Every match, modeled & scored.
A Dixon–Coles + Elo engine locks a scoreline before kickoff, then grades itself against reality. No hindsight, no edits — just the receipts.
Live Scorecard
How the model is performing across every graded fixture.
Getting Sharper
Accuracy per round and the running cumulative accuracy — the model re-fits its Elo, form and calibration after every matchday.
Knockout Bracket
Paul's path to the trophy. Each tie shows the predicted scoreline; the predicted winner is highlighted, and played ties are graded once the result is in.
Predictions & Results
Every pick is locked before kickoff. Each card shows what Paul predicted next to what actually happened — then grades itself.
Futures
The long bets locked at the start of the tournament — shown against Paul's current live favourite, so you can watch them hold or drift.
Golden Boot Race
Today's live top-scorer standings with each contender's current goal count — plus Paul's updated pick, re-projected from goals scored and how deep each player's team is expected to run. ⚽ marks a penalty taker.
Title Race
Championship odds from 20,000 simulations of the actual set bracket. Ties already played are locked in as certain — that's why a team can show 100% to reach the next round while the round itself is still underway — and every tie still to be played is simulated by the model, path and all.
Behind the Scenes
Paul isn't a lucky octopus. Every pick is an ensemble of opponent-aware signals fused into expected goals, then a probability matrix over every scoreline.
Final Report 2026 Retrospective
The tournament is over and the record is locked. Here's an honest look at what the model got right, what it missed, and why.
✅ What worked
- Both season-long futures — Champion (Spain) and Golden Boot (Kylian Mbappé) — were locked before a ball was kicked, and both held all the way through.
- Mbappé finished on 10 goals, clear of Messi (8) and Haaland (7) — the pick was never seriously threatened after the group stage.
- The Elo + Dixon–Coles core called the right side of the result in 75.8% of graded matches, and accuracy climbed as the model re-fit itself after every matchday (see "Getting Sharper" above).
- The Final: Paul had Spain to beat Argentina, and Spain did — 1–0, just not the 2–1 scoreline predicted.
❌ What missed
- Exact scorelines are brutally hard: only 11.0% (10 of 91) landed on the nose. Football's low-scoring, high-variance nature punishes point predictions even when the process behind them is sound.
- The single biggest miss: the Third-place playoff. Paul had France edging England 1–0 — instead it was a ten-goal thriller, England 6–4. No model saw that one coming.
- Mid-tournament, the per-match model and the whole-bracket simulation briefly told two different stories — one had France eliminating Spain in a single tie, the other still had Spain as the runaway favourite to win it all. Both were "correct" for what they measured; single-match upsets and tournament-long favourites are different questions.
Every model, script and result behind this report is public — the Elo ratings, the Dixon–Coles fit, the calibration, the Monte Carlo bracket sim, all of it. Fork it, point it at the next tournament, or just read the code that made these calls.