We don't name a champion in a 48-team field — that's a coin-flip dressed as a call. We lock a transparent model-desk ensemble (Opta + market + Elo) and clean, Brier-scorable partitions. The product is the calibration and the shown working, not a hero guess. The bracket Monte-Carlo locks at the Round-of-16.
Plus P(UEFA winner)=65.8% and P(≥1 of {France, Spain, England, Argentina} reaches the Final)=72.5%, with a full ensemble distribution (France 14.9 / Spain 14.2 / England 11.2 / …). We synthesise + calibrate public models — we do not claim to beat Opta. Scored by Brier.
We lock a full ensemble distribution over the title — no team above ~15% in a 48-team field, which is itself the honest signal.
| Team | Locked win probability |
|---|---|
| France | 14.9% |
| Spain | 14.2% |
| England | 11.2% |
| Argentina | 9.8% |
| Portugal | 7.1% |
| Brazil | 6.7% |
| Germany | 5.4% |
| Netherlands | 4.3% |
| Norway | 3.1% |
| All other 39 teams | 23.3% |
The actual locked claims are these bracket-independent partitions, each a clean yes/no scored by Brier against the published result.
| Partition | Locked probability |
|---|---|
| A first-time nation wins | 36.7% |
| Winner from UEFA | 65.8% |
| Winner from CONMEBOL | ~20% |
| ≥1 of {France, Spain, England, Argentina} reaches the Final | 72.5% |
Instead of eight geopolitical actor-desks (P-02), we run three forecaster desks — an Opta-supercomputer desk (weight 0.30, reduced for staleness), a betting-market-implied desk (0.55, absorbs live group-stage form), and an Elo desk (0.15, Bradley-Terry from current ratings) — each normalised to 100%, then weighted into a transparent RAOSCAFF consensus. We synthesise and calibrate credible public models; we do not claim a proprietary football engine. The full working — per-desk numbers and the Elo mapping — is shown so any third party can reproduce it.
The market moved France to solo favourite on one group result while Elo still rates Spain #1 globally. Our ensemble collapses that spread — France 14.9% vs Spain 14.2%, near co-equal — treating the one-game form signal as partly informative, not decisive. A mild lean against the market's France-first pricing.
A calibrated forecaster expects most individual outcomes not to happen; the test is whether the probabilities are well-calibrated, scored by Brier / log-loss on the distribution and each partition. We pre-commit: if a first-time nation wins, our 36.7% scores well on that partition; if a prior-winner takes it, the same number is tested the other way. The bracket Monte-Carlo — a separate simulation over the actual knockout tree — locks at the Round-of-16.
RAOSCAFF locks P-06 on 2026-06-20, during the group stage, before the knockouts. We do not name a champion. We lock a calibrated probability distribution and a set of clean partitions, and we will score them by Brier / log-loss against the published result — the same way we grade any forecast. The bracket Monte-Carlo locks separately at the Round-of-16.
A calibrated forecaster expects most individual outcomes not to happen; the test is whether the probabilities are well-calibrated, not whether one pick lands.