Model Performance
A probability model is judged by calibration over samples, not by single-match hits. Every engine change must move these numbers on out-of-sample matches — or it gets cut.
2026-27 Season · Cumulative (pregame predictions, no hindsight)
Matches scored
63
Model log loss
0.9864
League-prior baseline
1.0370
Edge vs baseline
+0.0506
Brier score
0.5892
Modal picks correct
33/63
Lower log loss = better. Baseline = constant league base rates (home 44%, draw 24%, away 32%). Sample is still tiny (63 matches): treat every number here as noise until at least a half-season has accumulated.
Over 2.5 Calibration
Matches scored
63
Actual Over 2.5 rate
71.4%
Model avg Over 2.5
68.2%
Over 2.5 Brier
0.1958
Over 2.5 log loss
0.5796
Hits (3+ goals)
45/63
Pregame Over 2.5 probability (same number as on each match analysis page) vs whether the final score actually had 3+ goals.
| Model said | n | Model avg | Actual Over rate | Δ |
|---|---|---|---|---|
| 40–50% | 1 | 49.1% | 100.0% (1/1) | +50.9 pp |
| 50–60% | 8 | 57.3% | 37.5% (3/8) | -19.8 pp |
| 60–70% | 30 | 64.9% | 70.0% (21/30) | +5.1 pp |
| 70–80% | 18 | 75.3% | 83.3% (15/18) | +8.0 pp |
| 80–90% | 6 | 81.4% | 83.3% (5/6) | +1.9 pp |
Well-calibrated bands sit near Δ ≈ 0 (actual rate ≈ model average). Positive Δ = model under-called Over; negative = over-called.
Per Gameweek
| GW | Matches | Model log loss | Baseline | Δ | Picks |
|---|---|---|---|---|---|
| GW 1 | 9 | 0.9029 | 0.9557 | +0.0527 | 5/9 |
| GW 2 | 9 | 1.0557 | 1.0652 | +0.0096 | 4/9 |
| GW 3 | 9 | 0.8785 | 1.0618 | +0.1833 | 5/9 |
| GW 4 | 9 | 1.0103 | 1.0264 | +0.0162 | 4/9 |
| GW 5 | 9 | 1.0029 | 1.0618 | +0.0589 | 6/9 |
| GW 6 | 9 | 0.9881 | 1.0264 | +0.0384 | 5/9 |
| GW 7 | 9 | 1.0665 | 1.0618 | -0.0046 | 4/9 |