Brier score
The mean squared error between predicted probabilities and actual outcomes. Lower is better.
The Brier score takes each prediction, squares the difference between the stated probability and the actual result (1 for a win, 0 for a loss), and averages across all predictions. A perfect forecaster scores 0. Always guessing 50 percent scores 0.25.
It rewards being both accurate and appropriately confident. A confident wrong call is punished far more than a hesitant wrong call, which is exactly the incentive an honest forecasting system should face.
Brier score is best read alongside a reliability diagram. The score tells you how good the forecasts are overall; the diagram tells you where the model is over- or under-confident.
Related terms
See it applied
Every published call on AI4Gameday carries a stated confidence and is graded in public once the game is final.