How 21D AI calls the World Cup.

What this is. An explainable forecast, re-run every morning. It is not a black box and not a trained statistical engine. It reads the confirmed draw, the live betting market, current form and squad data, then reasons team by team to a result, the way an experienced analyst would with every number in front of them. We are upfront about that, because the honesty is the point.

The data. Fixtures, the 12 groups and the bracket structure, plus actual results as they land, come from openfootball (public domain). Current FIFA rankings, recent form, squad and player data, injuries and odds come from API-Football. Market-implied probabilities are taken from published bookmaker odds as a cross-check, not scraped from betting sites.

How a prediction is built. Every team gets a strength score from 0 to 100, weighting current ranking and Elo, recent form, squad and key-player quality (downgraded for injuries), tournament pedigree, host advantage for Mexico, the USA and Canada, and market-implied odds as a cross-check. Each group is called on the top two plus the eight best third-placed teams, then the knockout bracket is walked tie by tie, every match carrying a predicted winner, a win probability and a one-line reason, through to the champion. Knockout pairings stay provisional until the eight best third-placed qualifiers are confirmed.

Played matches are fixed. Once a game is played the result is locked and never contradicted. From the Round of 32 onward the predicted bracket is replaced by the real one as qualifiers confirm, and everything still to play is re-projected against the new picture.

The market anchor. As of 8 June 2026, bookmakers had Spain and France as co-favourites at roughly 16% each, England third at roughly 11%, then Brazil, Argentina and Portugal. Our forecast sits close to that market, with England's run at the optimistic but defensible end of their range. It is an honest forecast, not rigged to favour England.

Where it is strong, and where it is not. The strength is transparency: every call traces back to a strength score and a stated reason you can argue with. The limits are just as clear. It is a single reasoning pass rather than thousands of Monte Carlo simulations, so the win probabilities are well-reasoned estimates rather than calibrated odds. The strength scores are reasoned from the inputs above rather than drawn from one rated system, and the knockout pairings depend on the best-third permutations. We would rather state those limits than overclaim.

What 'From the start' means. Two calls carry a 'From the start' tag: our first per-match forecast was generated, validated and recorded at 17:34 UTC on 11 June 2026, about ninety minutes before the opening kick-off, and published with the site's first fixture update on 12 June. Those two calls, covering the two opening games, count in the track record. Every other call on this site was published before the match it predicts, and that rule is enforced by the system, not by promise.

Updated daily. The whole forecast is re-baked every morning by 21D AI, reacting to injuries, form and results. You can see how its mind has changed, and how right it has been, on the track record page.