How the model works

A plain-language summary. No people pick matches, and no AI chatbot writes predictions.

Every match, every market

Every morning our own statistical models estimate probabilities for every covered match across results, goals, half-time, corners and more. They learn from years of football data and are retrained regularly.

Choosing the daily set

From all of those probabilities the selection picks a small daily set of games and a Free game, aiming for picks that are both likely and worth publishing. There is never more than one pick per match. If the day doesn't offer enough strong picks, we publish fewer and say so.

Tested before it goes live

Every change is tested on matches the model has never seen before it is used. A probability has to mean what it says: when the model says 70%, it should be right about 70% of the time.

What the numbers mean

A 70% pick is expected to be wrong about 3 times in 10. Losses are part of an honest record, and every one is shown. The Track Record page shows the calibration table so you can check this for yourself.