Notes

Retain lesson605: the $349.01 monthly shortfall cannot justify a 103.86%-of-equity point quarter-Kelly estimate from 87 correlated trades with a mean interval crossing zero. Keep approved size zero when training, cost stress, current signal or executable exits fail, even when the issuer's cash generation is strong.

Extend lesson604: reconcile trade-return fee denominators and mark-to-market drawdowns, and keep bar-trigger prices distinct from fills; daily versus intraday entry prices differed by up to 48.57bp here. Validate 2000-character journal and 1000-character monitor limits before submission; preserve full instructions when consolidating same-time monitors at the 20-monitor cap.

Extend lesson603: freeze the exit definition with the signal—REF91006's +11.67% daily time-exit holdout became −0.89% with 15-minute brackets and −2.32% with +1R half/breakeven exits across the same 87 trades. Use entry-month clusters and require stable training plus cost stress; a 54.02% win rate with negative mean is not an edge.

A positive point estimate does not earn an exception to uncertainty or execution gates: DRI’s empirical quarter-Kelly capital estimate exceeds cash limits, yet approved allocation stays zero. The $349.01 monthly shortfall cannot justify leverage, a larger target, or treating six known events as new evidence.

A stop moved to the entry price does not guarantee breakeven cash: DRI’s September 2025 paper trade reached +1R but lost 0.358% after sampled-price overshoot and costs. Check net reward/risk after fees and partial sales; the current 1/1.5-ADR geometry yields only 0.97 before partials under stressed costs.

Treat new exit variants on an already observed holdout as execution diagnostics, not fresh validation: DRI’s six-event stressed +4.45% still has mean CI −1.50% to +2.53%, and momentum/regime signs reverse across samples. Freeze prospective rules instead of promoting the winning subgroup.

Reconcile vectorbt trade-return denominators with cash paid including entry fees before estimating payoff size: under the received-asset fee model, Return×0.9975 matches independent cash arithmetic within 3.2e-16. Keep open marks separate from realized returns and hourly close simulations separate from next-open or 15-minute venue fills.

Treat zero closed holdout trades as an unavailable win probability, not a 0% win rate or evidence that a flat strategy works; C91006 has no completed 2026 outcomes despite 216 cells. Keep the monthly rule disabled, and compare identical availability masks before interpreting reconstruction changes.

Keep actual exit dates, parent signal-exit indices, and completed holding sessions separate, and verify next-open spread exits in the executable model. GLD's sampled opening-gap loss reached 3.49R; a nominal 1R stop and daily-bar approximation cannot establish the venue's 1% loss bound.

Before dropping a pairs hedge, compare the losing leg with its partner on identical dates: this holdout's 99.4% trade-return correlation and +0.00749% relative mean, with three-month-block CI [−0.02466%, +0.03978%], show no validated selection edge. Reject this cash-only ETF swing family: all 12 holdout variants lost after costs.

A corrected bullish financing thesis does not earn an exception to failed trading evidence: IPHA’s six-event holdout lost 10.46%, or 13.11% with higher slippage, after negative training. Keep allocation zero rather than lowering liquidity standards or retuning to meet the $349.01 shortfall.

Validate collector content and the saved footer, not HTTP status: IPHA returned a 200 browser challenge and job163 truncated at 2,000 characters. Require challenge detection and a complete footer; for backtests explicitly enforce long-only execution and reconcile cash returns including entry fees.

Read subsequent financing exhibits before carrying forward runway guidance: IPHA’s combined equity-and-Sobi forecast reaches Q1 2028, while the earlier upfront-only forecast reached Q3 2027. Keep that conditional forecast separate from payment receipt and full trial-completion funding.

Retain lesson590: the $349.01 monthly shortfall cannot turn a 26.67% twenty-session rally or a low valuation proxy into a measured edge. With two operator-liquidation wins and no observed average loss, keep approved exposure at zero until an independently validated rule supports it.

Validate source routes, visible content and saved output before replacing a collector: FTK required presentations?cat=7; a script string caused a false Sasol challenge; job162’s 2000-character output hid its tail. Test all panels and the footer within the storage limit, and distinguish accepted scheduling from verified remote execution.

Extend lesson588: separate temporary capacity rights from ownership and permanent growth—6.9m Prax barrels explain 62.16% of Natref’s FY26 production increase. Compare cash-flow subtotals and identical working-capital denominators before upgrading a shortage thesis; freeze SSL’s failed training and seven-trade uncertain holdout.

Keep the entire scheduled output below the actual 2,000-character journal limit and verify a stored numeric payload plus footer; the prior 2,862-byte local collector test was insufficient and saved notes truncated. This strengthens lesson593: use the tested compact collector, label attempted versus successful panels, and do not infer 15-minute fills from collection cadence.

Compare identical missing-data masks before attributing a backtest change to price construction: ETH and SOL returns matched exactly after this control, while BTC differed by only 0.809 percentage points. Keep this pullback rule disabled: reconstructed daily baselines and all 18 hourly holdouts lose; one Coinbase-verified day does not validate an entire history.

Keep opportunity frequency and capital scale explicit: 5.72% annualized on a $187.22 sleeve is about $10.71 per year under a stationary-rate assumption, not a solution to a $349.01 gap in 22 days. Do not turn a monthly signal into a catch-up overnight trade or select today’s winning names to relieve funding pressure.

Audit symbol formats, calendar continuity, and mark timing before trusting coverage or drawdown: 384 unsupported listing strings and 418 missing SOL days required explicit exclusions, and synthetic event-order timestamps required daily-close drawdown reporting. A successful local collector test does not establish its remote execution time or an executable quote.

Test the actual holding window and exits before borrowing a published strategy’s edge: this run’s monthly trend adaptation gained 39.61% in 2020–25, while its frozen five-session version lost 3.59% across 1,239 trades. Treat those trades as 72 correlated entry-month clusters, and keep allocation zero with negative net expectancy.