How it works

There are no "sure things" here. There are probabilities that try to be honest: what we give 60% should happen about 6 times in 10. And you can see where every number comes from.

How strong each team is

Each team has an attack strength and a defence strength. They update after every match. They are measured against the teams it really played: three goals against the bottom team count for less than three goals against the top team. The model learns from 161,317 matches in 149 leagues, since the 2020 season.

Expected goals

For each match, the model starts from the goals an average team would score against an average team. Then it multiplies them by three things: playing at home, the team's attack and the opponent's defence. You can see this sum under "Why?" on every match. It is not a summary: it is exactly the sum the model does.

From goals to everything else

With the expected goals of each team, we work out the chance of every score (0-0, 1-0, 2-1…). The result, the goals, both teams to score and double chance all come from there, so they never disagree with each other.

The league correction

A model can be right in general and still be wrong in the same direction in one league. When a league has enough past results, we correct its probabilities with them. Corners, cards, shots and fouls do not have this correction yet, and we say so on every match.

Everything is saved before kickoff

Every probability is saved before the match starts, together with the version of the model that made it. What you see is exactly what was saved. We will use it to publish how well the numbers held up, league by league, including what goes wrong.

What we do not do

• We do not use bookmaker odds to work out our probabilities.
• We do not promise to beat the bookmakers. For the result and the goals, the closing odds of the big bookmakers are very hard to beat, and our own tests show it.
• No number on this site is written by an AI. They all come from the model, and the text is built from them.