How it works
Honest test on seasons it never learned from: 2024, 2025
Average miss on the final margin: 12.7 pts (the betting market: 11.9 pts). Picked the straight-up winner 75% of the time.
| Against the closing line | W-L-P | Win % |
|---|---|---|
| All games | 872-929-37 | 48.4% |
| Lean (52%+) | 6-6-0 | 50.0% |
| Pick (54%+) | 0-0-0 | – |
| Strong (56%+) | 0-0-0 | – |
| Against the opening line | W-L-P | Win % |
|---|---|---|
| All games | 853-896-32 | 48.8% |
| Lean (52%+) | 4-4-0 | 50.0% |
| Pick (54%+) | 0-0-0 | – |
| Strong (56%+) | 0-0-0 | – |
Trained on 2016–2023. Built 2026-09-30T07:43.
- Preseason estimate for every team from last season's rating, how much production returns, and roster talent.
- Scoring rating: opponent-adjusted point margin, recent games weighted more, blowouts capped at 28.
- Efficiency rating: opponent-adjusted offense and defense expected points per play with garbage time removed. It is less noisy than the scoreboard.
- Model line: the two ratings plus home field, blended with weights learned from past seasons.
- Cover chance: how often past games covered when the model disagreed with the market by the same amount. Picks need 54%+, strong picks 56%+.
- Matchup breakdown: separate opponent-adjusted run and pass efficiency for every offense and defense, ranked across FBS, to explain each number in plain English.
- Line tracking: the line is recorded when picks lock and twice a day until kickoff, so you can see whether the market moved toward or away from each pick.
- Every Monday afternoon it grades last week, updates ratings, and locks the new picks.
For fun and information only. No model beats the spread every week; never bet more than you can afford to lose.