Games evaluated
351
Public validation record
2026 Football Model Performance
These results use official OSPR projections locked before kickoff and evaluated against final scores. Mirrored schedule records and live forecasts are excluded.
Evaluated through Aug 29, 2026
Accuracy calculated Aug 31, 2026 9:52 AM EDT
Performance summary
351 games evaluated with ospr-2026.1
Winner accuracy
61.0%
214 correct
Average margin error
23.3
Lower is better
Average score error
13.4
Per-team points
Brier score
0.229
Probability accuracy
Close projection rate
10.8%
Correct and within 7
Regular season vs. playoffs
A direct comparison using the selected season, week, team, and OSPR model version.
Regular season
- Games
- 351
- Winner accuracy
- 61.0%
- Margin error
- 23.3
Playoffs
- Games
- 0
- Winner accuracy
- —
- Margin error
- —
Performance by week
Results for the selected model version and season phase.
| Week | Games | Winner accuracy | Margin error |
|---|---|---|---|
| Week 2 | 351 | 61.0% | 23.3 |
Probability calibration
Observed win rate should approach the average predicted probability as sample sizes grow.
| Probability | Games | Predicted | Observed | Gap | Brier |
|---|---|---|---|---|---|
| 10–19% | 13 | 17.5% | 23.1% | +5.6 | 0.191 |
| 20–29% | 32 | 26.6% | 25.0% | -1.6 | 0.190 |
| 30–39% | 53 | 35.0% | 37.7% | +2.7 | 0.230 |
| 40–49% | 96 | 45.6% | 49.0% | +3.3 | 0.249 |
| 50–59% | 54 | 55.1% | 51.9% | -3.2 | 0.256 |
| 60–69% | 51 | 65.0% | 62.7% | -2.2 | 0.232 |
| 70–79% | 45 | 74.6% | 73.3% | -1.2 | 0.192 |
| 80–89% | 7 | 82.9% | 71.4% | -11.4 | 0.213 |
Performance by prediction confidence
Newer snapshots use historical calibration and comparable sample size. Legacy snapshots fall back to rating-evidence confidence.
| Confidence | Games | Winner accuracy | Average margin error |
|---|---|---|---|
| High | 252 | 55.6% | 22.8 points |
| Medium | 21 | 76.2% | 21.5 points |
| Low | 78 | 74.4% | 25.2 points |
Performance by risk indicator
A game may appear in more than one group because indicators describe different kinds of uncertainty.
| Indicator | Games | Winner accuracy | Average margin error |
|---|---|---|---|
| Limited Data | 351 | 61.0% | 23.3 points |
| Close Game | 273 | 57.5% | 23.1 points |
| Upset Possible | 253 | 55.7% | 22.8 points |
Largest misses
The largest margin errors in this selection. A miss is a model result to learn from, not automatically a data error.
| Date | Matchup | Projected | Final | Margin error |
|---|---|---|---|---|
| Aug 29, 2026 | Warrensville Heights vs. Cleveland JFK | 18–20 | 57–0 | 59.2 points |
| Aug 28, 2026 | West Clermont vs. Anderson | 24–17 | 19–70 | 58.8 points |
| Aug 28, 2026 | Ross vs. Thurgood Marshall | 19–20 | 57–0 | 57.4 points |
| Aug 29, 2026 | John Marshall vs. Rocky River | 17–24 | 0–63 | 56.8 points |
| Aug 28, 2026 | Wayne Trace vs. Spencerville | 18–19 | 55–0 | 56.2 points |
Evaluated games
The game-level record behind the summary above.
Aug 27, 2026
Perkins vs. Scott
- Projected
- 24–16
- Final
- 58–6
- Projected winner
- Perkins
- Margin error
- 43.8
| Date | Matchup | Projected | Final | Projected winner | Actual winner | Margin error | Result |
|---|---|---|---|---|---|---|---|
| Aug 27, 2026 | 24–16 | 58–6 | Perkins | Perkins | 43.8 | Correct |
How to interpret these numbers
Winner accuracy shows whether OSPR favored the team that won. Average margin error shows how far the projected point margin was from the final margin.
Brier score measures probability quality from 0 to 1; lower is better. Calibration compares predicted win probabilities with observed outcomes.
Small groups can swing sharply, so every percentage is shown with its evaluated game count.
Injuries, roster changes, weather, and information not reflected in recorded results may affect individual games.
