RaoscaffResearch
Prediction Series · Lock · Issue P-06
Prediction Series · P-06

The World Cup, as a calibrated forecast — not a pick.

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.

Type · Prediction Lock · model-desk ensemble + calibrated partitions Locked · 2026-06-20 · group stage · final July 19 Resolves · 2026-07-19 · FIFA result Scored · Brier / log-loss on the distribution + partitions
What we lock · a distribution, not a name
36.7%
P(a first-time nation wins) · our calibrated partition

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.

— 1 · The Locked Forecast

A calibrated distribution + clean partitions.

We lock a full ensemble distribution over the title — no team above ~15% in a 48-team field, which is itself the honest signal.

2026 World Cup · locked ensemble (Opta 0.30 · Market 0.55 · Elo 0.15) · sums to 100%
TeamLocked win probability
France14.9%
Spain14.2%
England11.2%
Argentina9.8%
Portugal7.1%
Brazil6.7%
Germany5.4%
Netherlands4.3%
Norway3.1%
All other 39 teams23.3%
— 2 · The Partitions

Clean, Brier-scorable bets — no named champion.

The actual locked claims are these bracket-independent partitions, each a clean yes/no scored by Brier against the published result.

Locked partitions · bracket-independent
PartitionLocked probability
A first-time nation wins36.7%
Winner from UEFA65.8%
Winner from CONMEBOL~20%
≥1 of {France, Spain, England, Argentina} reaches the Final72.5%
— 3 · Methodology — the model desks

The position-desk method, reborn for forecasting.

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.

— 4 · The Variant View

We fade the market's France-first move toward Spain co-equal.

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.

— 5 · Scoring & Pre-Committed Post-Mortem

Scored by calibration — not by 'did our pick win'.

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.

A calibrated forecast, locked and timestamped — scored by Brier, not by 'did our pick win'.

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.

Locked
2026-06-20 (commit timestamp on origin/main)
Resolves
2026-07-19 — FIFA World Cup Final, MetLife Stadium
Source
FIFA official result
Scored by
Brier / log-loss on the distribution + each partition (Big-4 frozen as France, Spain, England, Argentina)

A calibrated forecaster expects most individual outcomes not to happen; the test is whether the probabilities are well-calibrated, not whether one pick lands.