Our Method

A frozen EP perspective

The model combines opponent-adjusted current scoring margin, prior-season margin, first-down efficiency, EPA per play, success rate and explosiveness.

General rankings average exact neutral comparisons against all FBS opponents. Pair-dependent weighting means a ranking difference need not equal a specific matchup margin.

Preseason and every completed week have a fresh archived team assessment. These season forecasts are retrospective reconstructions, distinct from the original published predictions.

Historical 2025 matched-game MAE was 12.256 over 517 games, with a 371–146 straight-up record (fit on 2024 only). The difference from C2 (−0.020 points; 90% interval −0.089 to +0.052) is not statistically established. Play-by-play inputs use game-clock order and reconciled scores; 322 of 3,273 games that could not be reconciled contribute no EPA observation. No independent betting-profitability claim is made.

Game probabilities use a symmetric normal approximation fitted to historical development residuals; calibration is not independently established. Weekly Edge retains the expected-return uncertainty filter and a maximum of five picks. Historical spreads have unknown capture times and any −110 returns are hypothetical.