S166 — Automation Audit Report (Phần B)

Agent: Claude Code CLI | Ngày: 2026-03-26 Type: Investigation (READ-ONLY, no code changes)

B1. Contracts có tự sinh không?

NO — Manual JSON authoring required.

6 contract files in web/tests/contracts/:

• registries-l1.json: 15 hardcoded checks (CAT-ALL, column headers, badges)
• health.json: 10 hardcoded checks (health dashboard rendering)
• species-matrix.json: 10 hardcoded checks (species page elements)
• sync-directus-vector.json: 2 checks (knowledge_documents sync)
• watchdog.json: 1 always-fail check (runner liveness)
• schema.json: Contract schema definition (not a check contract)

All checks reference specific entity codes, CSS selectors, page URLs. Adding a new collection → must create a new JSON contract manually.

B2. Runner có metadata-driven không?

PARTIALLY YES.

Two modes in scripts/integrity/main.js:

• v2.0 PG-driven (default): Queries measurement_registry table (pg-client.js line 38: WHERE method=2 AND enabled=true). Auto-discovers. Adding a check = INSERT 1 PG row.
• v1.0 Legacy (--legacy): Reads web/tests/contracts/*.json. NOT auto-extending.

Conclusion: Default runner IS metadata-driven. But legacy checks (the ones that generate stale_check issues) use JSON files.

B3. PG counting có tự mở rộng không?

NO — Manual trigger creation required.

Generic counting function exists (update_record_count() from s134_m5b). But each collection needs its own trigger:

CREATE TRIGGER trg_count_[collection]
AFTER INSERT OR DELETE ON [collection]
FOR EACH STATEMENT EXECUTE FUNCTION update_record_count();

This is manual per-collection SQL. Adding a new collection requires:

1. CREATE TRIGGER SQL
2. INSERT INTO meta_catalog row
3. No auto-trigger creation exists.

B4. Nuxt API endpoints có tự mở rộng không?

MIXED.

Endpoint	Auto?	Evidence
counts.get.ts	YES	Queries SUM(record_count) from meta_catalog where identity_class=managed. Auto-includes new entries.
system-issues.get.ts	NO	Hardcoded collection name system_issues. Only counts this one table.
matrix.get.ts	NO	Hardcoded COLLECTION_MAP with 19 collections (lines 34-54). Adding new = edit code.
system-issues-detail.get.ts	NO	Hardcoded SUB_CLASS_MAP with 10 sub_classes (lines 10-21). Adding new = edit code.

B5. Tổng kết: TỰ ĐỘNG vs HARDCODE

Thành phần	Tự động?	Chi tiết
PG triggers đếm	KHÔNG	Manual CREATE TRIGGER per collection
PG view v_registry_counts	KHÔNG	Manual entry per collection in view
meta_catalog	KHÔNG	Manual INSERT row per collection
Contracts JSON	KHÔNG	Manual JSON file authoring per check set
Runner v2.0 (PG-driven)	CÓ	Auto-discovers from measurement_registry
Runner v1.0 (Legacy JSON)	KHÔNG	Reads static JSON files
Nuxt API counts.get.ts	CÓ	Queries meta_catalog dynamically
Nuxt API system-issues	KHÔNG	Hardcoded collection name
Nuxt API matrix.get.ts	KHÔNG	Hardcoded COLLECTION_MAP (19 collections)

Tóm tắt: ~30% tự động, ~70% hardcode.

Tự động: Runner v2.0 (PG measurement_registry) + counts API (meta_catalog aggregate). Hardcode: PG triggers, meta_catalog entries, contracts JSON, matrix API, system-issues APIs.

Adding new collection requires 6 manual steps:

1. PG trigger (SQL)
2. meta_catalog entry (Directus/DOT)
3. measurement_registry row (PG — for runner v2.0)
4. Contract JSON (if legacy checks needed)
5. matrix.get.ts COLLECTION_MAP (if matrix page)
6. dot-registry-count-refresh (initial count seed)

Recommendations for future Điều 31 evolution:

1. Auto-trigger creation: A DOT tool that reads meta_catalog and creates missing counting triggers
2. Dynamic COLLECTION_MAP: matrix.get.ts should read from meta_catalog instead of hardcoding
3. Auto-contract generation: Generate basic contracts from meta_catalog schema
4. Auto meta_catalog registration: Directus webhook on collection creation → auto-INSERT meta_catalog row