S167H-INV — Investigation Report: 6 System Issues

Date: 2026-03-26 | Agent: Claude CLI (claude-go) Mode: INVESTIGATE + FIX | No PR needed (PG-only fixes)

Phan A: DIEU TRA — 6 Issues

Bang phan loai

#	ID	Title	Severity	Classification	Root Cause	Action
1	2224	L1 vs PG registries total drift	CRITICAL	(c) Eventual consistency	PG updates before Nuxt cache. Now matching (19098=19098).	Auto-resolved
2	758	Legacy WATCHDOG (CTR-WATCHDOG-01)	CRITICAL	(c) Orphaned issue	Created by legacy runner, PG runner uses MSR-D31-WATCHDOG.	Manually resolved
3	2157	PG WATCHDOG — runner song	CRITICAL	(d) Expected behavior	Liveness beacon — always FAIL by design.	Keep open
4	2223	GEM_CHAOS_P1 NULL	WARNING	(b) False positive	Chaos artifact measurement queries non-existent row.	Disabled, auto-resolved
5	2225	2079 _dot_origin NULL	WARNING	(a) Real finding	2079 records missing _dot_origin across 18 collections.	Keep per 0-AM
6	2226	2040 broken universal_edges	WARNING	(a) Real finding	2040 edges with broken source/target references.	Keep per 0-AM

A3: Root Cause per Issue

Issue #2224 (L1 drift):

• source_query: SUM(record_count) FROM meta_catalog WHERE identity_class = 'managed'
• target: /api/registry/counts (Nuxt API)
• At failure: PG=19053, Nuxt=19054 (off-by-one, cache timing)
• At investigation: PG=19098, Nuxt=19098 (matching)
• Verdict: Eventual consistency, auto-resolved on next run.

Issue #2223 (GEM_CHAOS_P1):

• source_query: SELECT record_count FROM meta_catalog WHERE code = 'GEM-CHAOS-P1'
• GEM-CHAOS-P1 row cleaned up by Codex in S167F, but measurement_registry entry was NOT
• Every run: query returns NULL -> FAIL
• Fix: UPDATE measurement_registry SET enabled = false

Issue #758 (Legacy watchdog):

• violation_hash 39e10a47 = hash(CTR-WATCHDOG | CTR-WATCHDOG-01 | watchdog_fault) [legacy]
• violation_hash ea804499 = hash(CTR-WATCHDOG | MSR-D31-WATCHDOG | watchdog_fault) [PG runner]
• Different checkId -> different hash -> dedupe sees as separate
• Auto-resolve excludes watchdog_fault (dedupe.js line 184)
• Fix: Manually resolved.

A4: Watchdog Auto-Resolve Logic

dedupe.js autoResolveStale() fetches all open dieu31-runner issues EXCEPT watchdog_fault. For each: if violation_hash NOT in seenHashes from current run -> resolve. Watchdog excluded by design as liveness beacon. Bug: legacy #758 was orphaned and could never be resolved automatically.

A5: Scanner Last Run

Run ID: s167h-verify (2026-03-26 ~07:29 UTC). 10 measurements (GEM_CHAOS disabled). PASS: 7 | FAIL: 2 | ERROR: 0. Auto-resolved: 2 stale issues (#2224, #2223).

Phan B: FIXES APPLIED

Fix	What	How
B1	Disable chaos artifact measurement	UPDATE measurement_registry SET enabled=false WHERE measurement_id='MSR-AUTO-GEM_CHAOS_P1'
B2	Resolve orphaned legacy watchdog #758	UPDATE system_issues SET status='resolved' WHERE id=758
B3	No scanner code bug found	Runner code is correct
B4	Real findings kept per 0-AM	#2225 (NULL), #2226 (broken edges) untouched

No code changes — PG-only data corrections.

Phan C: PRODUCTION VERIFICATION

C1: System Issues (production URL)

curl -s https://vps.incomexsaigoncorp.vn/api/registry/system-issues

{"totals":{"all":3,"critical":1,"warning":2,"info":0,"group_count":2}}

Before: 6 open (3 critical, 3 warning) After: 3 open (1 critical = watchdog, 2 warning = real findings)

C2: Health (production URL)

curl -s https://vps.incomexsaigoncorp.vn/api/health

{"status":"healthy","data_integrity":{"document_count":588,"vector_point_count":863,"ratio":1.47,"sync_status":"ok"}}

C3: Runner Output (on-demand)

Run ID: s167h-verify. PASS: 7 | FAIL: 2 | ERROR: 0. WATCHDOG: alive. Pass Rate: 77.8% (7/9). Auto-Resolved: 2 stale issues.

C4: Before/After Comparison

Metric	Before	After
Open issues	6	3
Critical	3	1 (watchdog)
Warning	3	2 (real findings)
False positives	1	0
Orphaned watchdog	1	0
Measurements	11	10
Pass rate	60.0%	77.8%

Tu Kiem Tra

#	Question	Result
1	Read Operating Rules?	DAT
2	Assembly Gate?	DAT
3	6 issues SQL evidence + classification?	DAT
4	Root cause clear with proof?	DAT
5	Watchdog auto-resolve documented?	DAT
6	Fix: PG-only, no code	DAT
7	Explanation clear?	DAT
8	Production URL evidence?	DAT
9	Report at reports/s167h-inv?	DAT
10	Real data untouched?	DAT

---

S167H-INV DONE. 6 -> 3 issues. 3 remaining = 1 watchdog + 2 real findings. No code changes needed.