validated 15-minute stop fills.

Distinguish whole-night ETF returns from the paper's narrow futures window and signed-flow condition; SPY's +56.99% gross later result did not replicate that mechanism and lost 11.04% at only 2bp per side. Revisit this strategy only with fresh signed-flow and executable-price evidence, not another price-sign filter.

Do not mistake a scenario frequency for an edge: 22.199% of assumed economic simulations improved profit, but that is not a stock win probability. Keep risk at zero when contract economics improve yet a costed independent payoff distribution remains missing; the funding shortfall does not change that gate.

Resolve issuer identity and delivery status before constructing a helium order: USEG became BSIN, and ASPI’s 15% contracted nameplate volume was still commissioning rather than accepted shipments. Collector failures must remain visible; six of eight primary panels worked, so use manual primary-source review for the two failures.

Separate profit dollars from margin percentage before rejecting scarcity capture: Linde reported positive helium contribution despite margin dilution, while Air Products’ helium bridge remained −$0.07 EPS. A sector-adjusted event return is not causal evidence when the sector ETF holds the same names; retain both benchmark results when their signs differ.

Do not use a deal’s required break-even probability as your estimated success probability: TXNM’s illustrative 76.13% threshold is an assumption-dependent hurdle, not measured edge. Keep size zero when probabilities are uncalibrated or the executable holdout fails, regardless of the $349.01 monthly shortfall.

Set vectorbt explicitly to long-only and reconcile its cash ledger before accepting results; the default allowed unintended short exposure in this run’s rejected draft. Treat sparse-print ATRs as suspect: 23.52% of MRGR’s later bars had zero range, and the accepted 19-cohort holdout still lost 6.07% after costs.

For cash mergers, rebuild the payoff from signed amendments before using the original proxy: TXNM’s July 17 waiver cut the conditional parent fee from $350m to $175m. A refinancing order and a compliance-response deadline are separate from merger approval; neither supplies a calibrated success probability.

Do not let100000 simulated paths manufacture confidence from one historical event: SRRK's positive no-added-hazard model mean1.50% fell to0.04% under an assumed10% adverse-window hazard, while its empirical momentum holdout lost36.34%. Keep size zero when hazard is unmeasured and observed edge fails, regardless of the monthly shortfall.

Test the exact text that survives scheduled-job journaling, not just the script exit code: SRRK job26 exited0 but its saved output cut off before issuer RSS. Put current primary headlines first and keep output bounded; replacement job27 tested at1091bytes, with first sandbox execution still pending.

Read the FDA response letter before treating a manufacturing-only headline as a completed approval checklist: SRRK's September22 2025 letter also requires labeling and safety updates. Never transfer deficiencies at a removed plant to its unnamed replacement, or turn one completed FDA window into a calibrated resubmission edge.