Crowd wisdom from my LinkedIn poll → Raw vote shares, normalized across the final four. Hearts, not spreadsheets.
Elo ratings built from every international match since 1901 — each result nudges a team's rating, weighted by match importance (World Cup games count ~4× a friendly) and recency. 200,000 Monte Carlo runs of the remaining bracket.
An ensemble: a base rating from Elo and FIFA points (60/40), adjusted up or down for tournament form and player quality (Transfermarkt squad values — France €1.52B vs Switzerland €333M) — then blended with de-vigged betting-market odds.
Why Monte Carlo? A tournament isn't one prediction — it's a chain of them. To win, a team must survive its semifinal and beat whoever emerges from the other side, and each possible opponent changes the odds. Instead of trying to write one giant formula, Monte Carlo just plays out the whole bracket over and over, flipping a weighted coin for every match (weighted by Elo strength). A team's title probability is simply the share of simulated tournaments it wins. Same technique banks use for risk and forecasters use for elections.
Why 200,000? Because simulations are random, running them twice gives slightly different answers — that wobble shrinks with more runs. The noise on a probability estimate is roughly √(p(1−p)/n): at 10,000 runs, a "35%" could read anywhere from 34% to 36%; at 200,000 runs it's stable to about ±0.2% — steadier than the inputs themselves. Beyond that, more simulations add compute, not accuracy: the answer is only as good as the ratings going in.