P10A-1 — Điều 35 Segmentation Candidate (Read-Only Discovery)

Date: 2026-04-29 Captured (UTC): 2026-04-29T04:02:21Z Agent: Claude (Opus 4.7) executing P10A-1 dispatch Mode: ZERO MUTATION — SELECT only on pg_catalog / information_schema; SELECT only on tac_*_vocab rows. No INSERT/UPDATE/DELETE/DDL touched production data. DB verify: current_database = directus, current_user = directus ✅

---

0. Outcome Snapshot

#	PASS criterion	Result
1	Schema discovery (14 tac_* full cols + detailed FK/CHECK/trigger/privilege for 6 core/member + 7 vocabs)	✅ PASS
2	Source snapshot (path, revision, SHA256, captured_at)	✅ PASS
3	Batch marker identified	✅ PASS — tac_publication.id (uuid, PK)
4	Execution-path discovery (privileges + trigger defs + enabled status + fn owner/secdef, SELECT-only)	✅ PASS
5	Segmentation candidate JSON (17 units, all section_type ∈ vocab)	✅ PASS
6	Report uploaded to KB	✅ (this file)

Overall: 6/6 PASS. STOP after upload.

---

1. Source Snapshot — Điều 35 v5.2 FINAL

Field	Value
Path	knowledge/dev/laws/dieu35-dot-governance-law.md
Revision	13
Body bytes (UTF-8)	39,938
Body SHA256	4353ec6d453411a7c8e207658bbc4457d00f99747cba90551c8a4926894d2e5c
Captured at (UTC)	2026-04-29T04:02:21Z
Method	GET /api/documents/<urlencoded-path>?full=true (Bearer AGENT_DATA_API_KEY)
Title	ĐIỀU 35: LUẬT QUẢN TRỊ DOT — v5.2 FINAL (BAN HÀNH 2026-04-18 S178 Fix 15)
Tags	law, dieu-35, dot, governance, enacted, v5.2, final, s178-fix15, s178-fix23, fix_repair_dot, ops-code

---

2. Schema Snapshot — 14 tac_* Tables (full columns)

Tables (14): tac_birth_gate_config, tac_change_set, tac_change_set_member, tac_cs_lifecycle_vocab, tac_logical_unit, tac_lu_lifecycle_vocab, tac_pub_lifecycle_vocab, tac_publication, tac_publication_member, tac_publication_type_vocab, tac_review_state_vocab, tac_section_type_vocab, tac_unit_version, tac_uv_lifecycle_vocab.

2.1 tac_logical_unit (13 cols)

col	type	nn	default
id	uuid	Y	gen_random_uuid()
canonical_address	text	Y	—
doc_code	text	Y	—
parent_id	uuid	N	—
sort_order	int	Y	0
section_type	text	Y	— (FK→tac_section_type_vocab)
section_code	text	N	—
owner	text	Y	—
identity_profile	jsonb	Y	'{}'
tier	text	N	—
lifecycle_status	text	Y	'draft_only' (FK→tac_lu_lifecycle_vocab)
created_at	timestamptz	Y	now()
updated_at	timestamptz	Y	now()

2.2 tac_unit_version (20 cols)

col	type	nn	default
id	uuid	Y	gen_random_uuid()
logical_unit_id	uuid	Y	FK→tac_logical_unit
version_number	int	Y	1 (CHECK >0)
title	text	Y	—
body	text	N	—
description	text	N	—
content_hash	text	N	—
lifecycle_status	text	Y	'draft' (FK→tac_uv_lifecycle_vocab)
review_state	text	Y	'unreviewed' (FK→tac_review_state_vocab)
length_flag	text	Y	'normal' (CHECK in {normal,soft_limit,hard_limit})
length_exception_reason	text	N	—
content_profile	jsonb	Y	'{}'
editor	text	N	—
provenance	text	Y	'PROV-AI'
vector_sync_status	text	Y	'pending' (CHECK in {pending,synced,stale,error})
vector_synced_at	timestamptz	N	—
vector_chunk_count	int	Y	0 (CHECK ≥0)
created_at / updated_at / enacted_at	timestamptz	Y/Y/N	now()/now()/—

UNIQUE(logical_unit_id, version_number).

