wiki-audit: category gaps — architecture, CORS, versioning, flags, backfill, realtime (G1–G10)

[choiyounggi]

Finding (wiki-audit 2026-08-05, axis: category taxonomy)

Walked a 16-concern development-perspective checklist against the 60 existing categories (plans/wiki-audit/findings/categories.md). Taxonomy is coherent; the gaps are additive (no restructuring needed):

High-value genuine gaps (frequency-ranked for an implementation-planning wiki):

• G1 backend/common/architecture — sync-vs-async integration, module boundaries, event-driven adoption criteria. wiki-plan design decisions currently go [no-wiki] here.
• G2 api-design/cors-and-preflight — top-frequency LLM implementation stumble; nothing owns CORS today.
• G3 api-design/api-versioning-and-breaking-changes — call-site-enumeration covers internal contracts only.
• G4 feature-flag-lifecycle — referenced by rollout pages, owned by none.
• G5 databases/operations/data-backfill-migrations — online-schema-changes covers DDL locking, not batched backfill/dual-write verification.
• G6 backend/common/realtime (websocket/SSE lifecycle) — auth, reconnect, heartbeat, backpressure, shutdown draining.

Lower priority: G7 frontend/i18n, G8 load/capacity testing, G9 audit-trails, G10 cross-service writes (saga/outbox/delivery guarantees).

Judged out of charter: cloud cost, ML training (routing probe confirms clean MISS).

Proposed fix

Grow via wiki-ingest as real cases arrive, prioritizing G1–G6; each proposed seed page list is in findings/categories.md.