NINETYSTATS

Football probabilities that mean what they say.

Built from 30,920 matches of real data. Frozen before kick-off, graded in public, misses included.

See this week’s numbersHow we’re measured
Graded · Premier League19 May 2026

Chelsea vs Tottenham

2-1

Chelsea winhappened58%consensus 43%
Draw22%consensus 28%
Tottenham win20%consensus 28%

One entry from the graded record: our probability was frozen before kick-off, then judged against the result. The record keeps the misses too — see all of it.

The calibration proof

When we say 65%, it happens 65% of the time.

Anyone can publish a percentage. The only question that matters is whether the percentages are true. Ours are audited in every probability band, and the audit is public.

When we say 30–40%happened 34.7% of 354
When we say 40–50%happened 43.5% of 947
When we say 50–60%happened 57.6% of 668
When we say 60–70%happened 65.1% of 424
When we say 70–80%happened 73.2% of 198
When we say 80%+happened 88.4% of 69

Green bar: how often that outcome actually happened. White tick: the average probability we had stated. Premier League backtest, 2,660 matches, 2019–2026.

We tell you exactly how good we are. And how good we aren’t.

The betting market’s consensus is the sharpest football forecast on earth, priced with team news we cannot see. We measure ourselves against it on matches our model was never tuned on, and we publish the result either way.

Distance behind the market
+2.58%
measured on three seasons we never tuned on
Average drift from reality
1.5pp
our stated percentages vs what actually happened
Most likely outcome correct
52.7%
the market: 54.6% on the same matches

Open about results. Careful with the recipe.

  1. 01

    Real data in

    Three decades of English football — results, performance detail and context for every match since 1993. Bookmaker odds are never an input.

  2. 02

    Our own model

    Built entirely in-house to read how well teams are actually playing, not just what the scoreline said. The exact method is ours; every claim about its output is publicly testable.

  3. 03

    Public grading

    Probabilities freeze 24 hours before kick-off into a permanent record. After full time, every one is graded against what happened and against the market.

More on how we work, and what the model cannot see: read the methodology.

The 2026–27 record starts at matchweek one.

Every published probability this season joins the permanent record, hit or miss. Watch it accumulate here.

See the record