Prediction Methodology

Our prediction model draws from multiple authoritative data sources:

• FIFA Official Rankings — Updated monthly, used as the baseline for team strength comparison.
• Recent Match Results — Last 20 international fixtures per team, weighted by recency (more recent matches carry higher weight).
• Tournament History — Historical World Cup performance data from 1930 to 2022, including knockout round progression patterns.
• Squad Composition — Player availability, club performance levels (Champions League, top-5 European leagues), and injury reports.
• Head-to-Head Records — Historical results between specific matchups, weighted by relevance (competitive matches valued higher than friendlies).

Our ensemble model combines five weighted factors to produce a composite prediction:

For each match, our ensemble model runs 10,000 simulated outcomes using a Monte Carlo approach. Each simulation accounts for:

• Scoring rate estimation — Based on each team's expected goals (xG) for and against in recent matches against opposition of similar quality.
• Match state dynamics — How each team performs when leading, trailing, or level, based on historical data.
• Knockout adjustments — Extra time and penalty shootout probabilities for elimination matches beyond the group stage.
• Home advantage — Adjusted for host nations USA, Mexico, and Canada, plus proximity-based crowd support for other CONCACAF nations.

The win probability you see on each match page is the percentage of those 10,000 simulations in which the named team advanced (or won, for group stage matches).

We track prediction accuracy against actual match results. Our model is calibrated using:

• Back-testing — Retrospective predictions against known tournament results (2018 and 2022 World Cups, 2021 and 2024 Euros, 2024 Copa America).
• Brier Score — A proper scoring rule that measures the accuracy of probabilistic predictions. Lower scores indicate better calibration.
• Ranked Probability Score (RPS) — Evaluates how well our win/draw/loss probability distributions matched actual outcomes.

Last model update: June 9, 2026. We review and recalibrate model weights monthly during the competitive season and weekly during the tournament.