How every Results figure is measured

Cohort A — live bot book. Closed positions executed by the trading engine. This is engine evidence, not customer outcomes: it is not what a subscriber receives, fills, or earns.

Cohort B — published signal-model outcomes. The product being purchased: delivered signals under contract v1 (+1R target, 8-hour expiry, manual execution) settled by market-price observation. Each published signal counts once, no matter how many customers reveal it — purchasers are never counted as independent outcomes, and losing signals are never hidden.

Cohort C — historical simulations. Labeled models evaluated on historical data. Simulations are not live results and never substitute for delivered-signal evidence; every published model carries its label, frozen evaluation window, costs treatment, and sample size.

Cohort D — customer execution (future). Opted-in customer fills will appear as their own cohort. Nothing currently published is a customer return claim.

/signals and /results share one convention. Every aggregate names an explicit window start, window end, and as-of timestamp, all in UTC. The /signals aggregate view and the /results bot-book cohort render the same payload fields (window_start, window_end, as_of, period_days); the signal-model cohort renders its own published window.

A transient endpoint failure never rewrites history: when evidence goes stale, pages fail closed to “not currently verified” instead of showing old numbers as current.

Under contract v1, 1R = |entry − stop| and the delivered target is entry ± 1R. A billing WIN means the market touched the target inside the 8-hour window after entry activation (closed candles only) — a defined price-touch event, not proof that any customer filled at those prices or earned that amount.

Entry and target touching in the same candle, or stop and target touched in one candle after activation, classifies the signal AMBIGUOUS — never billed as a win. Signals never activated expire CANCELLED and are refunded; activated-but-unclosed signals at expiry stay in economic analysis with an executable exit mark and never disappear as zero-return cancels.

Modeled average R is reported with its R-sample size and denominator, using the mapping WIN=+1R, LOSS=−1R, AMBIGUOUS=−1R (conservative, loss-equivalent); CANCELLED signals keep their own count and never enter R; unknown states are counted separately and excluded from the total. Rates always carry their denominators (total signals vs decided signals); a percentage without a denominator is not published.

The signal-model cohort is price-touch verification only: no trading costs are netted in it(no fees, spread, slippage, or funding subtracted at signal level). Trading costs apply to any customer execution and are handled in the paid contract's economics study (B03/B04) — they are never subtracted inside this cohort, and the cohort never implies inclusion by silence.

The $59/month subscription and credit spend are service fees, not trading costs — they are never netted into trading expectancy. A positive trading signal can still be uneconomic for a subscriber after the service fee; customer-economics analysis keeps subscription cash, credit consumption, and trading expectancy in separate lines.

• Small samples carry low confidence; short windows annualize poorly (Sharpe needs ≥30 trades and is flagged otherwise).
• Sparse and negative periods remain visible — charts show losing months with their labels rather than dropping them.
• Market-price observation cannot prove customer fills, slippage, missed entries, or capacity.
• Past performance, backtests, and shadow-mode simulations do not guarantee future results.
• Missing or stale evidence reads unavailable, never optimistic.