How the model is doing

Every published prediction stays in the record.

Historical rolling backtestEstablished evaluation sample

This backtest simulates historical fixtures in chronological order. Its forecasts were never published publicly and are separate from the immutable published record.

Log Loss1.044Equal 33% baseline: 1.099
Brier Score0.628Equal 33% baseline: 0.667
RPS0.223Equal 33% baseline: 0.222
Top-outcome accuracy48.2%Equal 33% baseline: 33.3%
Matches scored987Evaluated matches
Predicted probability compared with observed frequency0%0%25%25%50%50%75%75%100%100%Predicted probabilityObserved frequency
View calibration data
PredictedObservedEvents
16%19%290
28%28%1872
48%47%652
66%60%136
84%73%11