Betting
Model Performance
Prediction accuracy and calibration analytics for the WElo model.
Model accuracy 61.8% 123 / 199 correct
Total predictions 199 settled matches with model prediction
Model beats market 8.5% 17 alpha vs 25 anti-alpha
Market accuracy 65.8% 131 / 199 correct
Accuracy over time
Rolling 50-match window accuracy
Calibration
Predicted probability vs actual win rate (favoured player). Diagonal = perfect calibration.
| Bucket | Matches | Predicted | Actual | Diff |
|---|---|---|---|---|
| 0.50-0.55 | 33 | 52.3% | 45.5% | -6.8% |
| 0.55-0.60 | 35 | 57.7% | 68.6% | +10.9% |
| 0.60-0.65 | 39 | 62.5% | 64.1% | +1.6% |
| 0.65-0.70 | 20 | 67.6% | 70.0% | +2.4% |
| 0.70-0.75 | 20 | 72.3% | 65.0% | -7.3% |
| 0.75-0.80 | 19 | 77.5% | 47.4% | -30.1% |
| 0.80-0.85 | 13 | 82.3% | 69.2% | -13.1% |
| 0.85-0.90 | 14 | 87.2% | 78.6% | -8.6% |
| 0.90-0.95 | 4 | 92.4% | 75.0% | -17.4% |
| 0.95-1.00 | 1 | 95.7% | 0.0% | -95.7% |
By surface
| Surface | Total | Correct | Accuracy |
|---|---|---|---|
| Hard | 108 | 67 | 62.0% |
| Clay | 91 | 56 | 61.5% |
By competition
| Competition | Total | Correct | Accuracy |
|---|---|---|---|
| Oeiras Challenger 2026 | 29 | 17 | 58.6% |
| Gwangju Challenger 2026 | 22 | 9 | 40.9% |
| Wuning Challenger 2026 | 22 | 16 | 72.7% |
| Rome Challenger 2026 | 17 | 10 | 58.8% |
| Busan Challenger 2026 | 16 | 11 | 68.8% |
| Tallahassee FL Challenger 2026 | 15 | 8 | 53.3% |
| Santa Cruz de la Sierra Challenger 2026 | 14 | 10 | 71.4% |
| Savannah Challenger 2026 | 13 | 5 | 38.5% |
| Shymkent Challenger 2026 | 10 | 6 | 60.0% |
| Abidjan Challenger 2026 | 7 | 6 | 85.7% |
| Monza Challenger 2026 | 6 | 3 | 50.0% |
| Sarasota Challenger 2026 | 6 | 4 | 66.7% |
| Campinas Challenger 2026 | 5 | 5 | 100.0% |
| Jiujiang Challenger 2026 | 4 | 3 | 75.0% |
| Madrid Challenger 2026 | 4 | 3 | 75.0% |
| Mexico City Challenger 2026 | 4 | 3 | 75.0% |
| Mauthausen Challenger 2026 | 3 | 2 | 66.7% |
| Ostrava Challenger 2026 | 2 | 2 | 100.0% |
Model vs market
Agreement 78.9% 157 / 199 matches
Both correct 106 model and market agree, both right
Model alpha 17 model correct, market wrong
Anti-alpha 25 market correct, model wrong
Both wrong 51 neither predicted the winner
Net alpha -8 alpha minus anti-alpha
Alpha rate -4.0% net alpha as % of all matches