Explore
Track record
Live predictions
Published before kick-off. 325 matches, 2026-10-23 to 2026-10-27.
Calibration
When the model says 70%, does it happen about 70% of the time? Points on the diagonal mean yes.
If you only took picks the model rated at least X% likely, this is how often they won. Base rate across all matches: 56%. Brier score 0.244 (lower is better).
| Model says ≥ | Picks | Won |
|---|---|---|
| 50% | 184 | 59.8% |
| 60% | 73 | 65.8% |
| 70% | 16 | 68.8% |
| 80% | 1 | 0.0% |
| 90% | 0 | – |
Value bets by market
Bets where model probability × average odds exceeded 105% of stake
| Market | Bets | Hit rate | ROI |
|---|---|---|---|
| Home win | 61 | 31.1% | -18.4% |
| Away win | 75 | 20.0% | -21.6% |
| Draw | 51 | 11.8% | -54.3% |
| Over 2.5 goals | 10 | 40.0% | -30.9% |
| Under 2.5 goals | 69 | 40.6% | -10.9% |
By league
Log loss on 1X2 for the same matches. Where the model is lower than the bookmakers, it was sharper.
Walk-forward backtest
Each week the model was refitted using only results available at the time, then asked to predict the following week. 14,205 matches, 2025-08-01 to 2026-10-21.
Calibration
When the model says 70%, does it happen about 70% of the time? Points on the diagonal mean yes.
If you only took picks the model rated at least X% likely, this is how often they won. Base rate across all matches: 53%. Brier score 0.245 (lower is better).
| Model says ≥ | Picks | Won |
|---|---|---|
| 50% | 7,287 | 58.3% |
| 60% | 2,577 | 64.6% |
| 70% | 440 | 73.4% |
| 80% | 46 | 84.8% |
| 90% | 0 | – |
Value bets by market
Bets where model probability × average odds exceeded 105% of stake
| Market | Bets | Hit rate | ROI |
|---|---|---|---|
| Home win | 3,410 | 30.1% | -17.0% |
| Away win | 3,636 | 22.2% | -16.3% |
| Draw | 1,525 | 20.7% | -11.1% |
| Over 2.5 goals | 1,005 | 50.8% | -4.3% |
| Under 2.5 goals | 1,910 | 45.3% | -8.5% |
By league
Log loss on 1X2 for the same matches. Where the model is lower than the bookmakers, it was sharper.