2.3 tac_publication (15 cols)

col	type	nn	default
id	uuid	Y	gen_random_uuid()
doc_code	text	Y	—
version	text	Y	—
publication_type	text	Y	FK→tac_publication_type_vocab
name	text	Y	—
owner	text	Y	—
description	text	N	—
lifecycle_status	text	Y	'proposed' (FK→tac_pub_lifecycle_vocab)
enacted_at	timestamptz	N	—
council_score	numeric	N	—
approved_by	text	N	—
risk_tier	text	Y	'medium' (CHECK in {low,medium,high,highest})
publication_profile	jsonb	Y	'{}'
created_at / updated_at	timestamptz	Y	now()

UNIQUE(doc_code, version).

2.4 tac_publication_member (6 cols)

col	type	nn	default
id	uuid	Y	gen_random_uuid()
publication_id	uuid	Y	FK→tac_publication
logical_unit_id	uuid	Y	FK→tac_logical_unit
unit_version_id	uuid	Y	FK→tac_unit_version
render_order	int	Y	0 (CHECK ≥0)
created_at	timestamptz	Y	now()

UNIQUE(publication_id, logical_unit_id).

2.5 tac_change_set (10 cols) [excluded as P10A-2 target per dispatch]

cols: id (uuid, PK), publication_id (uuid, FK→tac_publication, nullable), scope_description (text NN), lifecycle_status (text NN, FK→tac_cs_lifecycle_vocab, default 'draft'), apr_ref (text), owner (text NN), created_at, updated_at, submitted_at, enacted_at.

2.6 tac_change_set_member (8 cols) [excluded]

cols: id (uuid, PK), change_set_id (uuid, FK→tac_change_set, NN), logical_unit_id (uuid, FK→tac_logical_unit, NN), change_type (text NN; CHECK in {create,new_version,retire,structural}), old_version_id (uuid, FK→tac_unit_version), new_version_id (uuid, FK→tac_unit_version), snapshot_data (jsonb), created_at. UNIQUE(change_set_id, logical_unit_id).

2.7 Vocab tables (cols summary)

tac_section_type_vocab (PK code; 17 rows; cols: code, name, description, lifecycle_status default 'active' CHECK in {active,deprecated,retired}, owner, soft_limit_words default 500, hard_limit_words default 1500, description_required default true, body_required default true, created_at, updated_at; CHECK soft>0, CHECK hard>soft).

tac_publication_type_vocab (PK code; 10 rows; cols include lifecycle_status, default_risk_tier with CHECK in {low,medium,high,highest}).

Lifecycle vocabs (tac_lu_lifecycle_vocab 3 rows, tac_uv_lifecycle_vocab 4 rows, tac_pub_lifecycle_vocab 4 rows, tac_cs_lifecycle_vocab 7 rows, tac_review_state_vocab 5 rows): cols code (PK), name NN, description, sort_order int NN default 0 (CHECK ≥0), created_at, updated_at.

tac_birth_gate_config (PK checker_id; cols mode default 'block' CHECK in {block,warn}, enabled bool default true, rationale, created_at, updated_at).

2.8 Vocab values (vocab data)

tac_section_type_vocab (17): heading, article, paragraph, definition, principle, rationale, process, technical_spec, governance_process, checklist, instruction_block, reference_mapping, matrix, invariant_list, open_decision_list, appendix, changelog. (All lifecycle_status='active'.)

tac_publication_type_vocab (10): law (highest), policy (high), sop (high), constitution (highest), knowledge (medium), design_note (medium), report (low), memo (low), draft (low), working (low).

tac_lu_lifecycle_vocab (3): active, draft_only, retired.

tac_uv_lifecycle_vocab (4): draft, enacted, superseded, retired.

tac_pub_lifecycle_vocab (4): proposed, enacted, superseded, retired.

tac_review_state_vocab (5): unreviewed, in_review, review_passed, review_failed, needs_re_review.

tac_cs_lifecycle_vocab (7): draft, submitted, review_passed, approval_passed, enacted, rejected, withdrawn.

