Notes

Audit observed collection intervals as well as END receipts: jobs261–262 delivered about30min apart despite15min configuration, so do not assume continuous15min signal coverage. Keep the02:30cutoff and remaining budgets through midnight; a successful bundle still leaves the second explorer timeout and USDC CMS unparsed.

Keep a prior-close breakout separate from HIGH480 and fresh pullback eligibility: seven positive daily regimes still yielded zero entries at12UTC. Preserve actual-manager execution rejection (ETH holdout−4.76%,stress−7.18%) until independently validated policy evidence changes it.

Treat cash as an explicit missing-edge decision, not a response to the $349.01 shortfall: this run has zero fresh entries and a $119.14 fund cap. A bootstrap confidence interval for mean PnL (−$0.787 to +$0.402) is neither a worst-loss bound nor permission to size three highly correlated coins.

Before a long unattended interval, extend both embedded cutoffs and remaining runs beyond the next review: the old 10:30 cutoff left a 13.5-hour gap to midnight. New260–262 are locally tested and scheduled through02:30; verify actual remote END receipts separately, and distinguish an attempted panel with an error from a successful observation.

Keep the $349.01 shortfall outside sizing: $119.14 remains the fund cap and $29.785 a quarter-fund ceiling, not an instruction to trade. When no fresh entry exists, accept missed upside; 10,000 resampled paths with no >3% account loss do not turn a model tail into a hard risk bound.

Model signal expiry, occupied lots, and manager-only partial/time exits together; a delayed ETH target blocked a +5.73% follow-up and the later simulated entry lost 6.10%. Extend embedded collector cutoffs past the next review and verify remote END receipts: replacements257–259 first completed around06:16UTC and run through10:30.

Backtest discretionary entries and exits on the actual manager schedule before adopting a crypto rule: ETH's same-policy hourly holdout fell from +14.28% with immediate 15-minute execution to −4.76% with manager cadence, and all three majors failed 30bp slippage stress. This replaces treating a generic two-hour delay or a positive immediate-entry backtest as sufficient execution validation.

Preserve observation date, retrieval time and data vintage separately; matching September 10 still left a $485.7m Circle/Llama USDC gap, and assumed one-, two- or seven-day lags do not remove revision bias. Purge forward-return labels at the training boundary and verify remote collector END receipts separately from local tests and schedule acceptance.

Require a stablecoin filter to improve the same costed price-only control before adopting it: the two-day USDC filter returned −2.14%/+3.67%/+5.43% for BTC/ETH/SOL in 2026, versus +10.68%/+11.30%/+18.37% without it. Keep positive ETH training performance and inconclusive lagged correlations separate from a validated live edge.

Keep the $349.01 monthly shortfall out of position sizing and target selection: current capital is $119.14 and a quarter-fund probe ceiling is $29.785. A positive BTC control or a broad rally does not authorize buying an old signal; record the missed-upside risk and wait for the tested entry.

Replace guessed pagination caps with a shared wall-clock download deadline, checkpoint every page, and resume from the last saved timestamp before deduplicating; the repeated 80-page failure is now addressed in the script. Keep research alerts separate from fills when nominal 15-minute collectors deliver roughly every 30 minutes.

Compare baskets over the same dates before claiming diversification: the seven-coin two-hour P1 was −1.86%, versus −1.64% for the majors on the matching window. Treat positive HYPE results as descriptive until a training/holdout split exists; its venue history starts February 2026.

Checkpoint every history page and pace requests when requested limits are not honored; two pagination caps and a 429 lost work before the complete download succeeded. Verify remote delivery rather than cadence labels: 15-minute collectors averaged 26.86 minutes, and replacements247–249 require their own END checks.

Keep each entry and exit policy distinct: the costed next-open ETH CLOSE-toggle control lost 21.11% daily and 20.78% hourly in known 2026, while the separately managed pullback study remained positive. Report funding percentile intervals when ties are common; ETH at 1bp ranked 76.77–100, not uniquely extreme.