docs: sync from backend 8835068 — C2 chat query bounds closeout

09 - Audits/C2-DrizzleChatRepo-Partial-Fix-Report.md

@@ -0,0 +1,30 @@
+# C2: DrizzleChatRepo.findRows - Closeout Report
+
+**File:** `src/db/repositories/drizzle/DrizzleChatRepo.ts`
+**Method:** `findRows()`
+**Status:** Fixed in `backend@8835068` / v2.9.35.
+
+---
+
+## What Was Finished
+
+The earlier C2 fix moved the hot participant/archive/unread predicates into SQL, but this report correctly identified that edge paths could still fetch too much data. Commit `8835068` closes those remaining gaps:
+
+| Gap from the partial report | Final behavior |
+|---|---|
+| Empty `findRows({})` fetched the whole chat table | All chat row reads now go through a bounded query builder with a 1000-row max cap |
+| `findOne({})` could fetch many rows before returning one | `findOne()` now uses `LIMIT 1`; id-only queries use the existing `findById()` fast path |
+| `findForUser()` applied `skip`/`limit` after fetching all matching rows | SQL-pushable predicates now use DB-side `offset`/`limit` |
+| Search/fallback predicates still used in-memory matching | Fallback matching remains for unsupported predicates, but only after a bounded 1000-row scan |
+| Chat `type` was left as a fallback predicate | `type` is now pushed to SQL for `direct`, `group`, and `support` |
+| `settings.isArchived` had no matching B-tree index in the current schema/migrations | Added schema index and migration `0026_chat_settings_archived_idx.sql` for `chats_settings_is_archived_idx` |
+
+## Verification
+
+- `npm run typecheck` - passed.
+- `npm test -- --runTestsByPath __tests__/drizzle-chat-repo.test.ts __tests__/db-audit-high-indexes.test.ts --runInBand` - passed, 2 suites / 9 tests.
+- `scripts/smoke/db-audit-service-regressions.sh` - passed, 18 suites / 73 tests.
+
+## Remaining Long-Term Schema Work
+
+The C2 unbounded-fetch and pagination issue is closed. The larger chat storage design remains a separate schema project: moving `messages`, `participants`, and `unreadCounts` out of JSONB arrays into relational tables would enable exact indexed deep search, precise fallback pagination, and targeted message updates without rewriting large JSON blobs.