Walk-forward archive
Historical analysis
How the models performed on past games they had never seen — month by month, by confidence level, and game by game.
NFL — monthly Brier score
Lower is betterCalibration heatmap
Do predicted probabilities match how often teams actually won?Accuracy by confidence
Ensemble pick accuracy vs. its stated confidence (white tick)Prediction log
Every walk-forward prediction next to the actual result, newest firstpage 1
Backtest rows are regenerated on each nightly retrain; live pregame snapshots (immutable) are on each game page and in Model Performance → live record.