feat(fixtures): season-2026 practice corpus loader and integrity test (8.0)

[JoelA510]

Summary

Phase 8 task 8.0 (docs/PHASE_8_PLAN.md). Adds a read-only loader and integrity test for fixtures/season-2026/practice/, mirroring packages/core/src/fixtures/season2026{Loader,Parsers}.js and tests/season2026Fixture.test.js. Three supervisor review rounds (8 + 7 + 5 items) are folded in.

• packages/core/src/fixtures/season2026PracticeParsers.js — pure parsers, one .strict() Zod schema and one exact column contract per file (13 files, not the prompt's "seven"), a frozen finding-code table (28 codes) with a frozen severity table, one duplicate detector (noteDuplicate) for all five keyed files, compareDecoderRings() comparing the first occurrence of a code on both sides and carrying a kind per disagreement. Excel-date and unparsed availability rows are judged from the data, with the source's label read beside it. No Date construction; dates are YYYY-MM-DD, times are minutes past midnight; weekdays via availability/calendar.js's civil-date arithmetic; day names via a Map.
• packages/core/src/fixtures/season2026PracticeLoader.js — node:fs IO via the game loader's own helpers, the cross-corpus join, a deep-frozen result with findings, findingsByCode (every code, zero included), seasonLongClosures (all-day, all-fields rows covering ≥ half the days from first to last scheduled date), timeBoundedClosures (the same date test but a daily window — reported apart, never promoted) and meta (rowsRead / rowsParsed per file, plus what each check examined). The game corpus is the single producer of the season year; dir defaults season to the game corpus beside it and refuses one from anywhere else. One ALIAS_RESOLVES_TO_CLOSED_VENUE per season-long closure of the alias's venue.
• tests/season2026PracticeCorpus.test.js — 49 cases: read-only guarantee, every README/plan figure derived at test time, positive controls, including one that raises all three blocking codes beside an assertion on the severity table's blocking count.
• tests/reasonCodeReachability.test.js — the vocabulary is registered as audited. The corpus itself emits 22 of the 28 codes; the six it does not carry (AVAILABILITY_UNPARSED, PERMIT_DAY_MISMATCH, PERMIT_RESERVATION_OUTSIDE_SEASON_YEAR, CHANGE_LOG_DAY_MISMATCH, DUPLICATE_DECODER_CODE, PRACTICE_TEAM_NOT_ON_ROSTER) are driven from public parsers fed one- or two-row files. Header counts 19 / 424 / 413 / 11 are read back by the test itself.
• fixtures/season-2026/practice/README.md — the source's "65 teams" claim kept visible, the derived 44 / 53 added with where each is computed; 6 / 22 / 136 marked unreconciled. Two seams in the game-corpus modules: computeFixtureChecksums(dir, files) and export const trim. No CSV modified.

Claims table

Claim	Result	Actual
Seven files (plan prompt)	DID NOT HOLD	13 CSVs; all parsed
457 practice rows across 7 source sheets	HELD	457 / 7
88 distinct teams hold a slot	HELD	88
Exactly one practice team (16BSelect02) plays no game	HELD	PRACTICE_TEAM_PLAYS_NO_GAME = [16BSelect02]
65 game-playing teams hold no practice slot	DID NOT HOLD	44 from the roster (ROSTER_TEAM_HOLDS_NO_PRACTICE); 53 from every named side of combined_schedule.csv (computed in the test). README corrected
Two slot regimes, no third	HELD	45 @ 16:00/16:45/17:30; 60 @ 16:00/17:00/18:00
Practices Mon–Fri, 19 Friday rows	HELD	FRI = 19
201 coach registrations, 19 naming a second player	HELD	201 / 19
1153 players, 29 playing up	HELD	1153 / 29; no birth year ≥ 2026
13 constraint rows, 3 venues closed effectively all season	HELD	13; Cedarbrook Park, Fivepines Park, Quarrywood Park — all-day (00:00–23:00) all-fields rows covering 85/85, 71/85, 85/85 of the scheduled season's days (threshold ½); no time-bounded multi-day rows; the other all-fields rows are single days
4 permits, 767 reservations, 2026-08-10 to 2026-12-20	HELD	4 / 767; all in the season year; lead-in/tail counted in meta.examined.reservationsBeforeSeason/AfterSeason
14 inventory, 27 code names, 42 availability rows	HELD	14 / 27 / 42
Decoder rings share 20 codes, disagree on 12	HELD	20 / 12 = 11 label conflicts + 1 blank-vs-label (11v11 Field 2), asserted as such
Disagreements = 7v7 Field 1, 9v9 Field 1, Junior Field 1–7	DID NOT HOLD (incomplete)	those 9 plus 9v9 Field 2, 7v7 Field 2, 11v11 Field 2
Two 142-row sheets differ only in being a makeup week	HELD	row-for-row identical bar source_sheet
28 rows carry venue = (unresolved)	HELD	28, all in Select OLD; named finding, rows kept
22 select_coaches / 9 field_equipment / 167 game_change_log rows	HELD	22 / 9 / 167
Co-coach split 71/24/29/9/3/65	HELD, column 1 only	column 2: 10/3/4/7/177
"7v7 Field 1" → Cedarbrook Park, Offline 08-01..11-28	HELD	ALIAS_RESOLVES_TO_CLOSED_VENUE = 1 (one closure, one finding)
Eight aliases → Maplewood, closed 2026-10-23	HELD	8; Maplewood all-fields rows 09-24 and 10-23
Summit HS permit attaches Field Lights	HELD	all 39 PERMIT-04 rows, no other permit
Every practice team_code resolves in the game corpus except 16BSelect02	HELD	PRACTICE_TEAM_NOT_ON_ROSTER = 0
Every registration person_key on ../coach_roster.csv or minted	ASSERTED	coach_registration 166 / 35 minted; select_coaches 15 / 7 minted
Excel-date / unparsed availability rows reported, raw retained	HELD	15 (detected from the data; label agreement stated), 0 unparsed (judged from the data: every empty window is accounted for by unavailable or competitive-programme); 2026-MM-DD ⇒ (MM+12):00–(DD+12):00 on every row
6 venues / 22 fields / 136 team codes	UNRECONCILED	game corpus: 7 venues, 24 field ids, 132 roster teams. 136 = 131+5 or 132+4; 6 and 22 have no found reading

