Football Proof AIProbabilities with a public recordEvidence library

Evidence-first guide

How to read a football prediction probability

Direct answer: a 54% home-win probability means the home win is the largest of three uncertain outcomes. It does not mean the team is “54% certain” to win this one match, and it leaves 46% combined probability for a draw or away win.

Published by Football Proof AI · Updated · Deterministic examples, no generated picks

Step 1

Read all three 1X2 outcomes

Home, draw and away probabilities must form one complete set and add to 100%. The top call is simply the largest number.

Home54%
Draw26%
Away20%

54% + 26% + 20% = 100%. The alternatives still total 46%.

Step 2

Translate probability into long-run frequency

Across enough genuinely similar forecasts labelled 54%, the selected outcome should occur close to 54 times per 100. One result cannot establish whether the number was honest.

54 / 100

Expected occurrences over a sufficiently large, comparable sample

Definition: calibration →

Step 3

Fair odds are a conversion, not a recommendation

Fair decimal odds1 ÷ probability

1 ÷ 0.54 = 1.85

This mathematical reference contains no bookmaker margin and does not prove that any offered price is valuable. Football Proof AI does not turn it into a stake or profit promise. Read the full definition of fair odds.

The same probability discipline applies to yes/no markets. See how it changes the interpretation of Both Teams to Score and Over 2.5 Goals without turning either number into a guarantee.

To compare a complete 1X2 split with genuinely similar historical forecasts, read the deterministic, publication-time-safe Probability Neighbours protocol.

The three-part audit

How to tell whether a prediction model deserves attention

  1. 01

    Check publication proof

    The forecast needs a pre-match timestamp, immutable lead time and model version—not a screenshot posted after the result.

  2. 02

    Check the complete record

    Hits and misses must use the same settlement rules. Selective winning examples are not an accuracy record.

  3. 03

    Check probability quality

    Hit rate covers only the largest outcome. Calibration and multiclass Brier score test whether all three numbers behave honestly.