Final results across every competition we cover, each shown against the model's pre-match prediction — logged before kickoff, so the verdict is honest. ✓ Result = the model called the 1X2 outcome, ★ Exact = it nailed the scoreline.
This page is the match-by-match record. For whether the probabilities themselves hold up — calibration, and how the model scores against an uninformed baseline — see the model accuracy report.
Ranked by skill — how much of the available ground the model gained on a forecast that knew nothing and split every match evenly. Not by hit rate: a league of one-sided fixtures is easier to call, so a high hit rate there is partly a fact about the league rather than about the model.
| Competition | Skill | Correct 1X2 | Settled |
|---|---|---|---|
| +23.6% | 65.0% | 40 | |
| +19.8% | 62.9% | 35 | |
| +19.2% | 63.9% | 72 | |
| +18.8% | 59.8% | 82 | |
| +18.7% | 51.4% | 37 |
Competitions with at least 25 settled pre-match predictions. At these sample sizes a hit rate carries a margin of roughly ±11 to ±17 points, so read the order as a guide and not as a standing. Full table, all 75 competitions, ordered by hit rate →
The last 400 finished matches, each with the model's pre-match call and whether it landed.