Performance
Chronological holdout-style backtest: every fight is predicted from ratings as they existed immediately before that fight, then ratings are updated with the result.
Fights scored
6,013
Brier (out-of-sample)
0.242
baseline 0.248
Accuracy
56.2%
Log loss after prior fight
0.682
Recent calibration sample
| Predicted bin | Count | Mean predicted | Observed A win rate |
|---|---|---|---|
| 0-10% | 0 | — | — |
| 10-20% | 0 | — | — |
| 20-30% | 0 | — | — |
| 30-40% | 0 | — | — |
| 40-50% | 9 | 0.465 | 0.667 |
| 50-60% | 14 | 0.542 | 0.500 |
| 60-70% | 2 | 0.624 | 0.500 |
| 70-80% | 0 | — | — |
| 80-90% | 0 | — | — |
| 90-100% | 0 | — | — |
Consensus/market benchmark is not wired yet; this report verifies the chronological ELO+Glicko baseline on public UFC historical fights.