Signal explorer
Every input the intelligence layer produces becomes a named, timestamped signal — graded against outcomes and kept only if it earns its keep.
How to read this. Each signal's win rate is compared with the engine's base rate over the same graded fixtures (0 fixtures, base —). Lift is the difference; the verdict is earned, not asserted: keep needs a statistically meaningful lift over a real sample, monitor means not enough evidence yet, drop means it tracked the wrong way. No signal can touch a live decision until it has earned keep — measurement first, always.
SignalSourceStanceGraded nWin rateLift vs baseAvg CLVVerdict
Signals are accumulating — the shadow layer records them daily and grades them as fixtures settle.
Data from the engine snapshot · updated just now · all prices are real captured bookmaker odds, timestamped.