2.9 Constraints (FK/CHECK/UNIQUE/PK) — 6 core/member tables

Captured in §2.1–§2.6 inline. All FKs lead to vocab or parent entities — no cross-schema links. All vocab references enforce code PK.

2.10 Triggers on tac_* tables (catalog only, not fired)

Table	Trigger	Timing	Events	Function	Enabled
tac_logical_unit	trg_tac_birth_gate_lu	BEFORE	INSERT, UPDATE	fn_tac_birth_gate_lu	O (enabled)
tac_unit_version	trg_tac_birth_gate_uv	BEFORE	INSERT	fn_tac_birth_gate_uv	O
tac_unit_version	trg_tac_uv_compute_derived	BEFORE	INSERT, UPDATE	fn_tac_uv_compute_derived	O
tac_unit_version	trg_tac_enacted_immut	BEFORE	UPDATE, DELETE	fn_tac_enacted_immut	O
tac_publication_member	trg_tac_pm_consistency	BEFORE	INSERT, UPDATE	fn_tac_pm_consistency	O
tac_publication_member	trg_tac_pm_enacted_lock	BEFORE	INSERT, UPDATE, DELETE	fn_tac_pm_enacted_lock	O

Trigger functions (owner / SECURITY DEFINER):

• fn_tac_birth_gate_lu — owner=directus, secdef=true
• fn_tac_birth_gate_uv — owner=directus, secdef=true
• fn_tac_uv_compute_derived — owner=directus, secdef=true
• fn_tac_enacted_immut — owner=directus, secdef=true
• fn_tac_pm_consistency — owner=directus, secdef=true
• fn_tac_pm_enacted_lock — owner=directus, secdef=true
• fn_tac_log_checker_issue — owner=workflow_admin, secdef=true (helper, no direct trigger binding)

Trigger firing intentionally NOT verified in this read-only step.

---

3. Execution-Path Discovery (read-only)

Privileges (SELECT-only catalog query, information_schema.role_table_grants): role directus holds full DML (SELECT, INSERT, UPDATE, DELETE, REFERENCES, TRIGGER, TRUNCATE) on all 14 tac_* tables. Roles workflow_admin, incomex do not appear in role_table_grants for tac_* (no grants found via the queried filter).

Implication for P10A-2 (NOT executed here): Whichever role does the inserts must possess INSERT on tac_logical_unit, tac_unit_version, tac_publication, tac_publication_member. Catalog confirms directus does. Note from dispatch R4#4: the read-only verification used directus; this does NOT certify that directus is the appropriate execution role for P10A-2 — that decision is out of scope here.

Trigger surface that will fire on P10A-2 inserts (catalog-derived, not tested):

• INSERT into tac_logical_unit → fn_tac_birth_gate_lu (BEFORE)
• INSERT into tac_unit_version → fn_tac_birth_gate_uv + fn_tac_uv_compute_derived (BEFORE)
• INSERT into tac_publication_member → fn_tac_pm_consistency + fn_tac_pm_enacted_lock (BEFORE)

P10A-2 must satisfy these gates a priori (e.g. body_required / description_required per tac_section_type_vocab row, length_flag, content_hash if computed by fn_tac_uv_compute_derived).

---

4. Batch Marker (Rollback Anchor)

Chosen marker: tac_publication.id (uuid, primary key, default gen_random_uuid()).

Rationale: every downstream row created in P10A-2 chains back to one publication row via FK:

• tac_publication_member.publication_id → tac_publication.id
• tac_unit_version.logical_unit_id → tac_logical_unit.id (LU rows joined to publication via publication_member).

