Statistics
Real-data validation first; the demo backtest below uses simulated matches and is not real-world performance.
Real-data validation
Loading…
Demo backtest (simulated matches)
Model Performance
Demo- Correct: 37
- Incorrect: 43
- Total: 80
Walk-forward backtest on simulated demo matches. Lower Brier/log loss is better; a uniform guess scores 0.667 / 1.099.
Results over time (rolling 50 matches)
Calibration (top-pick probability vs observed hit rate)
Points on the dashed line mean the stated probabilities match reality. Expected calibration error: 0.082.
By league and market (all time)
| League | n | 1X2 | O/U 2.5 | BTTS | Brier | Log loss |
|---|---|---|---|---|---|---|
| Premier League | 144 | 48.6% | 50.0% | 44.4% | 0.614 | 1.025 |
| La Liga | 144 | 50.7% | 50.0% | 55.6% | 0.611 | 1.021 |
| Serie A | 144 | 50.0% | 45.1% | 50.7% | 0.621 | 1.033 |
| Bundesliga | 144 | 61.8% | 49.3% | 54.2% | 0.592 | 0.995 |
| Ligue 1 | 144 | 56.3% | 45.8% | 52.8% | 0.595 | 0.995 |
| All | 720 | 53.5% | 48.1% | 51.5% | 0.607 | 1.014 |
Research target: 90%+ on carefully defined categories. Not achieved or verified — 1X2 football outcomes are inherently uncertain and real-world models typically land around 50–55%.
Elo rating history — Premier League (demo)
Probabilities are estimates, never certainties. The long-term research target is 90%+ accuracy on narrowly defined markets — this is a goal, not a verified result. Figures shown are measured on simulated DEMO matches until real data is connected.