CS Quiz Question Pool Major Expansion + 5 New Categories + 'ALL' (Full Random) Filter

SummaryExpanded the CS quiz (/quiz) question pool from 150 to 350 (5 categories × 30 → 10 categories with augmented counts) and added a category 'ALL' (full random) filter. Wave A (infra): QuizCategory enum +5 + Palette added 20 CSS variable lines in globals.css for the 5 new categories (WCAG AA 4.5:1+ verified) + category-meta icon/var mapping. Wave B+C (content): 20 questions per new category + existing 5 categories 30→50 questions (HARD imbalance fix), globally unique IDs totaling 350, generated by 10 parallel Curator agents. Wave D ('ALL' filter): getQuestionsByFilter/getRandomQuestions accept category: 'ALL', QuizStart PillRadioGroup 'ALL' prepended (0 new tokens). Wave E (tests): +22 tests, global 1555 PASS. Lessons: per-file one-agent parallelism, single-quote string apostrophe syntax errors, within() scope queries, meta filter option zero-token pattern. Frontend only, merge ≠ live.

Date2026-06-07

ImpactCritical

Goal

Expand the CS quiz (/quiz) question pool from 150 to 350 (5 categories × 30 → 10 categories, existing 5 categories augmented from 30 to 50 each).
Add 5 new categories (Computer Architecture, Design Pattern, Web, Security, AI) with WCAG AA-verified CSS tokens and icons.
Add an 'ALL' (full random) filter to the category selector so users can practice across all categories at once.
Frontend only, pure data/UI expansion — merge ≠ live (post-redeploy live verification carries over to the quiz-ui-verification runbook).

Background

The CS quiz built from Sprint 215 onward had its UI, accessibility, and UX polished through Sprint 221~224, but the question pool remained limited at 5 categories × 30 questions = 150 items. Users exhausting a single category quickly, and the "shuffle all categories" practice scenario, were both unsupported.

Sprint 227 closes both gaps simultaneously:

Category expansion: Add Computer Architecture, Design Pattern, Web, Security, and AI alongside the existing Network, OS, Database, Algorithm, and Data Structure categories.
'ALL' filter: Add an 'ALL' (full random) option to the category PillRadioGroup, drawing from the full 350-question pool at random.

Decision

D1. 5 New Category Enum, CSS Tokens, and Icons (Wave A)

5 new values added to the QuizCategory enum:

COMPUTER_ARCHITECTURE / DESIGN_PATTERN / WEB / SECURITY / AI

Palette added 20 lines of --quiz-cat-*-color / --quiz-cat-*-bg CSS variables (light + dark) to globals.css. WCAG AA 4.5:1+ contrast verified directly:

Category	Light (fg)	Contrast	Dark (fg)	Hue distinction from existing 5
arch	`#B45309`	4.81	`#FBBF24`	amber
dp	`#BE185D`	5.78	`#F472B6`	pink
web	`#4338CA`	7.56	`#818CF8`	indigo
sec	`#B91C1C`	6.19	`#F87171`	red
ai	`#4D7C0F`	4.78	`#A3E635`	lime

category-meta.ts updated with lucide icons (CircuitBoard / Shapes / Globe / ShieldCheck / BrainCircuit) and var-slug mappings for the 5 new categories.

i18n categories object extended with 5 new keys (ko + en) + categories.ALL.

D2. Large-Scale Content via Parallel Generation (Wave B+C)

10 Curator agents run in parallel, one agent per file:

20 questions per new category (7E/7M/6H): computer-architecture.ts / design-pattern.ts / web.ts / security.ts / ai.ts
Existing 5 categories 30→50 questions (+20 each): HARD imbalance corrected (network HARD 4→17, os 5→17, database 7→17, targeting 16E/17M/17H each)

index.ts ALL_QUESTIONS integrates all 10 categories → QUIZ_CATEGORIES 10 entries, globally unique IDs totaling 350.

Integration-time discovery and fix: 2 unescaped apostrophes in English text inside single-quoted strings (Brewer's, process's) in database.ts and os.ts → converted to double-quoted strings to eliminate syntax errors.

D3. 'ALL' Filter Addition (Wave D)

index.ts getQuestionsByFilter / getRandomQuestions now accept category: QuizCategory | 'ALL':

'ALL': draws from the full ALL_QUESTIONS pool (mirrors the existing difficulty 'ALL' pattern)
Per-category filtering logic unchanged

QuizStart category PillRadioGroup gets an 'ALL' option prepended:

Shuffle icon + primary tone
Zero new design tokens — getOptionStyle returns undefined for 'ALL' (guard: 'ALL' has no entry in the category meta, so style falls back gracefully)
Default selected value changed to 'ALL'

page.tsx Session.category / start parameter propagate 'ALL'. Storage key is ALL::difficulty automatically. stats aggregation ignores 'ALL' key (QUIZ_CATEGORY_META own-key check, meta-mode guard).

