Favourite
Argentina
17.3% (±0.5% 95% CI)
Top final
Argentina vs Spain
5.4% of sims
Simulations
20,000
seed 2026
Rating σ
45 Elo
Dixon–Coles ρ = -0.0885
Title probability
Model vs. de-vigged market (overround )
12 groups · win-group share · top-2 advance share
Hover a row for ratings + champion odds. The bar shows the team's advance probability relative to the group leader. Most likely qualifier pair shown below each group.
16 Round-of-32 ties · click a team to highlight every slot it can fill
Each tie shows the 5 most likely occupants per side and the single most-likely pairing.
CHECK24 — point-maximizing tips per group match
4 pts exact · 3 pts tendency+goal-diff (incl. non-exact draws) · 2 pts winner only.
Orange pills mark matches where the EV-optimal tip differs from the most likely scoreline.
Expected total per group sums the per-match EV.
The big picture · 6 steps from rating to champion
One pass of the simulator walks each team through this pipeline. The Monte Carlo just runs it 20,000 times and counts.
Step 1 · Two numbers per team
Every team carries an attack (expected goals vs an average team) and a
defense (expected goals conceded — lower is better). Derived from Elo + style + a Poisson regression on recent results.
League average λ = goals — that's the "average team" benchmark.
Step 2 · One match, one formula
Step 3 · From λ to scorelines
Each side's goals are independently drawn from a Poisson with its λ. Real football under-counts 0-0 / 1-0 / 1-1 — the
Dixon-Coles correction (ρ = ) bumps probability into those four cells.
Poisson marginals — P(k goals)
Joint scoreline grid (Dixon-Coles applied)
Step 4 · Roll a match
Press the button to sample one scoreline from that joint distribution. The simulator does this 103× per tournament (72 group matches + 31 knockouts).
— · —
Rolls
0
Home wins
0
—
Draws
0
—
Away wins
0
—
After enough rolls these should converge to the analytic probabilities shown in Step 3 — that's exactly what the 20,000-sim tournament does, but for every match in the bracket simultaneously.
Step 5 · Group → bracket → champion
Each group plays a 6-match round-robin (every pair, sampled as above). Points: 3 for a win, 1 for a draw. Top 2 of every group plus the 8 best third-placed teams (32 in total) drop into the official FIFA bracket. From there it's straight knockout; a draw at 90' is resolved by a Bayesian-shrunk penalty-shootout skill.
Step 6 · Repeat 20,000× = probabilities
All of the above is one sample of how the tournament could play out. Run it 20,000 times, count how often each team wins the title, and the count divided by 20,000 is the title probability you see on the Champion tab.
Two extra layers fatten the realistic upset tail beyond a textbook Monte Carlo:
- Rating uncertainty · σ = Elo. Before each tournament sim, every team's true strength is resampled from a Gaussian around its rating — a point-estimate model is over-confident in the favourites.
- Travel · rest · altitude. The tired side scores a touch less and concedes a touch more, applied as a differential between the two sides.