How the ratings work

Three open, legal sources, no scraping of restricted sites. Squads and memberships come from Wikidata. Per-season appearances, goals and (for keepers) clean sheets are parsed from the "career statistics" tables on each player's Wikipedia article via the official MediaWiki API. Team defensive records (clean sheets, goals conceded, final table position) for the big-five leagues from 1993 come from football-data.co.uk. Text facts only: no images, crests, or kits. Wikidata/Wikipedia content is CC BY-SA.

A player's rating reflects how good they actually were that season, not how famous they are. For each season we combine three signals, weighted by position:

1 · Availability.What share of the season's games the player started. A regular starter outranks a fringe squad player.

2 · Output. Goals and assists per appearance for attackers and midfielders, benchmarked by position; clean-sheet rate for goalkeepers. A forward is judged on production, a keeper on shutouts.

3 · Team quality.For defenders and keepers, who do not score, the club's actual defensive record that season (clean sheets and goals conceded) plus its league finish. A regular on a title-winning defence rates high on real evidence.

Position matters: forwards lean on output, defenders and holding midfielders on availability and team defence, attacking midfielders on output. Age-at-season (peak 24-30) and a small cross-league adjustment apply on top.

Fame is the last resort. The old model rated everyone primarily by how widely documented they were. Now that number (Wikipedia language editions) is used only as a fallback when a player has no stats at all, so obscure players are not left blank. Around 150 all-time greats also carry a hand-set peak rating that lifts them toward it in seasons they played regularly, and caps a genuine off or fringe season below it.

An XI becomes two numbers: attack (forwards weigh most, then midfield, then defence) and defence (goalkeeper and back line weigh most). Chemistry — the share of player pairs that link by club, nation, or era — multiplies both by up to 6%.

Matches are simulated with a Dixon-Coles double-Poisson model, the standard in football analytics: expected goals scale exponentially with the attack-vs-defence gap, home advantage is worth about a third more goals, and a correction term keeps low-scoring draws realistic. Every match is drawn from a seeded random stream, which is why any season can be replayed identically from its seed — the "verify fairness" button does exactly that.

Ratings are opinions computed from public signals, not measurements. We use only stats that were actually recorded across the whole era: appearances, goals, assists, clean sheets, team defensive records. Per-player tackles, dribbles, passing and expected goals do not exist for most of football history (they were only collected from around 2017, and only behind paid or restricted sources), so we do not fake them. This makes outfield defenders and holding midfielders coarser than forwards and keepers.

Where a player's Wikipedia table has no row for a season, the rating falls back to their club spell, then to fame. Year-precision transfer dates can bleed a player into a neighbouring season at their old or new club. football-data.co.uk covers the big five from 1993, so the J-League and the early 1990s lean on availability alone for defenders. Corrections are welcome — see about & contact.

The aggregate squad ratings are published as an open dataset on the open data page (CC BY-SA, attribution: 38nil.app).