MODEL EVIDENCE

Forecast validation

Track whether predicted probabilities match real outcomes. Calibration is running in shadow: it does not change published picks or establish a betting edge.

Loading evaluation…

Lower Brier score and log loss are better. Market comparisons use the same games and stored pregame probabilities after removing the sportsbook margin. Missing quotes are excluded from paired comparisons. NFL ties are excluded from binary probability scoring. Historical input repairs limit what old results can establish.