Methodology

It's not guesswork. It's an ensemble that combines statistical and machine-learning models with the betting market and current news —the latter via AI—, and simulates the full World Cup thousands of times to estimate each team's odds.

01
Each team's strength
An Elo-style rating that learns from match history —official and friendlies— and measures each team's real level.
02
The market, cleaned up
Bookmaker odds (consensus/Pinnacle), stripped of their margin with the Shin method: by far the best-calibrated signal for predicting football.
03
Ensemble (Dixon-Coles)
A Dixon-Coles model —the academic standard for predicting scorelines— blends team strength with the market into a scoreline distribution per match.
04
News via AI
An AI agent monitors the news (injuries, rotations, context) and adjusts the model when there's relevant news.
05
Monte Carlo simulation
The full tournament is played thousands of times, with the official 2026 format and bracket, and the outcomes are counted.

Why trust the numbers

The key piece is anchoring to the market: bookmakers are, by far, the best-calibrated source for estimating football results. Combining our models with that signal —instead of competing against it— is what makes the odds credible. It's recomputed every few hours; that's why they move.

2026 format

48 teams, 12 groups of 4. The top 2 of each group plus the 8 best third-placed teams (32) reach the Round of 32, then it's single-elimination to the final.

Honesty

No model knows the future. This is statistical analysis for informational and entertainment purposes, not a guarantee of results or betting advice. The code and the change log are open.