Anyone can show winners. Two things are hard to fake: whether stated probabilities match reality, and whether the price you got beat the price at kick-off. Both are below, including the one we cannot show yet.
Expected calibration error is the average gap between a stated probability and the frequency actually observed on settled matches. Under 5% is good. A model above that ceiling never earns a Prime badge on the board, however large the gap looks.
The share of matches where the outcome we made favourite actually won. A coin flip is 50%; the market itself lands around 60 to 65% on these sports, so accuracy alone proves very little. It is here for completeness, not as the headline.