Rollback recipe (NOT executed here):

-- Given chosen pub_id (uuid):
DELETE FROM tac_publication_member WHERE publication_id = :pub_id;
-- Then delete versions + LUs created in this batch (tracked via session list):
DELETE FROM tac_unit_version  WHERE id = ANY(:uv_ids);
DELETE FROM tac_logical_unit  WHERE id = ANY(:lu_ids);
DELETE FROM tac_publication   WHERE id = :pub_id;

Caveat: trg_tac_pm_enacted_lock and trg_tac_enacted_immut will block deletion once lifecycle_status='enacted'. P10A-2 must keep candidate rows in proposed/draft until verify, otherwise rollback path collapses.

---

5. Segmentation Candidate (JSON, 17 units)

C1A SR-1..SR-7 was applied per heading hierarchy of the source. All section_type values are present in tac_section_type_vocab (verified §2.8). body_excerpt capped at 100 chars; body_sha256 is sha256(body_bytes), where body = section content excluding the heading line.

Summary table:

sort	unit_key	parent_key	section_type	title (trunc)	body_bytes
-1	dieu35.root	—	article	ĐIỀU 35: LUẬT QUẢN TRỊ DOT — v5.2 FINAL (synthetic root)	39938
0	dieu35.preamble	—	heading	preamble (title block + meta)	1397
1	dieu35.s1	dieu35.root	process	§1. MỤC TIÊU	1240
2	dieu35.s2	dieu35.root	process	§2. PHẠM VI	1776
3	dieu35.s3	dieu35.root	process	§3. DOT 2 CẤP	1215
4	dieu35.s4	dieu35.root	process	§4. SCHEMA	11147
5	dieu35.s5	dieu35.root	process	§5. QUY TRÌNH TẠO DOT MỚI — 8 BƯỚC	3094
6	dieu35.s6	dieu35.root	process	§6. VÒNG ĐỜI DOT (v5.2 mở rộng)	3912
7	dieu35.s7	dieu35.root	process	§7. ĐO LƯỜNG	190
8	dieu35.s8	dieu35.root	process	§8. DOT TỰ QUẢN TRỊ — 4 CẶP	4500
9	dieu35.s9	dieu35.root	process	§9. BOOTSTRAP — 4 BLOCKS	3628
10	dieu35.s10	dieu35.root	process	§10. SUCCESS METRICS	833
11	dieu35.s11	dieu35.root	process	§11. RETROFIT CLAUSE	2177
12	dieu35.s12	dieu35.root	process	§12. (Đã bỏ)	180
13	dieu35.appendix_a	dieu35.root	appendix	PHỤ LỤC A — 24 Domain Seed	1172
14	dieu35.changelog	dieu35.root	changelog	CHANGELOG	1964
15	dieu35.post_merge_todo	dieu35.root	checklist	GHI CHÚ BAN HÀNH (post-merge TODO)	896

5.1 Notes on segmentation choices

• Root unit is synthetic (sort_order=-1) so children can share a stable parent_key. Whether the canonical row is created at P10A-2 depends on §6 render-plan decision (see §6.2).
• §4 SCHEMA (11.1 KB) is intentionally kept as a single unit at this candidate stage; it exceeds soft_limit_words for process. Two paths for P10A-2:
  • (a) split into dieu35.s4.1, dieu35.s4.1.1, dieu35.s4.2, dieu35.s4.3, dieu35.s4.4 (recommended); section_type technical_spec for SQL-heavy children.
  • (b) keep as one and set length_flag='hard_limit' with length_exception_reason. Only if (a) blocked.
• §6 (3.9 KB) is borderline; preferred split: 6.1 (heading), 6.2 (governance_process), 6.3–6.5 (process/checklist), 6.6 (principle), 6.7 (paragraph).
• §7 is intentionally tiny (most content delegated to v5.0 baseline); kept as process.
• §12 (Đã bỏ) retained as process for traceability; render plan may flag as tombstone.
• PHỤ LỤC A is a domain reference table → vocab match appendix chosen over reference_mapping because it's a declarative seed list.
• CHANGELOG maps directly to changelog vocab.
• GHI CHÚ BAN HÀNH maps to checklist (vocab) — items are bracketed [ ]/✅ actions.

