The model's picks
Every week the model picks a winner for each game and a side against the spread. Here is its full record, untouched.
2026 season
No graded games yet this season.
How it works
Each team has an Elo rating that goes up when it wins and down when it loses, more so the bigger and more unexpected the result. At the start of each season, ratings move back toward the average.
For each game the model adds home-field advantage and the effect of rest (coming off a bye, for example).
If the starting quarterback is on the injury report, the team loses 80 points when he is out, 60 when doubtful and 20 when questionable.
The rating difference gives the win probability and the projected margin. The pick against the spread takes the side where the model margin beats the market line.
Against the spread, hovering around 50% is normal for any model: the market line already reflects almost all the available information.
Backtest on past seasons
Before publishing a single pick, the model was tested on seasons already played, using for each game only the information available before kickoff.
| Season | Games | Winners | Against the spread | Brier |
|---|---|---|---|---|
| 2015 | 267 | 68% | 60% | 0.224 |
| 2016 | 267 | 66% | 51% | 0.217 |
| 2017 | 267 | 67% | 50% | 0.217 |
| 2018 | 267 | 62% | 52% | 0.221 |
| 2019 | 267 | 64% | 51% | 0.224 |
| 2020 | 269 | 67% | 56% | 0.215 |
| 2021 | 285 | 61% | 48% | 0.233 |
| 2022 | 284 | 63% | 51% | 0.222 |
| 2023 | 285 | 60% | 46% | 0.231 |
| 2024 | 285 | 68% | 51% | 0.212 |
| 2025 | 285 | 65% | 51% | 0.221 |
| 2015-2025 | 3,028 | 65% | 51% |
Brier: the mean error of the probabilities (lower is better; always saying 50% scores 0.25).
Picks are the output of a statistical model, not betting advice.