Users noticed our highest-rated over-2.5 calls "rarely miss". We tested the whole band: across 44,728 walk-forward predictions, calls rated 67%+ landed 68.5% against an average claim of 70.2%. Calibrated — no better, no worse.
The band-by-band table: 60-65% claims landed 61.6%; 65-67% landed 66.0%; 67-70% landed 65.8%; 70-75% landed 71.3%; 75%+ landed 77.9%. Only 3.2% of games ever reach the 67% tier, and they cluster in high-scoring leagues. The tier also has seasons: summer leagues hit 72.9% historically, winter leagues 65.0%.
Two honest warnings ship with the star badge this study created. First, streaks happen: the historical record contains six consecutive misses inside the 67%+ band — that is what 70% looks like sometimes, not a broken model. Second, confidence is not value: a true 70% chance at odds implying 74% is a bad bet however safe it feels. The edge column, not the probability, is the value signal.
A larger re-test — every match from 2023 to 2026 replayed walk-forward — split the 67%+ band by how far into the season it was. From game four onward the band held up, or did slightly better than stated. In the first three games it did not: those ratings claimed about 70% and landed about 65%, because early on the model leans on last season and cannot see transfers or a change of manager. So a ★ now needs both sides four games in. On 40,705 later matches that rule leaves 1,579 high-confidence calls, which landed 71.8% against a claimed 69.8%, with a longest losing run of four. The figures above are the original study and are left as published.
Method: walk-forward testing — every prediction is made using only information available before kick-off, then graded against what happened, on the site's own dataset. Questions: contact us.