Every pick carries a confidence number. It's a probability estimate, not a promise: a 75% pick should lose about one time in four. If it never lost, the number would be wrong.
Things worth knowing when you read it:
- Higher confidence usually means shorter odds. A very confident pick often pays little; value lives in the gap between our estimate and the bookmaker's price, not in confidence alone.
- Calibration is checkable. Across many graded picks, the share of 70%-confidence winners should sit near 70%. For model picks we also track the Brier score — a standard measure of how honest stated probabilities are (0 is perfect, 0.25 is a coin flip).
- Small samples lie. A 5-0 week can be luck; that's why authors and models carry a "building history" label until they have a meaningful number of graded picks.
Probabilities, not promises — 18+, and never stake what you can't afford to lose.