⚽ World Cup 2026 — who should I cheer for?

See the odds as they stood after an earlier match: every probability re-derives using only the results played up to and including the match you pick.

• Model: goals are Poisson draws scaled by an Elo expectation on FIFA's 600-point scale (avg 2.7 goals/match); knockout ties break with a small skill edge as an extra-time/penalties proxy. Ratings are the official FIFA points from the June 11, 2026 release, the last ranking before the tournament. No host advantage, no injuries.
• Rules: official 2026 regulations end to end — head-to-head-first group tiebreakers, ranking of third-placed teams, and FIFA's real Annex C table (all 495 combinations) routing the best 8 thirds into the bracket.
• Live results: actual scores are merged in as the tournament is played, from ESPN's public scoreboard. Played games are pinned as fact; the simulation only fills in what hasn't happened yet, so every probability sharpens as the tournament goes. Knockout slots fill in with real teams as soon as the results force them (a clinched group winner, or a slot every simulation now agrees on), so "Group G winner" becomes the actual team across My matches, Probabilities, and Team path before the official fixture is published.
• Your preferences: teams start ranked by strength (the default assumption: you want to see the best teams); head-to-head picks personalize from there, and ⭐ pinned teams are locked above everything else. "Lineup quality" is the favorite-weighted appeal of the distinct teams you'd watch across your matches.
• Cheer guide: one big simulation records every game's outcome and who ends up in your attended matches; bucketing by outcome gives the causal effect of each result on your expected lineup quality. A loyalty guard stops it from telling you to root against a team you love unless the payoff is real (for that team or another favorite) — and it will never tell you to cheer against a team you've pinned.
• Must-watch: your team ranking becomes a personalized TV schedule — every remaining match scored by the expected appeal of the teams on the pitch (knockout slots weighted by who's likely to fill them), with one-tap export to your calendar (.ics).
• Accuracy: the model also publishes a win/draw/loss probability for every match (shown on each game in My matches). The Accuracy tab grades those calls against real results with a reliability diagram that checks whether, say, the games called ~60% home win really do end that way about 60% of the time. Scored underneath with the multiclass Brier score (0–2, lower is better), our model against a random guess. It's one tournament with correlated outcomes, and football is high-variance, so read it as descriptive, not a verdict.
• Team path: follow one team and the same simulation scores every remaining game by how much its result extends their expected run — group survival, but also which bracket slots and opponents each result steers them toward. It will tell you honestly when a loss would route them somewhere friendlier, but it never tells you to cheer against your own team.
• Privacy: everything runs in your browser; your picks live in localStorage and never leave your device. The only measurement is an anonymous, cookieless page-view count (GoatCounter).

Static site — simulations run client-side. Source on GitHub.