Per-engine quality: ROI, Sharpe, CLV, drawdown, statistical significance, recommended action. Closing-line-value column proves a winning engine is genuinely sharp rather than just lucky.
How these numbers are defined — the same definitions are used on
/track-record and /transparency.
Decided bets = wins + losses. Voids (stake returned) and pending bets are counted
separately and are never included in the bet total. ROI = profit ÷ stake risked on decided bets;
void stakes are not in the denominator. Where a figure uses a different basis it says so on the figure.
Confidence calibration
Does each engine's grade actually predict outcome?
—
Each row is one engine × grade. A well-calibrated engine shows hit rate dropping monotonically as the grade gets worse (A+ > A > B+ > B...). When the order is broken, the engine isn't ranking risk correctly — even if the average ROI is positive.