[CLAUDE] Docs: S41 close-out — record RAG corpus cleanup (3080->2406, 5/5 store_memory preserved)

STATUS + HANDOFF: S41 episode — exclude **/-anchor fix + AI_INFRA re-bootstrap +
store_memory reconcile (3 broadcast twin + audit-response twin + catch-up promoted-to-disk).
RAG chunks 3076->2406. SE post-verify PASS. Next S42 = product focus (Phase 11 / test gaps / Ops).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs/STATUS.md

@@ -3,7 +3,7 @@
 > **Update rule:** trước khi bắt đầu 1 task → ghi row `🔥 In Progress`. Xong → `✅ Recently Done`.
 > **Tiering rule (S40):** chỉ giữ **state hiện tại + 3 session gần nhất** ở file này. Session cũ hơn → `docs/changelog/sessions/`. Full history pre-S40 → `docs/_archive/STATUS-preS40-fullhistory.md`. (Tránh over-context — xóa double, không cắt nội dung.)
 
-**Last updated:** 2026-05-29 (Session 40 — init audit + **doc consolidation**: tier STATUS/HANDOFF 393KB→lean, re-ground count drift, fix stale credentials. KHÔNG feature/schema change.)
+**Last updated:** 2026-05-29 (Session 41 — **RAG corpus cleanup** w/ AI_INFRA: exclude `**/`-anchor fix + re-bootstrap 3080→2406 (−674 junk) + 5/5 store_memory preserved. KHÔNG feature/schema change.)
 
 ---
 
@@ -21,7 +21,7 @@
 | User memory | 27 | + `MEMORY.md` index |
 | Skills | 6 | 3 domain + 3 ops |
 | Sub-agents | **7** | Opus 4.8 1M (S39 split 4→7) |
-| RAG chunks | 3076 | ⚠️ S37-S39 content gap (re-ingest pending) |
+| RAG chunks | **2406** | ✅ S41 re-bootstrap clean (3080→2406, −674 junk: node_modules+_archive now excluded; user-memory 60 chunks/10 files slug-fixed + S38-S41 indexed) |
 
 **Bundle hash live (prod):** admin `cWAXid0q` · user `CX79e2kZ` (Run #247, S38).
 **Phase:** ✅ **Phase 10 COMPLETE 11/11** Plan G-* · ⬜ Phase 11 polish NOT started · 🚫 Phase 9 Ops blocked (anh main coordinate).
@@ -44,6 +44,12 @@
 
 ## ✅ Recently Done (newest on top — 3 session; cũ hơn → session logs)
 
+### S41 (2026-05-29) — RAG corpus cleanup (w/ AI_INFRA)
+- AI_INFRA RAG audit → SE-side prep: `.claude/rag.json` exclude root-anchored→`**/`-anchored (defeats gotcha #10: `node_modules/**`+`docs/_archive/**` weren't matching nested paths) + retired stale `_decision_log` "+321%/11,922".
+- **store_memory reconcile (anti-data-loss, NAMGROUP lesson):** unified at-risk rule = content reproducible từ file {exists ∧ matches corpus glob ∧ not excluded}. 5/5 accounted: 3 broadcasts disk-twinned + `16a6b6db` audit-response twin-safe + `0307141b` S37-S40 catch-up **promoted-to-disk** (`docs/changelog/sessions/2026-05-29-S37-S40-rag-catchup.md` — anchor was virtual, not real STATUS section).
+- AI_INFRA re-bootstrap host-side `--config`: **3080→2406** (−674 junk), node_modules=0 / _archive=0 / manual-marker=0, user-memory 60 chunks (slug fix). SE post-verify PASS (promoted catch-up returns file-indexed real heading, not `(manual)`). Commits `282cbd0`+`e8cbbe5` pushed (docs/infra → CI skip).
+- ⏳ Standing infra backlog (AI_INFRA, non-blocking): bootstrap glob→0 warn · auto_reindex hook fire · search_code corpus gap (src/*.cs+fe/*.tsx, v0 design) · registry sync.
+
 ### S40 (2026-05-29) — Init audit + doc consolidation
 - 7-agent smoke verify (🟪 test-specialist + 🟦 investigator-api load OK), RAG re-rank PASS (top 0.8789 Voyage rerank-2.5), 130 test confirm.
 - Doc consolidation: STATUS 170KB→5.7KB + HANDOFF 224KB→3.4KB (archive full → `docs/_archive/`). Count re-ground (endpoints 211, FE pages 65, menu ~53; tables 84 confirmed). Stale credential `admin@solutionerp.local`→`admin@solutions.com.vn` fixed. CLAUDE.md root patch (mig/table/test).