Current baseline strength uses a FIFA-rank fallback from the imported team dataset; no stable structured feed of official ranking points is connected yet. Form and goal rates use only completed official 90-minute results from this World Cup, limited to the latest 18 months and ten matches per team, and activate after five matches. Score distributions use Poisson with a Dixon-Coles low-score correction.

Completed group matches use real 90-minute scores, while completed knockout matches lock the official advancing team. Only unfinished matches are simulated, and 90-minute, extra-time, penalty, and advancement results remain separate. The engine runs 10,000 Monte Carlo simulations over the remaining tournament.

Only the FIFA official calendar and results feed is currently a structured prediction input. Official ranking, squad, match-centre, and government weather pages are recorded as fact and audit sources and are not claimed to affect probabilities. Facts require manual review or a versioned deterministic rule before activation; media and LLM extraction always enter review first.

As an analysis tool, we recognize that models cannot replace human intuition and experience. Match detail pages now support local adjustments for team form, attack strength, and home context to generate a personal forecast without changing the global baseline snapshot.

No model can perfectly predict the high randomness of a single football match (red cards, penalties, wonder goals). Our models provide baseline probability distributions over large samples to quantify risk, and must never be treated as guaranteed factual outcomes.