DSA Track RecordBacktest
Men's College Basketball 2024–25
A backtest, not a live record. Each week the ratings were rebuilt using only games played before that week, then used to predict that week's games. No game was predicted by a model that had already seen it. Live tracking for basketball hasn't started.
Updated Oct 23, 2026
- Favorite won
- 68.6%
- Right–wrong
- 3,611–1,650
- Games graded
- 5,261
- Brier score
- 0.205
Lower is better. Always guessing 50% scores 0.250.
Picking winners isn't the whole story. Plenty of games are close to coin flips, and a 55% favorite is supposed to lose almost half the time. The real test is whether the probabilities are honest, which is what the calibration chart below checks.
01
Calibration: does 70% mean 70%?
For each confidence band, how often the favorite actually won. A 70–80% favorite should win about 75% of the time. Points on the dashed line are perfectly calibrated; below the line means the model was overconfident, above means it was too cautious. Bigger points mean more games.
| Band | Games | Predicted | Actual | Gap |
|---|---|---|---|---|
| 50-60% | 1,759 | 55.0% | 56.9% | +1.9 pts |
| 60-70% | 1,512 | 64.8% | 67.5% | +2.6 pts |
| 70-80% | 1,181 | 74.6% | 77.3% | +2.7 pts |
| 80-90% | 668 | 84.3% | 81.6% | -2.7 pts |
| 90-100% | 141 | 92.7% | 93.6% | +0.9 pts |
02
Week by week
Right and wrong calls for every week of the backtest. Games with an exact 50/50 prediction count as neither.
| Week | Right–wrong | Accuracy | Right versus wrong |
|---|---|---|---|
| 1 | 242–78 | 75.6% | |
| 2 | 221–95 | 69.9% | |
| 3 | 207–115 | 64.2% | |
| 4 | 210– |
Schedule and results: CollegeBasketballData.com.