Plan N 2 commits:
- N1+N2 atomic (commit 0326458) BE fix line 765 ApproverUserId discriminator + 2 regression test
- N4 (this) docs + memory update
Docs update:
- docs/STATUS.md — Last updated S23 t4 + stats 108 test (+2 từ 106)
- docs/HANDOFF.md — TL;DR S23 t4 với Investigator pre-flight catch root cause
- docs/changelog/sessions/2026-05-15-s23-turn4-plan-n-per-nv-lookup-bug.md — Session log Plan N
Memory user-level update (1 entry CRITICAL HOTFIX section):
- feedback_per_nv_permission_scope.md — Wire checklist 9 surface points (NOT 8)
+ Point 9 mới: Handler lookup site `currentLevelOptions` MUST discriminate
ApproverUserId. 3× cumulative gap analysis Mig 29 + 30 + 31.
Agent MEMORY drift (auto-flush):
- Investigator (.claude/agent-memory/investigator/MEMORY.md) — S23 t4 spawn N0
audit Hypothesis B verdict + sqlcmd verify + API curl verify
- CICD Monitor (.claude/agent-memory/cicd-monitor/MEMORY.md) — S23 t3 Plan M
Run #200 anomaly note (docs-only trigger CI mặc dù paths-ignore)
Stats final S23 t4:
- 31 mig · 59 tables · ~145 endpoints · 34 FE pages
- **108 test PASS (+2: per-NV lookup discrimination regression)**
- 47 gotcha · 20 memory (1 entry reinforced CRITICAL section)
- 6 skills · 4 sub-agents (1 Investigator spawn ~80K)
- 2 commits Plan N `0326458..HEAD` ready push
Pending: CICD Monitor post-deploy verify Plan N
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
11 KiB
Session 23 turn 4 — 2026-05-15 — Plan N HOTFIX per-NV lookup site discrimination
Dev: Claude Opus 4.7 1M (4 sub-agents active + em main coordinator)
Duration: ~1h
Base commit: f4055a1 (S23 t3 Plan M wrap)
Final HEAD: <pending> (Plan N batch + docs)
Total commits Plan N: 2 (atomic N1+N2 BE+test + N4 docs)
🎯 Trigger session
Bro UAT screenshot post-Plan M deploy phát hiện disconnect cực nghiêm trọng:
"Hiện User UAT test v2 vẫn chưa thấy cho chuyển tiếp đến người cuối hoặc các option trả lại trong menu Duyệt NCC → Duyệt → Chưa thấy xuất hiện option duyệt thẳng, trả lại theo option và cho Edit"
4 screenshots evidence:
- Workflow Designer: Admin tick TRUE 7 flag cho NV Test (UAT V2) slot Bước 2 Cấp 1, 3 NV khác CHỈ Drafter flag
- Dialog ✓ Duyệt: Chỉ có "Ghi chú (tùy chọn)" + Xác nhận — KHÔNG có checkbox F2 skipToFinal
- Dialog ← Trả lại: Chỉ có 1 radio "Trả về Người soạn thảo (mặc định)" — KHÔNG có 3 mode F1 khác (OneLevel/OneStep/Assignee)
- PE detail UI: KHÔNG có F3 Edit Section 2 + F4 Edit ngân sách
🔍 Investigator pre-flight audit (~80K spawn, Hypothesis B verify)
🟦 Investigator spawn audit DB + API + FE wire cross-stack:
API curl verify (admin token):
- GET
/api/purchase-evaluations/<id>→currentLevelOptions = {allowReturnToDrafter: true, 6 flag khác: false}— KHÔNG match flag admin tick cho NV Test - GET
/api/approval-workflows-v2/<wfId>→ 4 Level cùng Order=1, NV Test slot có 7 flag TRUE, 3 NV khác CHỈ AllowReturnToDrafter=true
sqlcmd prod verify:
- Schema
ApprovalWorkflowLevels4 rows cùng (ApprovalWorkflowStepId, Order=1) trong Step 2 — đúng OR-of-N Mig 29 pattern - NV Test (UAT V2) row có 7 flag TRUE: AllowReturnOneLevel + OneStep + ToAssignee + ToDrafter + EditDetails + EditBudget + SkipToFinal
- 3 NV khác chỉ AllowReturnToDrafter=true
FE wire grep verify:
fe-user/src/components/pe/PeWorkflowPanel.tsx:51,335-425đọcevaluation.currentLevelOptions.allow*đúng tên field, conditional render đủfe-admin/src/pages/system/ApprovalWorkflowsV2Page.tsx:889-946Designer wire 7 checkbox per slot đúng
Verdict: HYPOTHESIS B confirm — BE handler bug line 765:
// BUG ([PurchaseEvaluationFeatures.cs:765])
var curLevel = curStep?.Levels.FirstOrDefault(l => l.Order == curLevelOrder);
Schema Mig 29 (S21 t5 2026-05-13) refactor: 1 ApprovalWorkflowLevel row per ApproverUserId (OR-of-N cùng Order). 4 NV cùng Cấp = 4 row có Order=1. FirstOrDefault(Order==X) không có discriminator → luôn lấy row đầu DB (Trần Xuân Lưu / Lê Văn Bính / etc. tuỳ ordering EF nội bộ, đều chỉ Drafter flag) → bỏ qua admin tick per-NV của NV Test.
