Free evidence tool · No sign-up

Football probability & Brier score calculator

Turn a complete home-draw-away forecast into fair odds, visible uncertainty and a reproducible probability-quality score.

Interactive · Formula v1

Build a 1X2 forecast. Score the result.

Move any outcome and the other two rebalance automatically, so the split always totals 100%. Then choose a hypothetical actual result to calculate the class-averaged three-outcome Brier score used on this site.

Probability check100%Always a complete outcome set

Choose the actual result
Class-averaged Brier score0.1064

Lower is better. This is one outcome's probability error—not enough evidence to judge a model on its own.

Top call
Home win · 54%
Top-two gap
28 points
Uncertainty
High uncertainty
Normalised entropy
0.915

Probability made visible

100 comparable forecast worlds

Expected mix across 100 comparable forecasts if calibrated—not 100 simulations of this fixture.

  • Home win54
  • Draw26
  • Away win20

Same prediction, three possible outcomes

See why the actual result changes the score

Home win occurs0.1064Selected result
Draw occurs0.2931Alternative result
Away win occurs0.3331Alternative result

Exact formula

How is a three-outcome Brier score calculated?

Subtract the observed outcome—1 for what happened, 0 for what did not—from each probability, square all three errors, add them, then divide by three.

Brier3-class average((pH − yH)² + (pD − yD)² + (pA − yA)²) ÷ 3

54 / 26 / 20 + home result = (0.46² + 0.26² + 0.20²) ÷ 3 = 0.1064

Reference points

What does the number mean?

0.0000
A perfect probability of 100% assigned to the result that occurred.
0.2222
An equal 33.33% / 33.33% / 33.33% forecast, regardless of which result occurs.
0.6667
The maximum: 100% assigned to one outcome while a different outcome occurs.

One match can reward or punish a probability split, but it cannot establish model quality. Judge a model across a complete sample alongside calibration, sample size and publication proof.

Continue the audit

Calculate here. Verify on the public ledger.

Football Proof AI uses the same class-averaged definition for its settled 1X2 record. The glossary fixes the terminology; the accuracy page shows how the metric behaves across real samples.

Open accuracy ledgerAudit a complete CSV recordDefine Brier scoreRead probability guide