The Janus record — goals, cards & corners, graded

Across 12358 graded claims our stated chances miss what happened by 1.9 points on average.

Alongside them, 566 are frozen and await grading, 1252 stayed on the league norm, 5817 sat out and 73 were voided by cancelled games. A forecast that stayed on the norm carried no match-evidence claim of its own, so it is graded in the table but kept out of the bands.

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

Every lane below is the same test. Claims are sorted by the chance we gave the headline side — over for the goals, cards and corners lines, yes for the scoring calls, and all three outcomes for first team to score. The grey bar is the promise, the filled one is what football delivered.

On each “it happened” bar, the two thin marks bracket where the true rate could sit by chance alone — flip a fair coin 20 times and you can easily get 13 heads, not 10; small samples wobble. A gap between said and happened only starts to matter once “what we said” falls outside the marks, and more claims pull the marks closer together.

Does the match evidence beat the league norm?

Every forecast starts from its league's norm — how often games in that competition actually go over 2.5, see both teams score, and so on — then moves off it when the match evidence says to. The honest question is whether those moves help. Both reads were frozen before kickoff and are graded on exactly the same claims.

On the 12336 claims where the frozen league norm exists, both reads are graded side by side. The two land equally close — we miss by 1.9 points on average, the norm by 2.2.

Pooled error score: us 0.204 · league norm 0.2109.

The forecast-error score — statisticians’ Brier score — counts how far every claim sat from what actually happened (lower is better). The league norm is where each forecast starts, never a separate engine. Every term on this page, explained in the glossary. How these forecasts are made — the full methodology.