Track Record
How the model actually performs on Greek Super League matches • Predictions through: 2026-09-11 12:41:08
Matches evaluated
28
finished matches with a stored prediction
Pick accuracy
57.1%
most likely outcome was correct
Log loss
0.911
uniform guess = 1.099 — lower is better
Brier score
0.534
0 = perfect, lower is better
Probability calibration
When the model says 60%, the event should happen about 60% of the time. The dashed line is perfection; the solid line is the model across all three outcomes.
Overall reliability curve, all three outcomes pooled. Sparse buckets (few matches) naturally wobble more — the current sample is small this early in the season.
Per-outcome diagnostics
| Outcome | Precision | Recall | F1 | Matches |
|---|---|---|---|---|
| Home win | 67% | 86% | 75% | 14 |
| Draw | 0% | 0% | 0% | 7 |
| Away win | 40% | 57% | 47% | 7 |
Draws are the hardest outcome in football — precision/recall stay low for every honest model.
Latest results
| Date | Match | Pick | Result | |
|---|---|---|---|---|
| 2026-09-13 | PAOK FC – Aris Thessaloniki FC | Home win (63.9%) | 2–0 | ✓ |
| 2026-09-13 | Panathinaikos FC – Panetolikos FC | Home win (66.6%) | 3–0 | ✓ |
| 2026-09-13 | Kifisia FC – Levadiakos FC | Home win (43.0%) | 0–0 | ✗ |
| 2026-09-13 | Kalamata FC – Volos NFC | Away win (41.3%) | 3–1 | ✗ |
| 2026-09-13 | Asteras Tripolis FC – AEK Athens FC | Away win (61.1%) | 1–1 | ✗ |
| 2026-09-12 | Olympiacos FC – OFI Crete FC | Home win (67.5%) | 0–1 | ✗ |
| 2026-09-12 | Iraklis FC – Atromitos FC | Away win (44.2%) | 1–1 | ✗ |
| 2026-09-10 | Panathinaikos FC – Kifisia FC | Home win (65.2%) | 3–1 | ✓ |
| 2026-09-07 | Asteras Tripolis FC – Iraklis FC | Home win (44.0%) | 0–2 | ✗ |
| 2026-09-06 | Panathinaikos FC – PAOK FC | Home win (48.0%) | 3–1 | ✓ |
| 2026-09-06 | OFI Crete FC – Kifisia FC | Home win (48.6%) | 2–0 | ✓ |
| 2026-09-06 | Levadiakos FC – Panetolikos FC | Home win (40.6%) | 0–2 | ✗ |
What we don't publish: betting simulations, ROI figures or “value picks”. The numbers above are about the quality of the probabilities, not about wagering — see the methodology for how they are produced.