Scoreboard
Since week 7 · 67 ranked-team games: Model 69% · AP 69% · Split picks 12 (Model 6–6 AP)
Warm-up (weeks 2–6) · 90 ranked-team games: Model 76% · AP 80% · Split picks 14 (Model 5–9 AP)
ML engine · 194/290 (67%) · our model on the same games 184/290 (63%) · Brier 0.210 · about the ML engine
Backtest, 2014–2025 (ex. 2020), ranked-team games: the model trailed the AP by 3.5 points before week 7 (74.9% vs 78.4%) and led it from week 7 on (77.5% vs 75.8%).
All FBS games (AP has no pick when neither team is ranked): Model 65% · Brier 0.215 · 511 games graded
| Week | Ranked-team games | Model % | AP % | Split picks | Splits won (Model–AP) | All FBS games | Model % (all) | Brier | ML % (games) |
|---|---|---|---|---|---|---|---|---|---|
| 2 | 18 | 83% | 83% | 2 | 1–1 | 49 | 63% | 0.219 | — |
| 3 | 20 | 75% | 80% | 3 | 1–2 | 57 | 68% | 0.175 | — |
| 4 | 18 | 78% | 72% | 3 | 2–1 | 58 | 66% | 0.208 | — |
| 5 | 15 | 87% | 93% | 1 | 0–1 | 56 | 73% | 0.181 | — |
| 6 | 19 | 58% | 74% | 5 | 1–4 | 58 | 64% | 0.239 | 60% (57) |
| 7 | 16 | 56% | 63% | 5 | 2–3 | 59 | 59% | 0.248 | 64% (59) |
| 8 | 17 | 76% | 65% | 2 | 2–0 | 56 | 61% | 0.252 | 64% (56) |
| 9 | 18 | 72% | 72% | 0 | 0–0 | 56 | 70% | 0.196 | 79% (56) |
| 10 | 16 | 69% | 75% | 5 | 2–3 | 62 | 65% | 0.213 | 68% (62) |