How the forecast works
A probability forecast should show its workings. Our engine fits team strengths from international results, simulates the tournament repeatedly, and reports how often each outcome occurs.
1. Estimate team strength
The underlying open-source model uses a Bayesian Dixon-Coles formulation designed for football scores. Recent international results and team-strength priors inform separate attacking and defensive estimates.
2. Simulate the complete tournament
Each run samples group-stage scores, applies tournament ranking rules, fills the knockout bracket and resolves every knockout match. Repeating this process produces a distribution of plausible tournament paths.
3. Turn frequencies into probabilities
If a team wins 1,400 of 10,000 simulated tournaments, its estimated champion probability is 14%. Match win, draw, loss and scoreline probabilities come directly from the fitted score model.
What the model cannot know
Late injuries, tactical changes, squad selection and rare match events affect model inputs as new data arrives. We publish the data cutoff, simulation count, source commit and historical accuracy with each forecast version.
Open-source attribution
The prediction engine is based on FootballTournamentPrediction by the Alan Turing Institute, used under the MIT license. Our application layer adds data syncing, versioned exports and an interactive presentation of the model output.