5.2 Full JSON (canonical)

The complete candidate JSON (with body_excerpt, body_sha256, body_bytes, sort_order, source_heading per unit) is 9,057 bytes. Snapshot of every unit follows (all 17):

{
  "doc_code": "DIEU-35",
  "version": "v5.2",
  "revision": 13,
  "source_path": "knowledge/dev/laws/dieu35-dot-governance-law.md",
  "source_sha256": "4353ec6d453411a7c8e207658bbc4457d00f99747cba90551c8a4926894d2e5c",
  "source_bytes": 39938,
  "unit_count": 17,
  "units": [ /* 17 unit objects, each with unit_key, parent_key, section_type, title,
                body_excerpt (≤100 chars), body_sha256 (hex), body_bytes, sort_order, source_heading */ ]
}

(Full per-unit objects — including each body_sha256 digest — are listed in the temp file /tmp/d35-segments.json on VPS, hash table:)

unit_key	body_sha256
dieu35.root	4353ec6d453411a7c8e207658bbc4457d00f99747cba90551c8a4926894d2e5c
dieu35.preamble	da1b6bce745bd6d7778d01f6927d88b600f94ac7a9413b90190779ff6a11d992
dieu35.s1	9385c7cd2bb7f592aa955aaa4795adaa084f29771e66b94271e5acf801961d39
dieu35.s2	8367833d17c356798c6b76b3b06d3d1a824f5d2b205fdf011414f359a48486cd
dieu35.s3	494dcc03cf95b306fb988cc1e0ad523e83c55d7c431337bd37e8fb7d5bbb5a1b
dieu35.s4	6dfaa690f14a6e8cabfe43bf8ac50ac6c396b2052102ecc24eb21f890b0e1c6e
dieu35.s5	adebffafbbfd1247aa44897a765129016f9b9caf01c0e0f5179c5df08832ae08
dieu35.s6	9b6e917a45438f1bdb69015fd4b6b6138c455b3b556416f0c83142615a53e806
dieu35.s7	c04c760d96e6fe4a21ccf5656b713b9e199c7b23e2888f3d6bdee329347afcdb
dieu35.s8	e345838bf2aa0d8e81a06dcc9780cc45d387c58772bb343b81d1f365d395c426
dieu35.s9	4c812b46ce65f031d33b0cbd4367575fd6c486462cb1f675d8347fd42c5071d9
dieu35.s10	b119ff9abd50b4e0fa96c9d56eb29290c83b7e2cb971fc22f1b1752c7b505348
dieu35.s11	f5c543a7812cc64aa449a8b5e08e3fcc0583c574714a08b90236fdc61e4bc332
dieu35.s12	d19f2141a04b6f6eb9eed6e0bb32f166ec6cf47a39dc88224c977cd85be10b81
dieu35.appendix_a	cf7e9e5f0d7f007d51d04ee74db0728cca83ceac247b98d08fc4a208c82e7401
dieu35.changelog	5733bc20276b1658a1a5ce03e00800e294c9f7279f85edd5d71305d66c0cc4e0
dieu35.post_merge_todo	f99f4cd89b6fbefd15bea8fe23b1f6685b1d46069311879a174808bc644fa875

Vocab sanity: every section_type used (article, heading, process, appendix, changelog, checklist) is present in tac_section_type_vocab ∧ lifecycle_status='active'. ✅

