README.md
In hoopR: Access Men's Basketball Play by Play Data

player_core parity fixtures

Golden-master oracle for espn_basketball_player_core(), captured from the Python implementation that currently produces the released player_core dataset. The R function is a port; these files are what it must reproduce.

| | | |---|---| | Oracle function | sportsdataverse.nba.helper_nba_player_core | | sdv-py version | 0.0.75 | | Captured | 2026-08-03 | | Payload source | hoopR-nba-raw/nba/player_core/json/{athlete_id}.json (committed raw tree, unmodified) | | Capture command | helper_nba_player_core(json.load(f), athlete_id=<id>) per payload, pl.concat, write_csv |

10.json, 1000.json, 1011.json — ESPN core-v2 athlete payloads, copied unmodified from the raw tree (git may normalise line endings on commit; the parsed JSON was verified identical() to the raw-tree originals). Not hand-written: a hand-made payload is the failure mode this fixture exists to avoid — three Savant parsers shipped wrong against synthetic fixtures.
expected_player_core.csv — the oracle. 3 rows x 35 columns.

They are not a random sample; each covers a branch of the projection:

| athlete | covers | |---|---| | 1000 Brendan Haywood | college and draft present — the fully-populated path | | 1011 Raul Lopez | no college — college_id must be NA, not 0 or an error | | 10 Malik Allen | college but no draft — draft_year/round/selection all NA |

No payload in the 2,577-file tree has fewer than 12 keys, so there is no "sparse" real case to capture; the empty/non-dict input path is covered by a unit assertion in the test rather than a fixture.

35 columns, dtypes as the released dataset:

athlete_id — Int64, the join key into player_box / player_season_stats. Never a float-origin string ("123.0").
age, position_id, college_id, current_team_id, experience_years, status_id, draft_year, draft_round, draft_selection — Int32.
height, weight — Float64.
active — logical.
everything else — character.

Nulls serialize as empty cells (not NA/NaN text), so readr::read_csv() reads them as NA of the column type.

current_team_id is the athlete's team today, not their team in any past season, and height/weight/jersey are a current snapshot — ESPN overwrites them in place, so era-correct bio is not obtainable from this endpoint. The season dimension of the released dataset is participation (who appeared in season Y, from player_box), not the bio's vintage.

Ids come from parsing the $ref URL (/colleges/{id}, /teams/{id}) — the port must never fetch the ref.

Re-capture only when the Python oracle changes deliberately. Re-copy the payloads from the raw tree in the same pass — a refreshed oracle against stale payloads is worse than no oracle.