Backtest results
The model did not beat the historical break-even rate at −110. Historical performance does not establish this week’s win probability.
ATS measures covering the spread. MAE is mean absolute margin error in points; lower is better. Binomial p measures evidence against a 50% cover-rate baseline, not the probability a pick will win.
Backtest results
Break-even at −110: 52.4% · n= 505 graded (ATS) · n=516 projected (edges)
| Bucket | N | Accuracy |
|---|---|---|
| 0-3 | 98 | 51.0% |
| 3-7 | 134 | 64.9% |
| 7-14 | 186 | 71.5% |
| 14+ | 98 | 92.9% |
| Bucket | N | Accuracy |
|---|---|---|
| 0-3 | 225 | 51.6% |
| 3-7 | 175 | 50.9% |
| 7-14 | 97 | 46.4% |
| 14+ | 8 | 75.0% |
| Week | Matchup | Proj | Actual |
|---|---|---|---|
| 12 | San José State @ Nevada | -10.4 | +45.0 |
| 11 | San Diego State @ Hawai'i | -14.9 | +32.0 |
| 8 | Old Dominion @ James Madison | -5.2 | +36.0 |
| 10 | Fresno State @ Boise State | +17.7 | -23.0 |
| 10 | Arizona @ Colorado | +4.1 | -35.0 |
| 6 | Boston College @ Pittsburgh | +2.1 | +41.0 |
| 7 | Old Dominion @ Marshall | -13.3 | +24.0 |
| 7 | Louisiana Tech @ Kennesaw State | -9.3 | +28.0 |
| 7 | South Florida @ North Texas | +10.2 | -27.0 |
| 7 | Rice @ UTSA | +10.9 | +48.0 |
| 7 | Iowa @ Wisconsin | -0.5 | -37.0 |
| 14 | Fresno State @ San José State | +8.4 | -27.0 |
| 12 | Northern Illinois @ Massachusetts | -6.8 | -42.0 |
| 10 | James Madison @ Texas State | +3.1 | -32.0 |
| 8 | UTSA @ North Texas | +2.9 | +38.0 |
Projected winner · confusion matrices
| Actual ↓ / Called → | Home win called | Away win called |
|---|---|---|
| Home won | 1,678 (correct) | 517 (wrong) |
| Away won | 638 (wrong) | 1,015 (correct) |
Accuracy 70.0% on 3,848 games · home recall 76.4% · away recall 61.4% · called home 60.2% vs actual home share 57.0%
| Actual ↓ / Called → | Home win called | Away win called |
|---|---|---|
| Home won | 1,694 (correct) | 493 (wrong) |
| Away won | 567 (wrong) | 1,084 (correct) |
Accuracy 72.4% on 3,838 games · home recall 77.5% · away recall 65.7% · called home 58.9% vs actual home share 57.0% · 10 no-call
Straight-up winner calls from the sign of the projected home margin (model) and the recorded line (market) on the same walk-forward games. Highlighted cells are correct calls and gray cells are misses; stronger shading means more games. Both call the home side more often than it wins; the market is the harder benchmark, as the backtest above already says.