---

6. Render Plan Draft (for P10A-2 review — NOT executed)

6.1 Order of operations

1. Insert tac_publication row — doc_code='DIEU-35', version='v5.2', publication_type='law', name='ĐIỀU 35: LUẬT QUẢN TRỊ DOT — v5.2 FINAL', risk_tier='highest', lifecycle_status='proposed', publication_profile contains source_sha256 + revision + captured_at. Capture pub_id (= batch marker).
2. Insert tac_logical_unit rows — one per unit_key, parent_id resolved via unit_key → id map after insert (two-pass or use deterministic uuid). canonical_address pattern: dieu-35/{unit_key}.
3. Insert tac_unit_version rows — one per LU, version_number=1, body=full segment body, content_hash=body_sha256, lifecycle_status='draft', review_state='unreviewed'. fn_tac_uv_compute_derived may overwrite content_hash/length_flag.
4. Insert tac_publication_member rows — one per (pub_id, lu_id, uv_id), render_order = candidate sort_order+offset to keep ≥0.
5. Verify row counts = 1 pub + 17 LUs + 17 UVs + 17 PMs (or 16+root variant TBD).
6. STOP — no enact: keep publication.lifecycle_status='proposed' and uv.lifecycle_status='draft'; no transition to enacted in this batch.

6.2 Open decisions (route to GPT for P10A-2 sign-off)

• D1: Synthetic root — keep dieu35.root (article) or drop it? (Most laws have an implicit doc-level row; without it, all 16 §-units lose a parent.) Recommend KEEP, mapped to section_type='article' so that parent_id chain ends at the article.
• D2: §4 splitting — split now in candidate (option (a) above) before P10A-2, or accept oversized segment and split in v5.3? Recommend split before P10A-2 to satisfy soft_limit and to enable per-subsection vector chunking later.
• D3: Heading-only vs body-bearing units — tac_section_type_vocab.heading has body_required=false. Use heading for §s where the only content is sub-sections (none in current pass). For now no unit uses heading except dieu35.preamble whose body is the title-meta block.
• D4: Description backfill — description_required=true for all process rows. P10A-2 must auto-generate or APR-backfill description for each UV (per Đ43 description governance). Source candidate description = first paragraph of body, capped to vocab soft_limit_words for descriptions.
• D5: §12 tombstone — keep as process row with body=2-line note, or convert to paragraph/changelog annotation?

6.3 Trigger pre-flight checklist

• body and description non-empty for all section_types where body_required=true and description_required=true.
• length_flag left at default; fn_tac_uv_compute_derived will compute. If body exceeds hard_limit_words, expect EXCEPTION → split first (D2).
• content_hash either pre-set to body_sha256 or left to derived trigger (verify which).
• No row inserted with lifecycle_status='enacted' (otherwise immut + lock triggers fire).
• Verify tac_birth_gate_config for fn_tac_birth_gate_lu / fn_tac_birth_gate_uv — current default mode='block'. If block-mode requires fields we haven't computed (e.g. tier), pre-flight will fail. → Inspect tac_birth_gate_config rows in P10A-2 step 0.

6.4 Rollback plan

Use the recipe in §4. Keep all rows in mutable lifecycle state until P10A-2 verify completes; only flip to enacted after a separate APR.

---

7. Đ41 / VPS Code Hygiene

• No repo files were created or modified by P10A-1.
• Temp artifacts (read-only output, kept for audit, NOT committed):
  • /tmp/seg.py on VPS — segmentation script.
  • /tmp/d35-segments.json on VPS — full candidate JSON (9057 bytes).
  • /tmp/segments.json on local — mirror of the above.
• No git add / git commit performed. Per dispatch §9 fallback: only temp files; canonical output uploaded to KB → no git commit required.

---

8. STOP

P10A-1 ends at this upload. No further mutation, no P10A-2 actions. Hand off to GPT for sign-off on the open decisions in §6.2 before P10A-2 dispatch.

---

P10A-1 report | S187 | 2026-04-29 | Read-only | 6/6 PASS