Findings the corpus README does not state

• select_coaches.csv vs ../coach_roster.csv, both directions: 8 sheet rows name a person the roster does not have on that team (7 minted + teagan hobbes); 1 names a rostered coach at another slot (emerson crane, 19BSelect01, sheet slot 1 / roster slot 2 — slot is a clash-breaker, not a role); 8 roster coaches of Select teams are omitted by the sheet.
• Venue names: practice says Maplewood, the game corpus Maplewood Back / Maplewood Front; Larkfield Green hosts no game. VENUE_NOT_IN_GAME_CORPUS = 33 across all 9 venue-bearing files (list derived from the column contracts).
• field_constraints.csv Gardening Day row has fields = 2026-01-07 (Excel-corrupted 1-7).
• 9 coach-registration player birth years are 2026.
• Duplicates: verity bramteford registered twice; players imre hesgaley, ilse wyndafield; Willowmead Park inventoried twice.
• 9 named registration players have no player_registration.csv row; of 10 unresolvable co-coach keys, 7 are registered players' keys (CO_COACH_KEY_IS_PLAYER) and 3 nowhere.
• 11v11 Field 1 carries a label but no venue (DECODER_RING_ALIAS_VENUE_BLANK); 11v11 Field 2 carries neither. confirmed is empty on all 27 code-name rows and is carried on every disagreement.
• Not changed (operator decision, carried in the progress log): apparent real organisation and place names in game_change_log.csv matchup cells.

Positive controls (assertions proven able to fail)

Inline, kept as tests: header-only and full extra column, long row, empty file, unknown enum value, unknown record key, non-ISO and impossible calendar dates (Feb 30), a third slot regime, a reconciled decoder ring (11 < 12), a duplicated alias row (still 20 / 12 plus DUPLICATE_DECODER_CODE), a duplicated code-name row on the fields side in both orders (agree-with-first → 0, differ-with-first → 1; a last-wins index gets both wrong), unparsed and unlabelled date-shaped availability rows, an empty unlabelled window (AVAILABILITY_UNPARSED) against unavailable / competitive-programme (not), a permit on the wrong weekday, a 2027 reservation (finding) and a 2026-01-05 one (figure, not finding), a change-log note on the wrong day, birth year without a season year for both registration parsers (refuse), a 59-day mid-season closure (season-long) vs a 60-day January one (not), the 2026-09-05..11-28 16:00–19:00 all-fields row (time-bounded, not season-long, no alias finding), two season-long closures of one venue (two alias findings; corpus stays at one), a sheet with one team's rows removed (its three roster coaches reported), 'constructor' / '__proto__' as day names, a season from a non-sibling directory (refused), and one control that raises all three blocking codes beside the severity-table count.

Executed against corrupted copies of the corpus in a scratch directory (removed), across the rounds: dropping a grid row, a non-roster team code, Rookery→Rookerie, an unparsed row, resolving one (unresolved) venue, shortening the Cedarbrook closure, Friday→Thursday, 16BSelect02→16BSelect01, a minted select coach on a roster key, an added column, a new class value, a wrong-weekday reservation, an edit inside the 142-row makeup sheet, a 30-minute slot, a venue misspelt only in the equipment sheet, a wrong change-log note, a duplicated decoder code, a confirmed mark, dropping a team from the sibling coach_roster.csv, every reservation re-dated to 2027 beside the 2026 corpus, dropping one select_coaches row, swapping 19BSelect01's slots, filling the 11v11 Field 1 venue, a player born 2026. The noteDuplicate generalisation was proven by a byte-for-byte comparison of the serialised corpus finding list before and after (111,484 bytes, identical).

Verification

Statically reviewed: /code-review at high and medium (11 items, 10 fixed, 1 handled by carrying confirmed rather than resolving on it); supervisor rounds 1, 2 and 3 (20 items, all fixed).

Executed (actual output at head 7ce44d9):

npm run lint            0 errors, 1 warning (baseline: 0 errors, 1 warning)
npm run typecheck       0 errors
npm run test            before (origin/main b048022): 2165 passed | 34 skipped | 6 todo (2205), 158 files
                        after: 2216 passed | 34 skipped | 6 todo (2256), 159 files
npx vitest run tests/season2026Fixture.test.js tests/season2026PracticeCorpus.test.js tests/reasonCodeReachability.test.js
                        110 passed (34 + 49 + 27)
npm run frontend:build  clean
npm run check:advisors  PASS
npm run check:bundle    PASS
npx prettier --check fixtures/season-2026/practice/README.md   clean

Test-verified: everything in the claims table and the inline controls above. The corrupted-copy controls were executed, not kept as tests.

Left open, deliberately: the 65 / 6 / 22 figures (unknown derivation, now stated in the README); the change-log organisation names (operator); whether used_for / remainder on field_code_names.csv should ever be load-bearing (retained as record data); what a time-bounded season closure means for practices (reported as timeBoundedClosures with its window; the corpus has none).

🤖 Generated with Claude Code

https://claude.ai/code/session_011ppMMPnorTKrb9ZVaGTpS9

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
squadlogic	Ready	Preview	Sep 4, 2026 9:20pm UTC