AI vs Humans

Pre-tournament · vs the model

How the brackets people submitted on /predict stack up against the model. Before kickoff there are no results yet, so every bracket is scored on how closely it tracks the model’s own projection — not on being right.

Benchmark: model elo-dc-v1.0 · 50,000 sims · pre-tournament · updated 10 Jun 2026

Not enough brackets yet0 / 5 needed

Crowd insights unlock at 5submitted brackets. Until then, here's the model's projected bracket as the benchmark — submit yours on /predict to see how it compares.

The model's projected bracket

the benchmark

ChampionESPSpain

Finalists

ESPSpainARGArgentina

Semi-finals

FRAFranceESPSpainBRABrazilARGArgentina

Quarter-finals

FRAFranceNEDNetherlandsESPSpainBELBelgiumBRABrazilMEXMexicoARGArgentinaPORPortugal

Group winners

MEXMexicoCANCanadaBRABrazilUSAUnited StatesGERGermanyNEDNetherlandsBELBelgiumESPSpainFRAFranceARGArgentinaPORPortugalENGEngland

Human brackets are scored on how much they overlap with the model · run forecast__elo-dc-v1.0__seed42__cond0__rerun20260610.

How scoring works

max 172 pts

A simple round-weighted scheme: deeper correct calls are worth more. Each entry earns points for agreement with the model — no upset bonuses or tie-breakers (richer leaderboard scoring is a V2 idea, deliberately out of scope pre-launch).

Group winner — each of the 12 group winners1 pt
Round of 32 — each team you advance to the Round of 162 pts
Round of 16 — each team you advance to the quarter-finals4 pts
Quarter-final — each team you advance to the semi-finals8 pts
Semi-final — each finalist you call16 pts
Champion — the champion32 pts