Out-of-sample performance
Walk-forward backtest of the walk-forward Elo home-team win probability. Every game is predicted using only ratings from earlier games, then scored against the result. Task type: binary-head-to-head.
0.2238
Brier score
lower is better
63.5%
Accuracy
home/away pick
0.2460
Baseline Brier
always predict home base rate 56.3%
7,276
Games scored
out-of-sample
The model's Brier score of 0.2238 beats the naive always predict home base rate 56.3% baseline (0.2460) by 9.0%, confirming genuine out-of-sample skill across 7,276 games.
Log loss 0.6392 路 generated 2026-06-09 路 Out-of-sample walk-forward backtest over full nflverse history. Brier and accuracy are for the binary home-win head-to-head; baselineBrier is the constant home-base-rate prediction.