Prediction guide · World Cup 2026
How Bawler predicts the World Cup
Predicting a World Cup is not the same as predicting a Premier League weekend, and pretending otherwise is how prediction sites get caught out. Here is exactly how the tournament is modelled — and where the model is weaker than usual.
Where the numbers come from
The engine estimates each side's attacking and defensive rate, then derives a full distribution over scorelines — and every market — from it. For clubs those rates come from rolling expected goals. For nations, team strength is anchored on the FIFA World Ranking, adjusted for recent form and squad data, and fed into the same scoring rates.
FIFA rankings are a robust, slow-moving measure of national-team quality — good enough to separate contenders from minnows and set sensible base rates — but they miss squad-specific detail: an injury, a hot striker, a new manager.
Group stage vs knockouts
- Group stage — sides are fresher and motivation varies, since a team already through may rotate. Goal and double-chance forecasts are usually where model accuracy is highest.
- Knockouts — single matches with extra time and penalties. The model forecasts 90-minute outcomes, so we flag where a tie could go to extra time rather than implying certainty.
- The 48-team format — 12 groups of four, with the best third-placed teams advancing, means more games involving mismatched sides early on.
The honest limitations
- Fewer recent competitive matches per team than a club season — more uncertainty.
- Tournament one-offs — red cards, weather, refereeing, shootouts — add variance no model removes.
- Early-tournament predictions are noisier; accuracy shows over the full month, not over a single matchday.
Same rules as everything else
Every World Cup pick is logged before kickoff with a tamper-evident hash and settled automatically — no deletions, no edits — so you can see precisely how the model did across the tournament on the public track record.
Model estimates are not guaranteed outcomes.