Every game modeled. Few worth betting.
Opponent-adjusted ratings and per-game efficiency, validated by walk-forward backtest. We publish the model’s edge, and say so when there isn’t one.
One Model, Shown Plainly
Ridge regression over 184 features, trained on 9,306 FBS games with closing lines going back to 2013.
Opponent-Adjusted
Per-game efficiency ratings that account for who the opponent was, with garbage-time snaps excluded.
No Look-Ahead
Every feature is computed from games played before kickoff. No rating can see the result of the game it is predicting.
Calibrated, Not Claimed
Win probability is fitted to 7,022 walk-forward games, not asserted. A 12-point edge is worth about 56% — we say that instead of claiming 80%. Prior season: 28.6% on 14 graded picks.