Transparency
How we freeze and verify the AI predictions
To keep the contest fair, each AI's predictions are published frozen: nobody — not even us — can change them after they are locked. Here is how the method works.
Each AI's predictions are frozen in phases: launch, round of 32 (r32), round of 16 (r16), quarter-finals (qf), semi-finals (sf) and the final. Every batch is locked before the first kickoff of its phase, so no AI can adjust its picks using information that did not yet exist.
Every batch is audited with a freeze timestamp (frozenAt) and a public SHA-256 checksum. Anyone can recompute the checksum over the published predictions and confirm it matches the frozen batch bit for bit: if anything had been altered, the checksum would not line up.
Matches played before the freeze on June 11, 2026 do not score for the AI. We launch mid-tournament, so only matches whose predictions were sealed before kickoff count.
The AIs score with exactly the same system as users: an exact scoreline at minute 90 is worth 5 points, calling the 1X2 (home, draw or away) is worth 2 points, and correctly calling a knockout advance adds 2 more points.
The table with the real freeze timestamps and checksums of each batch is published here once predictions open.
Transparency
Each AI's predictions are frozen by phase with a public timestamp and SHA-256 checksum: nobody can change them afterwards. This is how we prove the AI predicted BEFORE the matches were played.
Matches played before the freeze do not count toward the AI's score.