Prem ProjectionsYour complete Premier League hubVillaVerdictWorld Cup 2026

How accurate is the model?

Every probability this site prints was replayed against history: 9,874 Premier League matches, 2000-01 to 2025-26, forecast exactly as the live site would have — the model’s own numbers blended with the betting market where match odds exist, walking forward through each season with no knowledge of what came next. This page is the result, updated when the model changes.

This season so far: 12 of 20 played matches called from the frozen pre-kickoff forecasts — the narrowest: Brentford 3–0 Tottenham, called at 37%.

When we say 60%, does it happen 60% of the time?

Every win, draw and loss probability from the replay, grouped by its stated value and checked against what actually happened. Points on the dashed line are perfect calibration: a forecast of 45% that comes true 44.8% of the time. The purple line is the site’s published forecast; amber is the model without the market blend. One honest wrinkle: our strongest calls are slightly conservative — matches we price at 65% land about 69% of the time.

0%0%25%25%50%50%75%75%100%100%forecast 8% → happened 5.8% (n=939)forecast 16% → happened 14.3% (n=4,149)forecast 26% → happened 25.8% (n=12,054)forecast 35% → happened 33.5% (n=4,306)forecast 45% → happened 44.8% (n=3,265)forecast 55% → happened 57.2% (n=2,404)forecast 65% → happened 68.8% (n=1,409)forecast 74% → happened 77.3% (n=904)forecast 83% → happened 88.5% (n=191)

Called it

The simpler question: when the site’s most likely outcome was the answer, how often was it right? For reference, always picking the home side hits about 46%, and no public model clears the mid-50s over decades.

WindowMatchesHit rate
All matches9,87453.9%
Early season (first 8 rounds)2,36252.8%
Mid-season (rounds 9–25)4,43552.7%
Run-in (round 26 on)3,07756.4%

The live version of this record runs on the fixtures page, where every finished match keeps its frozen pre-kickoff forecast and a called/upset verdict.