WORLDCUPAI · HOW IT WORKS

The model & methodology

4,200 AGENTS · ELO + POISSON · MONTE-CARLO · OPEN METHOD

The swarm engine

Every probability you see on this site comes from a Monte-Carlo swarm — 4,200 independent simulations of the full World Cup bracket. Each agent perturbs the 48 teams' Elo ratings with Gaussian noise (σ = 58), then plays every match of the 104-match tournament, sampling goal counts from a Poisson distribution informed by each team's attack and defence rate. Aggregate the 4,200 outcomes → you get the title %, final %, semi %, and advance % for every team.

The whole engine lives client-side in src/lib/swarm-engine.ts. Hitting Re-run swarm on the /odds-model page reseeds the random number generator and runs all 4,200 agents again — typically about 4 seconds on modern hardware, with progress streamed in 350-agent chunks so the UI stays responsive.

⚠️ Elo ratings are informed June-2026 estimates, not official FIFA figures. The Poisson scoring is a deliberate simplification — real football has correlated goals, fatigue, red cards, set-piece variance and all sorts of dynamics this model ignores.

Pick'em scoring

Pick HOME, DRAW or AWAY for every match. Locked picks are saved to your account (auth-gated).

Correct group pick
+3
per match
Correct knockout pick
+5
per match
Exact scoreline bonus
+2
stacks with the pick
Champion preseason
+25
locked before kick-off
Data sources
Glossary
Elo rating
A numeric strength index for each team. Higher = stronger. Updates after every match based on the result, opponent strength, and venue. 2026 Elos are informed June-2026 estimates, not official FIFA figures.
Title %
Estimated probability that a team wins the entire tournament, derived from running the full bracket thousands of times with perturbed Elos.
Advance %
Probability of progressing past the group stage.
Final / Semi %
Probability of reaching the Final / Semi-finals respectively.
GD
Goal Difference — goals scored minus goals conceded across all group games. Used to break ties.
xG
Expected Goals — a measure of shot quality. Sum of probabilities that each shot would result in a goal.
Poisson model
A statistical distribution used to simulate match scorelines based on each team's attack/defence rates.
Swarm
The collective output of 4,200 parallel simulations, each with slightly different inputs. Robust against single-run variance.