How we compute the AI Score

The AI Score is a synthetic assessment of a player's quality across 7 dimensions (technical, physical, mental, tactical, potential, durability, consistency), expressed on a 0 to 100 scale, with a confidence band.

Inputs

For each player we collect: season statistics (goals, assists, minutes, cards), match-by-match statistics (per-match rating, shots, passes, tackles), injury history, declared role and effective role from lineups, recent transfers, and football-relevant news.

Model

The structured data is composed into a prompt and passed to a Large Language Model (Claude from Anthropic in production, Ollama locally as fallback). The model returns a score for each of the 7 dimensions and a confidence score that measures how solid the data foundation is. Responses are validated via JSON Schema; out-of-range values are discarded.

Confidence band

Players with few recent matches, missing data, or very recent transfers receive a lower confidence score. Below a minimum threshold the score is not displayed, to avoid suggesting a precision we can't guarantee.

What the AI Score is NOT

It is not a prediction of future performance: it summarises observed performance and structural signals at compute time. It does not replace a scout's judgement. It is not optimised for fantasy football (a separate calculation handles league-specific bonuses). It is not betting advice.

When we recompute

Scores are recomputed nightly after the day's stats refresh. Players inactive for over 30 days are skipped to avoid wasting API quota.

Methodology version: v1, last updated April 2026. Any formula change will be flagged here.