Methodology

How we build each prediction and why we publish our misses as well as our hits.

1. Data

Fixtures, results and tables come from sports data providers. We also use historical results from several seasons, the weather forecast for kick-off time and sports news coverage (always summarised in our own words and linked to the source).

2. Our own statistical model

A bivariate Poisson model estimates each team's expected goals from their home and away performance (giving more weight to recent matches) and combines it with an Elo-style strength rating. The model does not use artificial intelligence: it is reproducible mathematics.

3. Artificial intelligence agents

Eight specialised agents work every day: news, teams, conditions (team news and weather), prediction, verification, SEO, Premium analysis and writing in five languages. The agents can only use the data they are given, and an automatic verifier rejects any text containing figures or claims that are not supported.

4. Honesty rules

• We publish estimated probabilities, never certainties. A 60% probability means that, in similar matches, that outcome should happen about 6 times in 10.
• If the analysis departs by more than 8 points from the statistical baseline, it must be justified with specific data; otherwise it is corrected automatically.
• Premium analysis always includes a probability range and the reasons why it could go wrong.
• Every result is recorded and published on the Track record page, misses included.

5. What we don't do

We don't sell "sure bets", we don't promise winnings and we are not a bookmaker. PronosLab is sports information and analysis content for over-18s.