Bug present từ Mig 29 deploy 2026-05-13 (2 ngày prod) nhưng chỉ bộc lộ khi lần đầu admin tick selectively per-NV. Trước đây tất cả slot FALSE → behavior giống nhau (mọi actor đều thấy "không có options"), không lộ.
🌳 Plan N 2 chunk execution
Chunk N1 + N2 — BE fix + regression test (atomic commit)
👤 Chủ trì Solo (bug fix reasoning chain cross-stack — Implementer auto-refuse criteria #4).
N1 BE fix (PurchaseEvaluationFeatures.cs:765-781):
- var curLevel = curStep?.Levels.FirstOrDefault(l => l.Order == curLevelOrder);
+ // Match actor.UserId trong slot (per-NV admin opt-in flag).
+ // Admin / non-approver fallback row đầu (giữ behavior view detail).
+ var curLevel = curStep?.Levels.FirstOrDefault(l =>
+ l.Order == curLevelOrder && l.ApproverUserId == currentUser.UserId)
+ ?? curStep?.Levels.FirstOrDefault(l => l.Order == curLevelOrder);
5 LOC delta. ICurrentUser currentUser đã có DI sẵn line 652.
N2 Regression test (tests/SolutionErp.Infrastructure.Tests/Application/GetPurchaseEvaluationCurrentLevelOptionsTests.cs):
2 test method (~210 LOC test new file):
GetPe_PerNvLookup_ActorMatchesSlot_ReturnsActorSpecificFlags:- Seed 1 Workflow + 1 Step + 4 Level cùng Order=1, mỗi Level distinct flag profile (A=Drafter / B=OneLevel / C=SkipToFinal / D=EditBudget)
- 4 actor scenarios → assert mỗi actor nhận flag-set RIÊNG (KHÔNG profile khác)
- Critical assertion: Actor C →
AllowApproverSkipToFinal=true(regression bug bro UAT)
GetPe_PerNvLookup_AdminNonApprover_FallsBackToFirstRow:- Admin actor (KHÔNG match) → fallback FirstOrDefault EF SQLite non-deterministic ordering
- Weak assertion: NOT null + match 1 trong 4 distinct profile (mỗi Level distinct flag → fallback pick exactly 1)
Verify:
dotnet build src/Backend/SolutionErp.Applicationclean (0 warning, 0 error)dotnet test SolutionErp.slnx108/108 PASS (+2 từ 106: 58 Domain + 50 Infra)- 2 test scenario N2 PASS individually + full suite no regression
Chunk N4 — Docs + memory update + commit (this commit)
docs/STATUS.mdLast updated S23 t4 entrydocs/HANDOFF.mdTL;DR S23 t4 đầy đủ- Session log file này
- Memory user-level
feedback_per_nv_permission_scope.mdreinforcement S23 t4 wire checklist 9 surface points (NOT 8)
📊 Stats Plan N chốt
| Metric | Trước (S23 t3) | Sau (S23 t4) | Δ |
|---|---|---|---|
| DB tables | 59 | 59 | 0 |
| Migrations | 31 | 31 | 0 |
| Endpoints | ~145 | ~145 | 0 |
| FE pages | 34 | 34 | 0 |
| Unit tests | 106 | 108 | +2 (per-NV lookup discrimination regression) |
| Gotchas | 47 | 47 | 0 |
| Memory entries | 20 | 20 | 0 (1 entry feedback_per_nv_permission_scope.md reinforced — wire 9 points checklist) |
| Skills | 6 | 6 | 0 |
| Sub-agents | 4 | 4 (1 Investigator spawn ~80K) | active |
| Commits Plan N | — | 2 (atomic N1+N2 + N4 docs) | pending push |
🎯 Multi-agent ROI evidence Plan N
| Spawn | Agent | Cost (tokens) | Output | Catch |
|---|---|---|---|---|
| Pre-flight | 🟦 Investigator | ~80K | Hypothesis B verdict + file:line + DB sqlcmd verify + API curl verify + FE wire grep verify | Critical root cause — em main lần đầu hypothesize 3 candidates (A: admin chưa tick + B: BE handler bug + C: FE wire bug), Investigator confirm B với evidence-based file:line, avoid em main spam fix WRONG |
| N1 fix | 👤 Chủ trì | ~self | 5 LOC BE Service handler line 765 | — |
| N2 test | 👤 Chủ trì | ~self | 210 LOC test new file 2 scenario | First iteration EF SQLite ordering non-deterministic → fix weak assert fallback test |
| Reviewer | (skip) | 0 | — | LOW effort 5 LOC BE + isolated test, em main solo per criteria small scope < 30 min |
| Post-deploy | 🟩 CICD Monitor | TBD (pending) | Bundle hash + smoke + sqlcmd | Spawn sau push |
Total Plan N spawn cost (excl CICD): ~80K tokens — Investigator chính, em main solo execute.