D4. Test Suite Full Refresh (Wave E)

Test file	Change
`QuizStart.test`	Full refresh — two `'ALL'` ("전체") labels in category + difficulty radiogroups caused `getByRole` ambiguity → `within()` scoping, default category `'ALL'`, 4 new ALL-category tests
`index.test`	`+5` — 'ALL' filter behavior
`data-integrity.test`	New `+12` — 350-question unique ID, category, bilingual integrity
`category-meta.test`	5 new category var-slug assertions
`page.test`	Key updates reflecting 'ALL' default

Global: 1533 → 1555 tests (+22).

Implementation

Total atomic commits (start 66820e0):

Commit	Agent	Content
`229da66`	Palette + Herald	Wave A — QuizCategory +5, globals.css 20 lines, category-meta icons/vars, i18n +5 keys + ALL
`f8660d1`	Curator×10 + Herald	Wave B+C — 5 new categories 20 questions each, existing 5 +20 each (HARD fix), index.ts ALL_QUESTIONS merge, apostrophe fix ×2
`09dc639`	Herald	Wave D — getQuestionsByFilter/getRandomQuestions 'ALL' support, QuizStart 'ALL' prepend, page.tsx propagation, stats 'ALL' ignore
`2912955`	Herald + Librarian	Wave E — test suite refresh +22, data-integrity.test new, ADR sprint-227 KR+EN, README 164→165

Verification

tsc: 0 errors (apostrophe fixes included)
next lint (rtk proxy actual binary): 0 errors / 487 warnings (all pre-existing inline token var() warnings, 0 new)
jest: 152 suites / 1555 PASS (+22, 0 failures)
Global coverage: lines 88.03% / branches 79.37% (gate 83/71 — pass)
Quiz components and data: lines 100% / branches 98.59% (data-integrity /* istanbul ignore */ branches excluded)
next build: ✓ /[locale]/quiz 84.8kB (Sprint 224 was 40.2kB → increase: 350 bilingual questions now included in the client bundle per the existing architecture; see Lesson ⑤)
ADR gates: index count (sprint 165) / adr-en coverage (KR/EN 1:1) / adr-links 0 broken / doc-refs no broken

Lessons Learned

Large-scale content parallelizes cleanly at one-agent-per-file with zero conflicts — 10 category question files were generated by 10 Curator agents concurrently; only index.ts wiring was done in a serial integration step. No file conflicts, and each file being an independent unit made per-category tsc and test isolation trivial.
English apostrophes (possessives/contractions) inside single-quoted strings are a syntax error — 'Brewer's' and 'process's' caused the parser to see the string as terminating mid-word. For bulk English content generation: use double quotes, or catch early with an integrated tsc --noEmit pass.
Two radiogroups sharing the same 'ALL' / "전체" label make getByRole name queries ambiguous — getByRole('radio', { name: '전체' }) matches both groups. Use within(categoryGroup) / within(difficultyGroup) to scope queries to the correct radiogroup.
Meta filter options need zero new design tokens — getOptionStyle undefined guard is sufficient — 'ALL' has no entry in the category metadata, so returning undefined from getOptionStyle lets the caller fall back gracefully without touching existing category style logic. Reuses the existing difficulty 'ALL' pattern.
Client-bundle question data scales linearly with pool size — expanding from 150 to 350 bilingual questions grew /[locale]/quiz from 40.2kB to 84.8kB. Future work: per-category dynamic import (lazy-load) to reduce initial bundle (carry-over seed).

New patterns:

One-agent-per-file parallel content generation pattern — generate large content file sets with one agent per file in parallel; wire them together in a single index.ts integration step.
Meta filter option zero-token pattern — meta options like 'ALL' can be added with zero new design tokens by returning undefined from getOptionStyle and leaning on existing primary-tone styling.

Critic Cross-Review

Tool: Codex codex-cli 0.130.0, codex review --base 66820e0 -c model=gpt-5.5
Rounds: 2

R1 — [P2] 1 finding: The new default category 'ALL' and the 5 new categories were rejected by the backend quiz-record DTO allowlists (original 5 only), so authenticated saveResult (best-effort, swallowed error) silently fails — best records are never saved or merged up. Fixed in fcb1800: identity entity enum +5 / identity & gateway DTO allowlist = 10 categories + 'ALL' / regression specs (gateway +12, identity +13). DB column is varchar(30) so no migration is needed.

R2 — CLEAN (0 P-findings): "The changes type-check across the frontend, gateway, and identity service, and the updated category/ALL flow appears consistently propagated through data, UI, and persistence DTOs. I did not find a discrete introduced bug that would warrant an inline finding." (Codex ran tsc across frontend/gateway/identity.)

Overall verdict: ✅ Mergeable — R1 P2 (missing backend allowlist) fixed, then R2 CLEAN. Global 1555 PASS + backend regression specs; coverage and ADR gates all pass.