📊 2026 World Cup · Data Methodology

• Official match feed: FIFA-licensed real-time event data stream (25 fps)

• Optical tracking system: 12 high-speed cameras per stadium, recording player/ball coordinates (x,y) at 25Hz

• Manual verification: Key events (goals, red cards, penalties) verified by at least two independent analysts

• Data partners: Opta / StatsBomb / CSL Data Lab

• Missing value handling: Kalman filtering for trajectory interpolation; event gaps filled via video review

• Multi-source alignment: Synchronize optical data with referee signals, millisecond timestamps

• Outlier removal: Physical-implausible sprint/speed records automatically flagged and reviewed

• Timezone normalization: All timestamps stored in UTC, front-end displays localized to user’s timezone

xG (Expected Goals) — Probability of a shot resulting in a goal based on distance, angle, defensive pressure, etc. See xG model for details.

PSxG (Post-Shot xG) — xG adjusted for goalkeeper save ability, used to evaluate keeper performance.

Shot conversion rate = Goals / Total shots (excl. blocked)
On-target conversion rate = Goals / Shots on target

Possession percentage — Share of total passes per team (excluding clearances, throw-ins).

Pass success rate = Successful passes / Total pass attempts (forward passes weighted higher)
Progressive passes — Passes that advance the ball toward the opponent's goal by at least 10 meters.

PPDA (Passes Per Defensive Action) = Defensive actions in opponent's half / Opponent passes in that zone.

Tackle success rate = Successful tackles / Total tackle attempts
Interceptions — Passes cut out (non-tackle)
Clearances — Balls kicked away from dangerous zones
High press success rate — Percentage of recoveries or forced errors in opponent’s half

xPts (Expected Points) — Simulated points based on per-match xG and xGA; measures "luck".

ELO rating — Dynamic strength rating adjusted for opponent difficulty: R_new = R_old + K * (actual - expected).

Final third touches — Touches inside opponent's final third (including wide areas).

Logistic regression using live ELO, last‑5 form index, injury weight, and home advantage:

Parameters refitted daily to capture latest momentum.

10,000 simulations of remaining fixtures based on current standings and match probabilities.

• Group ranking strictly follows FIFA tiebreakers: points → GD → H2H → fair play.
• Knockout outcomes sampled from AI prediction engine’s probability distribution.
• Penalty shootouts modeled using historical World Cup data (player conversion rates + keeper tendencies).

Unsupervised K‑means clustering categorizes teams into 4 strength tiers for draw simulation and visualisation.

Feature vector includes: ELO, last‑10 xG differential, key passes, defensive resilience.

Hyperparameters (e.g., weighted learning rate) optimised via Bayesian methods as the tournament evolves.

Also applies Bayesian smoothing to “random” events like woodwork hits or deflections to reduce small‑sample bias.

• Raw data sourced exclusively from FIFA‑authorized providers.
• Every aggregate metric includes 90% confidence intervals to avoid “false certainty”.
• Backtesting against the last three World Cups shows 74% accuracy in projecting semi‑finalists.
• Open‑source validation: core metric definitions are publicly available on GitHub.

• Inherently unpredictable factors: locker‑room atmosphere, referee bias, late‑breaking injuries.
• Last‑matchday “collusion” in group stage cannot be reliably modeled.
• Individual player explosion / collapse (e.g., goalkeeper heroics) not pre‑captured.
• Extreme weather (heavy rain) impact on xG not yet fully incorporated, but flagged during knockout stage.