Multi-agent value caught:
- Investigator catch root cause precisely — file:line + sqlcmd + API + FE evidence-based. Em main solo có thể đã spawn fix sai (vd suspect admin chưa tick → ask bro re-tick → vô ích vì admin đã tick đúng).
- 3-spawn cumulative pattern caught — Mig 29 + 30 + 31 + Plan N reveal: 3× refactor wire 8 surface points đúng nhưng MISS point 9 lookup discrimination. Pattern reinforced cho future flag F5+.
⚠️ Pattern reinforced — Per-NV admin opt-in wire 9 surface points (KHÔNG 8)
Old (S23 t1 K9 catch — 8 points): Mig 29 + 30 + 31 wire checklist 8 points (Domain + EF config + Mig 3-file + Service guard + PE bundle DTO + Admin AwLevelDto + CreateAwLevelInput + FE Designer).
NEW S23 t4 Plan N catch: Thêm point 9:
9. Handler lookup site
currentLevelOptionsMUST discriminateApproverUserId == currentUser.UserId(fallback row đầu cho admin/non-approver). Áp dụng cho TẤT CẢ handler đọc Allow* flag từ Level slot khi schema OR-of-N approvers cùng Cấp.
Audit checklist trước commit cho future flag F5+:
- Grep
FirstOrDefault.*Order ==trong codebase → enumerate all lookup sites - Verify mỗi site có discriminator role-context (ApproverUserId hoặc role-specific column)
- Test regression: seed N rows cùng Order × N actor distinct → assert actor-specific flag returned
3× cumulative gap analysis (Mig 29 + 30 + 31):
| Migration | Wire 9 points | Bug surface | Caught when |
|---|---|---|---|
| Mig 29 F1+F3 | 1-8 OK, 9 GAP | Lookup row đầu DB | Plan N S23 t4 (2 ngày prod) |
| Mig 30 F4 | 1-8 OK, 9 GAP | Same | Same (cumulative) |
| Mig 31 F2 | 1-8 OK, 9 GAP | Same | Same |
→ 3× cumulative refactor wire SAME bug — point 9 lookup discrimination chưa có precedent → em main + Reviewer + Implementer all MISS xuyên 3 plan.
⏭ Pending S23 t5+
- 🟩 CICD Monitor post-deploy verify — spawn sau push remote
- 🟦 Investigator follow-up gotcha #41/#48 candidate controlled test paths-ignore (carry from Plan M)
- 🟡 Plan B Contract V2 wire (Mig 32+33) — HIGH priority next
- ⛔ Plan F drop V1 — defer sau Plan B + UAT 2-3 tuần
- 🟢 Plan H PE PDF Export — LOW priority carry
- 🟡 Plan I RAG Hybrid setup — defer chờ bro confirm
- 🔧 Gotcha #47 paths-ignore agent-memory — PENDING bro confirm
📋 Pattern reinforced Plan N
- Per-NV admin opt-in flag wire 9 surface points (NOT 8) — point 9 handler lookup discrimination added. Memory user-level
feedback_per_nv_permission_scope.mdupdated CRITICAL HOTFIX S23 t4 section. - Investigator pre-flight audit BẮT BUỘC cho bug fix cross-stack — em main solo có thể spam fix WRONG hypothesis. Investigator file:line + sqlcmd + API + FE grep evidence base.
- Atomic commit BE fix + regression test cùng commit (per
feedback_uat_skip_verifyupdated rule Plan L lesson) — Service semantic refactor BẮT BUỘC test cùng commit. - EF SQLite non-deterministic ordering test gotcha — weak assertion cho fallback path (NOT null + match 1 distinct profile), KHÔNG assume "row đầu" deterministic.
References
- Memory user-level updated:
feedback_per_nv_permission_scope.mdCRITICAL HOTFIX S23 t4 section - Files: PurchaseEvaluationFeatures.cs:765-781 (N1) + GetPurchaseEvaluationCurrentLevelOptionsTests.cs (N2)
- Rules: §3.9 mirror 2 FE, §7 test timing test-before (regression bug),
feedback_per_nv_permission_scope(S23 t4 wire 9 points),feedback_uat_skip_verify(Plan L Service refactor + test atomic)