TWO ERAS. ONE PITCH.

Historical teams. Hypothetical matches.

Choose club seasons from the Premier League, La Liga, Bundesliga, Serie A and Ligue 1, or national-team editions from the men's World Cup, Euros and Copa América since 2010. Club years refer to the season-ending year: Manchester City 2023 means 2022/23, not the calendar year. National teams use their tournament edition.

Where the strengths come from

Real scoring and conceding totals are normalized against the competition's scoring environment, with an eight-match average-team prior to reduce small-sample extremes. A neutral 1.35-goal baseline combines one team's attack with the other's defense. Home advantage is optional. Rates are capped between 0.15 and 4.5 goals per side. Cross-league and club-versus-country comparisons assume equal competition strength; they are experimental, not calibrated forecasts.

Measured accuracy

Loading the held-out model benchmark…

New match-lab controls

Six formation layouts shape suggested XIs. Five tactical approaches apply transparent heuristic multipliers to goal rates; they are not learned tactical effects. The forecast grid and batch experiments use the pre-match rates, while the scoreboard probabilities condition on score and time. Player ledgers and key moments reflect only the current replay prefix. Saved fixtures and completed results stay in this browser’s local storage.

What is simulated

A seeded generator samples shots, shot quality, goals and passes. Historical squad snapshots and their recorded coaches are shown in the builder. A suggested XI is selected by positional availability and can be edited. Simulated actions are assigned to eligible selected players using position-based heuristic weights, not historical player ratings. The engine does not reconstruct an actual match, injury situation or manager's tactics. Missing rosters retain anonymous event units. The same team editions, venue, model version and seed reproduce the same events. A new seed produces a new possible match. Matches finish at 90 minutes; extra time and penalties are not simulated.

Scouting models

The Scout lab adds two models fitted offline by scripts/build_scouting.py from this repository's own data, with no network access and no third-party model. A ranker predicts which club a player moves to next, trained on the club-to-club moves the squad snapshots actually record. A classifier projects which calibre of squad list a player first seen at 21 or younger appears on two to six seasons later. Both are validated on a forward split in time: earlier seasons fit the weights, later seasons are held out, and the Scout lab's model card publishes the held-out numbers next to the baselines they are measured against. A club's profile for a destination season describes that club the season before, so no candidate carries information from the season a prospect would be arriving into.

These two targets are narrower than the words "scouting" and "potential" suggest, and the Scout lab's model card states the gap plainly. A squad list records who a club registered, never who played, so the progression target is which calibre of list a career later appears on. A teenager who stays registered at an elite club for years without playing reaches the top class exactly like one who starred there. Being at a strong club at 18 is therefore the strongest single predictor of being on a strong club's list at 21, and the highest projections cluster on big-club academy intakes for that reason rather than by insight. The ranker, likewise, predicts recorded destinations with loans included, which this source cannot distinguish from permanent transfers.

There is no academy data in this project. The pool is every under-21 player in the bundled senior-squad snapshots, which is the youngest cohort these sources can observe, not a youth-team intake. Squad lists carry no minutes, fees, valuations, injuries or scouting grades, so none of those inform the models. Fit is learned from where players actually moved, which reflects the recruitment market rather than merit. A player who leaves these five leagues becomes unobservable and is reported as censoring, never as failure. Hand-entered prospects are scored from the attributes you supply, which the model cannot verify.

Squad coverage

Loading historical squad coverage…

Source of Material is http://www.footballsquads.com. Material: © FootballSquads.com, 1999 - , All Rights Reserved. Used for this non-commercial project under the source’s terms of use. A listed coach is a source snapshot, not a complete managerial history.

Source coverage

Club results: OpenFootball and FootballCSV / Soccerverse. International results: Mart Jürisoo. Source datasets are CC0; local download hashes and result cutoffs are included in the catalog.