Where the model actually knows something
Every number below comes from walk-forward testing: the model is trained only on matches before each month, then scored on that month. “Improvement” is how much better its Brier score is than always predicting the league average. Positive means signal; zero or negative means we switch that market off.
| League | Matches tested | Home win | Over 2.5 | Corners 9.5 | Cards 4.5 |
|---|---|---|---|---|---|
| EPL | 790 | +11.2% | +0.1% | -0.1% | +0.7% |
| LaLiga | 801 | +14.2% | +2.5% | +2.4% | +1.2% |
| SerieA | 790 | +13.6% | -2.3% | +1.9% | +4.1% |
| Bundesliga | 630 | +14.3% | +2.0% | +0.9% | -0.1% |
| Ligue1 | 639 | +9.2% | +0.2% | -0.8% | +2.8% |
| Eredivisie | 656 | +11.5% | +0.1% | -0.3% | -2.6% |
| Portugal | 656 | +19.8% | +3.8% | +0.5% | -2.2% |
| Belgium | 668 | +11.0% | -4.4% | -0.6% | +2.9% |
| Turkey | 684 | +11.0% | +3.0% | +0.6% | +0.2% |
| Greece | 490 | +17.4% | -0.1% | -0.8% | +1.0% |
| Scotland | 489 | +14.5% | -0.1% | +0.2% | +0.8% |