How accurate is the model, honestly?
The public scoreboard: overall accuracy, calibration by confidence bucket, and the bookmaker-price paper ledger. No real bets or guaranteed returns are represented here.
Overall accuracy
1545resolved predictions
62.8%accuracy (favored side wins)
0.711log-loss (lower is better)
Calibration
When the model says a side wins with probability X, how often does it actually win? A calibrated model's actual win rate tracks the bucket midpoint.
| Confidence | Predictions | Actual win rate |
|---|---|---|
| 0.5-0.6 | 440 | 49.1% (target ≈ 55%) |
| 0.6-0.7 | 353 | 59.8% (target ≈ 65%) |
| 0.7-0.8 | 311 | 66.2% (target ≈ 75%) |
| 0.8-0.9 | 291 | 72.9% (target ≈ 85%) |
| 0.9-1.0 | 150 | 84.0% (target ≈ 95%) |
Bookmaker-price paper signals
11settled signals
9wins
Small sample: 11 settled bookmaker-price paper signals. Aggregate return is withheld until 100 results.
Signals use the last observed bookmaker quote before the scheduled start; that quote may be older than the true closing line. See the full archive →