worldcup.infratek.ai
Next match
the pre-match forecastWho wins the cup?
P(win) · 32 teams · 10,000 sims · interval shown for every team| # | Team | P(win cup) | Interval |
|---|---|---|---|
| 1 | 🇦🇷Argentina | 32% | 31%–33% |
| 2 | 🇫🇷France | 13% | 12%–14% |
| 3 | 🇪🇸Spain | 12% | 12%–13% |
| 4 | 🏴England | 8.4% | 8.0%–8.9% |
| 5 | 🇧🇷Brazil | 7.9% | 7.5%–8.3% |
| 6 | 🇵🇹Portugal | 4.5% | 4.2%–4.8% |
| 7 | 🇳🇱Netherlands | 4.2% | 3.9%–4.5% |
| 8 | 🇩🇪Germany | 3.7% | 3.3%–4.0% |
| 9 | 🇧🇪Belgium | 2.1% | 1.9%–2.4% |
| 10 | 🇺🇸United States | 1.7% | 1.5%–1.9% |
| 11 | 🇨🇴Colombia | 1.7% | 1.5%–1.9% |
| 12 | 🇲🇽Mexico | 1.3% | 1.1%–1.5% |
| 13 | 🇭🇷Croatia | 1.2% | 1.0%–1.4% |
| 14 | 🇨🇦Canada | 1.2% | 1.0%–1.4% |
| 15 | 🇲🇦Morocco | 1.1% | 0.9%–1.3% |
| 16 | 🇨🇭Switzerland | 0.9% | 0.7%–1.0% |
| 17 | NOR | 0.6% | 0.5%–0.7% |
| 18 | 🇯🇵Japan | 0.4% | 0.3%–0.5% |
| 19 | 🇸🇳Senegal | 0.4% | 0.3%–0.5% |
| 20 | 🇪🇨Ecuador | 0.2% | 0.1%–0.3% |
| 21 | ALG | 0.1% | <0.1%–0.2% |
| 22 | 🇨🇮Côte d'Ivoire | 0.1% | <0.1%–0.2% |
| 23 | 🇦🇺Australia | 0.1% | <0.1%–0.2% |
| 24 | GHA | 0.1% | <0.1%–0.2% |
| 25 | SWE | 0.1% | <0.1%–0.2% |
| 26 | 🇦🇹Austria | 0.1% | <0.1%–0.1% |
| 27 | 🇵🇾Paraguay | <0.1% | <0.1%–0.1% |
| 28 | BIH | <0.1% | <0.1%–0.1% |
| 29 | 🇪🇬Egypt | <0.1% | <0.1%–0.1% |
| 30 | COD | <0.1% | 0%–<0.1% |
| 31 | CPV | <0.1% | 0%–<0.1% |
| 32 | RSA | 0% | 0%–0% |
Road to the final
the bracket shape — results fill in as matches are playedHow this forecast works — and what "calibrated" means
The headline number is P(win the cup): out of many simulated run-throughs of the remaining bracket, the share a team lifts the trophy. It is not a prediction that any one team will win — a 16% favorite is still far likelier to go home than to win it all. That is the point.
The model is an ensemble. A strength model (Elo-style team ratings) gives per-match win/draw/loss odds; a goals model (Dixon–Coles bivariate Poisson) gives scorelines; market-implied odds act as a prior and a benchmark to beat. The blend feeds a Monte-Carlo simulation over the fixed knockout bracket, which tallies how often each team reaches each round and wins the cup.
Every number carries an interval. We sample the model's own parameter uncertainty before simulating, so the range you see is real spread, not decoration. We never publish a point without its band, and we never animate the band — a quivering needle reads as panic; a static range reads as honesty.
"Calibrated" means our 70%s should happen about 70% of the time. Once matches are played we score ourselves on exactly that — Brier score and reliability — and we publish every miss in a public error log, with a one-line cause. The model learns ratings from results as the tournament runs, but its feature weights are tuned on history (past World Cups, Euros, Copas), never on these few games.