What a probability means
When Vesta gives a team 45%, a team in that position wins about 45 times in 100: not every time, not never. A good model is one whose 45% really happen 45% of the time. That is called calibration, and we measure it in public on the track record.
Club matches
- Data. Every match of the Premier League, La Liga, Serie A, Bundesliga and Ligue 1 since 2014-15: results and expected goals (xG), the quality of the chances each team created. The site also shows the results and line-ups back to 2014-15; Vesta does not learn from them, since those seasons have no expected goals.
- Strength. Vesta estimates each team's attack and defence from the quality of the chances it creates and allows and from its results, giving recent matches more weight, together with home advantage and the players a team is likely to field.
- Every score. The strengths give each team's expected goals; from them Vesta computes the chance of every exact score. Every other number is added up from the scores, so they always agree.
- Calibration. A last step adjusts the probabilities so that, on past seasons, what Vesta says matches what happens. Only information available before kickoff is used.
National teams
Every international since 1872. Strength is a rating updated after every match, with more weight for competitive matches and for home advantage unless the ground is neutral.
How it was tested
Vesta was built and chosen on seasons up to 2023-24 only, then run once, frozen, on the 3,504 club matches of 2024-25 and 2025-26. It was more accurate than simpler models on the same data, such as a plain Elo rating or a goals-only model. National teams were judged the same way on the matches from 2022 to 2026.
Reliability labels
- Tested: checked on past matches at the same level of detail.
- Tested in aggregate: checked only on a coarser measure it comes from.
- Model estimate: not checked on real data yet.
What Vesta cannot know
News after the forecast was computed, like a late injury or a surprise line-up; weather, travel, motivation. That is why a 70% favourite still fails to win three times in ten. No tips, no odds, no bookmakers: the numbers are for information and entertainment.