CFB Predictor

Backtested, walk-forward, 3,113 games: straight-up 70.0% · against the spread 49.9% · totals 52.2% · 12.96 pt MAE. Break-even at −110 is 52.38%, so only the moneyline column has ever cleared it — and even there the closing favourite beats this model. Spread and total picks are shown because you asked to see them; they are the model's opinion, not a tested edge.

These are not shown as a ranked shortlist, because ranking them makes things worse rather than better. Over 2,398 CFB games, sorting spread picks by how far the model sits from the number puts the worst ones on top: 14+ point disagreements went 46.6% ATS (−11.1% ROI) against 49.4% for 0–3 point ones. Totals are now a real per-game projection — NFL from EPA + pace, CFB from PPA + pace, which cut CFB total MAE from 13.59 to 13.18 and replaced a constant that gave every game the same number. They are still not a tested edge: the best NFL fit reached 50.8% against a 52.38% break-even, and the CFB version has never been graded live. Break-even is 52.38%, so nothing here clears it and a "strongest" list would just be the loudest noise.

Every game's spread and total projection is in the full slate below. Read them as the model's opinion of the number, which is all they have ever been.