Quiz lint cleanup (rule 8 surface-form duplicates + apiStore GET cache assertions)

SummaryA small cleanup sprint clearing the two optional follow-up items deferred from Sprint 218. (1) check-quiz-content.mjs rule 8 (post-normalization duplicate WARN) compared acceptedAnswers after normalizeAnswer stripped all whitespace and punctuation, so it flagged 156 intentional surface-form variants such as '연결리스트'/'연결 리스트' and 'btree'/'b tree'/'b-tree' (whitespace/hyphen presence) as noise. The comparison key was switched to caseFoldKey (NFKC→lowercase→trim, preserving inner whitespace/hyphens/punctuation) so the rule only catches real duplicates that differ by case or leading/trailing whitespace. 156 WARN → 0. The narrowed rule immediately caught one genuine duplicate (db-22 '2PL'/'2pl', case-only), which was cleaned up. (2) Sprint 218 Critic herald P4 follow-up — the authenticated-path integration test (page.test.tsx Scenario 2) asserted only the saveResult POST and not the getBest-triggered GET/cache behavior; this was shored up with nth-call ordering assertions on finish() plus toHaveBeenCalledTimes(2), and a getBest cache-hit assertion was added to api-store.test.ts. User decision: rule 8 basis is 'case-insensitive + punctuation/whitespace preserved' (intentional variants preserved). No jest unit test for the lint script (jest roots=src/ only, scripts/*.mjs out of scope — keeping the 216/218 convention, node-run verification). 4 atomic commits. Verification: frontend tsc 0 / content-lint --strict exit 0 (rules 8 & 9 zero) / 147 suites 1498 tests PASS · global lines 87.6% · branches 78.96% gates met / next build ✓ /[locale]/quiz 37.5kB (bundle unchanged) / next lint 0 errors. Critic single round CLEAN. Code-side cleanup only — independent of live deploy. start commit 1813a5f.

Date2026-06-03

ImpactCritical

Goal

Clear the two items deferred as optional follow-ups from Sprint 218.
Remove rule 8 noise: narrow check-quiz-content.mjs rule 8, which flagged 156 intentional surface-form variants as noise, so it only catches genuine duplicates.
Shore up apiStore GET cache assertions (Sprint 218 Critic herald P4): fill the GET/cache assertion gap in the authenticated-path integration test.
Code-side cleanup only — independent of live deploy (separate from the deferred operational migration:run).

Background

Rule 8 flagged intentional surface-form diversity as noise

The content lint rule 8 added in Sprint 218 (checkNormalizedDuplicates) WARNed any acceptedAnswers entries that became identical after normalizeAnswer (NFKC → lowercase → strip non-alphanumeric/Hangul → strip all whitespace). But all 156 current WARNs were whitespace/hyphen surface-form pairs like ['연결리스트','연결 리스트'], ['btree','b tree','b-tree'], ['logn','log n'].

Because normalizeAnswer strips whitespace and punctuation, these synonyms all match against the same grading key — but in the content they are a legitimate case: both surface forms are shown for readability since people type them differently. Rule 8 reported this intentional diversity as duplication, emitting 156 lines of noise on every run despite being WARN-only (--strict exit 0, no CI block).

apiStore GET cache assertion not shored up (Sprint 218 Critic herald P4)

page.test.tsx's authenticated-path Scenario 2 (completes game via apiStore) asserted only the saveResult POST, not the GET triggered by the preceding getBest nor its cache behavior (no redundant re-GET during result-screen render). Sprint 218 Critic deferred this as P4 (non-blocking).

Decision

D1. Rule 8 basis — case-insensitive + punctuation/whitespace preserved (user-confirmed)

Switch rule 8's comparison key from normalizeAnswer (strips all whitespace/punctuation) to caseFoldKey (NFKC → lowercase → trim, preserving inner whitespace/hyphens/punctuation). As a result:

Whitespace/hyphen surface-form variants like '연결리스트'/'연결 리스트', 'btree'/'b tree'/'b-tree' become distinct keys and pass (synonym diversity preserved).
Genuine duplicates differing only by case or leading/trailing whitespace like 'stack'/'Stack', '2PL'/'2pl' become the same key and are caught as WARN.

156 → 0 on the current data. The user chose this over "strict exact-string" (case-sensitive too) — to keep a safety net for future case-only duplicates.

D2. No unit test for the lint script (Oracle call)

No jest unit test is added to check-quiz-content.mjs. jest roots only includes src/ and transform handles only .tsx?, so scripts/*.mjs is out of scope. We keep the Sprint 216 (lint introduced) / 218 (rules 8 & 9 added) convention of node ... --strict run verification. CI (ci.yml:427) gates with --strict, and this sprint verified caseFoldKey/checkCaseFoldedDuplicates behavior directly via node -e (case duplicate WARN / whitespace difference passes).

Implementation

Branch chore/sprint-219-quiz-lint-cleanup, start commit 1813a5f, 4 atomic commits. Wave A (Curator lint/content) → Wave B (Herald tests) → Wave C (Librarian ADR).

