Methodology

Two models, one projection. Each game's margin comes from Wolf Kalman, our most accurate margin model. Its total comes from Wolf Decay, our most accurate totals model. Both were chosen by pre-registered backtests on 2014–2021, checked on 2022–23, and confirmed on the 2025 season.

Wolf Kalman gives every team an offense and a defense rating on a points scale. Each result moves a rating by how uncertain that rating was. Ratings barely drift within a season. Between seasons they keep 80% of last year (pulled toward their division's average) and become uncertain again, so early games this season count heavily.

Wolf Decay fits opponent-adjusted offense and defense ratings by weighted least squares. This season's games count fully, last season's count less (0.5× for margins, 0.2× for totals), and two seasons back less again. FBS and FCS teams shrink toward their own division's average.

What we found about recency. Weighting last month more than September did not help. Counting this season more than last season did, by a lot.

No look-ahead. Every projection uses only games completed before the week being projected.

The honest part. The betting market's closing line is still more accurate than both models. We publish leans to measure disagreement with the market, not as advice.