In one sentence: FATE plays the whole World Cup nine thousand times — from each team's strength rating, honoring real results as they lock into history — and reports how often each outcome happens. Probabilities are frequencies, not opinions.

1. Strength — one scalar per team, s = ln(ev/38)·0.42; team EV is the only strength input.
2. Match goals — opponent-aware Poisson, λ = 1.32·exp(±ha + Δs); the home edge ha applies only to a 2026 host (USA/MEX/CAN) at home — a World Cup is otherwise neutral, so we add no phantom edge from fixture order. Live matches taper by minute / xG / red cards.
3. Scoreline — Dixon-Coles τ correction (ρ = −0.11) for realistic draw rates. A played match uses its real score every simulation.
4. Group tables — FIFA 2026 Art. 13: head-to-head before overall goal difference, then a coin.
5. Third place — the real best-8-of-12 by (points, GD, GF) fill the 32-team knockout.
6. Knockout — strength-seeded bracket: regulation → extra time → penalties.
7. Monte-Carlo — n = 9,000 tournaments; P(reach a round) is just how often it happens. River mass halves 48→32→16→8→4→2→1.
8. Sources — results: ESPN (live) · odds: Polymarket + Kalshi (live, labeled separately) · PELE: Nate Silver's model (snapshot) · the headline: our own Monte-Carlo.
9. Independence (honest) — those sources are correlated estimators of one truth, not independent votes: Polymarket≈Kalshi (one market, two quotes); PELE's live rating & its champion-% are one model. So agreement doesn't multiply confidence — only the Divergence view is truly informative. And the simulation runs on EV + real results only — markets/PELE are comparison, never double-counted into the projection.
10. Determinism — a seeded RNG makes every run replayable; 14/14 conformance checks lock the behavior.

It does NOT model: lineups, injuries, suspensions, rotation, xG history, fatigue, travel, weather, referees, in-game momentum, transfers, or motivation/dead rubbers. Disciplined uncertainty — explicitly not prophecy. Full written breakdown →