Skip to content

README.md

Latest commit

 

History

History
126 lines (103 loc) · 6.41 KB

README.md

File metadata and controls

126 lines (103 loc) · 6.41 KB
Raw

Football 11 Prototype Data

Active source

The active catalogue uses the complete 2023/24 English Premier League results and all 20 club squad tables from OpenFootball England, plus player-profile matches from OpenFootball Players. Both repositories describe their data as free and public domain. The exact input revisions are pinned in the generated manifest and adapter:

data/content/openfootball-pl-2023-24-named-squads-v2.json

The source contains 380 results and 1,041 named squad entries across 20 clubs. Player name, club membership, shirt number, broad position, and 1,011 birth years are observed. OpenFootball Players supplies an unambiguous height match for 538 entries; missing heights and 30 missing birth years use deterministic position-population models. The source has no player appearances or event statistics, so those remain zero rather than being invented.

Rating model

The adapter derives percentile indices within each season from points, goals for, goals against, clean sheets, goal difference, second-half goal difference, home/away results, and match-to-match consistency. It applies position-specific weights: forwards and wingers emphasize attack, defenders and goalkeepers emphasize defence and clean sheets, and midfielders blend control with attack or defence according to role.

The observed GK, DF, MF, and FW values are deterministically mapped to playable sub-positions. Age affects pace, leadership, and distance from a position-specific peak; height affects aerial presence. Position templates, club-season indices, and small seeded variation then produce attributes and raw ratings. Primary raw ratings are a 55/45 blend of club-role quality and relevant attributes, with an age-curve penalty; secondary positions receive a four-point step-down. A monotonic, piecewise-linear calibration then preserves the 55-point floor and middle of the pool while stretching the strongest modeled cards into a scarce 90–94 elite band.

In the frozen v2 catalogue, the median best-position rating is 72, the 95th percentile is 86, 14 of 1,041 players (1.34%) reach 90 or higher, and the maximum is 94. Automated generation tests freeze those distribution gates. The curve improves game differentiation; it does not add player-performance evidence or turn these values into real-world scouting claims.

The catalogue conforms to football11-player-catalogue-v1. The observedStatsBombPositions field is a legacy v1 property name; OpenFootball cards label the observed squad position explicitly.

Public-domain notice and remaining rights

Public-domain data status does not independently settle player publicity rights or grant rights to league or club names, crests, kits, or other marks. A public release using real player identities and club/competition identifiers still needs explicit legal review.

football-data.org remains unsuitable as a checked-in free source: its API is subscription-bound, its published free service is non-commercial, and its terms restrict continued use after cancellation.

Rebuild

From the repository root, clone and check out the two pinned inputs in the ignored cache, then regenerate and validate:

git clone https://github.com/openfootball/england.git .cache/openfootball-england
git -C .cache/openfootball-england checkout 0690446f794fde748ea4b994244def699c6a65b2
git clone https://github.com/openfootball/players.git .cache/openfootball-players
git -C .cache/openfootball-players checkout 125d20f7cc06cac7e758b40df535a7695632680a
python tools/openfootball/prepare_openfootball.py
python -m unittest discover -s tools/openfootball -p "test_*.py" -v
python tools/catalogue/validate_catalogue.py
python tools/phase0/prepare_phase0_fixtures.py
python -m unittest discover -s tools/phase0 -p "test_*.py" -v

The output has no generation timestamp, is sorted deterministically, and stores the source revisions, generator version, seed, public-domain notice, disclaimer, and SHA-256 player hash. Re-running the pinned inputs reproduces it byte for byte.

Phase 0 contract and fixtures

  • schema/player-catalogue-v1.schema.json — machine-readable catalogue shape.
  • config/assessment-v1.json — formation, chemistry, tactical, composite, and team-unit coefficients.
  • config/league-profile-v1.json — nineteen abstract opponents played home and away.
  • fixtures/phase-00-golden-rosters-v1.json — three valid reference XIs and one duplicate-identity rejection case.
  • fixtures/phase-00-draft-cases-v1.json — selection, cooldown, reroll, and forward-feasibility cases.
  • fixtures/phase-00-match-cases-v1.json — match delta and probability vectors.

The v2 rating-only revision preserves every v1 player identity, club, era, and legal position. A refreshed 10,000/10,000 complete-draft analysis found no dead ends and exposed every card at least once. Each offer groups one 2023/24 club squad under the 2020s era constraint.

Offer presentation is configured by the server-owned AGS Game Record football11_game_config_v1. Sites validates and freezes its active revision in every new challenge; the domain accepts 3–15 offers and 0–5 rerolls, while revision 1 uses 15 and 5. Browser generation and trusted replay consume the same challenge snapshot, and no draft-limit environment variables remain.

Current limitations

  • Names and club squads are real 2023/24 records, but exact formation positions are inferred from broad source positions.
  • Missing birth years and heights are modeled; appearances remain unavailable; all attributes and ratings are synthetic.
  • Club results are the only performance evidence; there are no player events, minutes, goals, assists, or defensive actions in the selected inputs.
  • The catalogue materially increases the browser bundle and currently triggers the build's large-chunk warning; catalogue loading/code-splitting is a follow-up.
  • Competitive records created under the retired content version must not share a leaderboard with this differently calibrated catalogue without an explicit migration or new-board decision.