All your compliments and your cutting remarks are captured here in my quotation marks. "Everyday I Write the Book" - Elvis Costello

Accuracy 2026 — every FINAL game scored on leakage-free pre-game predictions

Metric Model 3 Book Closing Line
Spread MAE (n=5,531) 9.41 8.92
Winner % 70.9% 72.7%
Total MAE (n=5,531) 14.27 13.37
Log loss 0.5553 0.5129
Brier 0.1885 0.1732

Head-to-head on identical games: model spread MAE 9.41 vs book 8.89 over 5,487 games; taking the model against the closing spread hits 50.3% (52.4% is breakeven at −110). Full breakdowns on Compare Models. Not betting advice.

What it knows — the 27 inputs per game

Form & schedule (15)

home_win_pct_l5, home_avg_margin_l5, home_avg_total_l5, home_margin_stddev_l5, away_win_pct_l5, away_avg_margin_l5, away_avg_total_l5, away_margin_stddev_l5, home_games_played, away_games_played, home_days_rest, away_days_rest, season_week, is_neutral_site, is_conference_game

Ratings (12)

massey_beta_home, massey_beta_away, massey_beta_diff, massey_gamma_home, massey_gamma_away, massey_gamma_sum, bt_theta_home, bt_theta_away, bt_logodds, bt_theta_weighted_home, bt_theta_weighted_away, bt_logodds_weighted

Training

Trained
2026-07-27
Train seasons
2021, 2022, 2023, 2024, 2025
Held-out test
2026
Test spread RMSE
12.00
Test Brier
0.1885

Walk-forward (each season held out in turn)

SeasonSpread RMSEBrier
2022 11.64 0.1897
2023 11.63 0.1914
2024 11.68 0.1890
2025 11.84 0.1842
2026 12.01 0.1876

Calibration when it says 70%, does the home team win 70% of the time?