Commits

Hash	Content
`a087ba9`	chore(frontend) — Curator: narrow `check-quiz-content.mjs` rule 8 to surface-form duplicates — switch to `caseFoldKey` (NFKC→lowercase→trim) comparison key, rename `checkNormalizedDuplicates`→`checkCaseFoldedDuplicates`, update header comment & WARN message. 156 WARN → 0
`9cf0dc2`	fix(frontend) — Curator: remove db-22 `acceptedAnswers` `'2PL'`/`'2pl'` case-only duplicate (genuine duplicate found by the narrowed rule 8)
`2783db5`	test(frontend) — Herald: `page.test.tsx` Scenario 2 GET cache assertions + `api-store.test.ts` getBest cache-hit assertion
(ADR)	docs(adr) — Librarian: sprint-219 KR+EN + README index 156→157 · range 62~~218→62~~219

Wave A — Curator (lint + content)

Narrow rule 8 (a087ba9): add a caseFoldKey(raw) helper — raw.normalize('NFKC').toLowerCase().trim(). It preserves inner whitespace/hyphens/punctuation, so intentional variants remain distinct keys. Switch checkNormalizedDuplicates's grouping key from normalizeAnswerJs → caseFoldKey; since the meaning shifts from "normalization duplicate" to "case/leading-trailing-whitespace-only duplicate," rename the function to checkCaseFoldedDuplicates. Update the file header (rule 8 definition/examples), WARN message, the checkWarnOnlyRules caller, and the output label consistently. Rule 9 (checkEmptyNormalized) and normalizeAnswerJs are kept — rule 9 still uses normalizeAnswerJs (whitespace/punctuation stripped) to check "empty after normalization," so normalizeAnswerJs remains exported. 156 WARN → 0.
Clean up the db-22 genuine duplicate (9cf0dc2): the narrowed rule 8 caught exactly one real duplicate — ['2PL','2pl'] (db-22) — case-only, so normalizeAnswer's lowercasing already grades them identically. Keep the standard abbreviation '2PL' and drop '2pl'. A case where narrowing the rule removed noise and simultaneously surfaced one real cleanup target.

Wave B — Herald (apiStore cache assertions)

page.test.tsx Scenario 2 GET cache assertions (2783db5): the finish() flow is store.getBest() (1 GET + memory cache) → store.saveResult() (POST + cache invalidate). The existing POST assertion was strengthened into nth-call ordering — 1st = option-less GET (getBest→fetchAllBest), 2nd = POST (saveResult). toHaveBeenCalledTimes(2) pins that the GET result is cached so there is no re-GET during result-screen render. (Scenario 2 has empty localStorage, so merge-up makes 0 fetchApi calls — only finish()'s GET+POST occur.)
api-store.test.ts getBest cache hit (2783db5): previously only getAllBest cache-hit was covered. Asserted directly that calling getBest twice with different keys reads from the cached map, so fetchApi fires only once.

Wave C — Librarian (ADR)

sprint-219 KR+EN + docs/adr/README.md index 156→157 · range 62~~218→62~~219.

Verification

Confirmed directly by Oracle (frontend):

tsc --noEmit → 0 errors
check-quiz-content.mjs --strict → exit 0 (rules 1–7 pass, rules 8 & 9 zero — 156 WARN removed + db-22 cleaned)
node -e direct check — caseFoldKey: stack/Stack same key (✓), 연결리스트/연결 리스트 distinct keys (✓), btree/b-tree/b tree all distinct (✓), leading/trailing whitespace folded (✓). checkCaseFoldedDuplicates: 1 case-duplicate WARN, 0 for whitespace difference (✓).
jest --coverage → 147 suites · 1498 tests PASS / 0 fail (218's 1497 + getBest cache 1), global lines 87.6% (gate 83) · branches 78.96% (gate 71)
next build → ✓ Compiled, ƒ /[locale]/quiz 37.5kB (same as 218 — script/test-only, bundle unchanged)
next lint → 0 errors (only pre-existing sidebar/sonner/useAutoSave warnings, unrelated to quiz)

Lessons Learned

WARN-only lint rules must surface only real risk without blocking intentional surface-form diversity — when rule 8 used the grading normalization (normalizeAnswer, whitespace/punctuation stripped) for duplicate detection, it flagged readability-oriented intentional synonym pairs as noise. Separating a comparison key fit for the check's purpose (caseFoldKey, surface-form preserving) lets intentional variants pass while catching only genuine duplicates like case-only ones. Grading normalization and duplicate-detection normalization have different purposes, so keep them separate.
The narrowed rule immediately found one genuine duplicate — once the 156 noise lines were removed, the previously buried db-22 '2PL'/'2pl' case duplicate surfaced. Evidence that noise had been masking signal — refining a rule reduces false positives while raising the visibility of true positives.
Respect test-infra boundaries to avoid scope creep (D2) — scripts/*.mjs sits outside jest's roots (src/), so adding a unit test would entail a jest config change. Following the 216/218 convention with node run + node -e direct verification kept a small cleanup sprint from sprawling into infra work.

Critic cross-review

R1 — CLEAN (Codex gpt-5.5, codex review --base 1813a5f -c model=gpt-5.5, codex-cli 0.130.0, session 019e8c2b-1feb-7cd0-ad56-c32cec2494ae)

Critical / High / Medium / Low 0.
"The changes are limited to the quiz content warning logic, one redundant accepted answer removal that remains covered by normalization, and test assertions. I did not identify any introduced correctness, security, performance, or maintainability issue that warrants an actionable finding."

Final — CLEAN. Single round, 0 actionable findings.