11 — Law / Gate / Risk Self-Review

Final self-review of the command pack against the prompt standard and the Constitution.

1. Prompt-standard compliance (v1.2)

Standard rule	This pack
Design-Only Macro Mode	YES — authored, no implementation; no DDL/DML/migration/trigger/worker/cron/mutation
Survey-Then-Design Gate	YES — design accepted (Rev5) → design_pack_macro allowed; implementation deferred to named macros
No-underload / adaptive scale	YES — 12 deliverables, 5 author-ready prompts, full gate + DOT matrices, U1–U11 backlog, risk register; did not stop at a checklist
Author-ready-followup (§4F)	YES — drift feeds a surgical patch (U11), not a redesign; existing functions reused not re-authored
Live-ready-means-use-not-rewrite (§4C)	YES — reuse fn_iu_*/dot_iu_*/envelope/routes; no duplicate logic proposed
Execution Channel Pack	YES — each mutating prompt carries SSH workflow_admin channel proof + Hard-Gate-0; this macro = AUTHOR_MODE_ONLY by design
Dependency Closure Pack	YES — each mutating prompt lists upstream functions/tables/FKs/rollback primitives to confirm
Self-contained-after-clear	YES — each prompt names sources, baseline counts, allowed/forbidden, PASS criteria
No-hardcode	YES — identifiers resolved from registries; no literal uuids; tag keys validated against iu_metadata_tag_registry
FORBIDDEN section + no-mutation report	YES — deliverable 00 §7 + this section

2. Constitutional principles

• NT13 PG-first / NT8 Assembly-First / Điều 7: all tests read/assemble over existing PG substrate; no Nuxt logic, no OSS substitution; commands are DOT pairs. ✔
• NT14 six execution-readiness questions answered per test (data where / who runs / when / threshold / verified by what / failure handling) across deliverables 02–08. ✔
• NT6 five-layer sync / no bespoke Nuxt: no UI authored; tests are DB+DOT only. ✔
• Evidence-first / no fake PASS: every test defines explicit assertions + evidence rows; gated tests cannot report PASS without verify-close. ✔

3. Per-law mapping

Law	How honored
Điều 0-G (Birth)	d's split/merge must not create orphan children; first-use of split/merge sets carries birth/lineage supersede; orphan checks in c and d.
Điều 7 (Assembly First)	reuse-first; no new substrate where live exists; OSS only as adapter (none introduced).
Điều 28 (Nuxt/template)	no UI/template work; tests are headless DOT-callable.
Điều 30 (Reversibility)	every mutating step has a rollback primitive (fn_iu_structure_op_rollback, retire test objects, gate auto-close, BEGIN/ROLLBACK).
Điều 31 (Audit/Integrity)	every command writes dot_iu_command_run; gate flips write transition records.
Điều 32 (Approval)	bounded gate open requires approval_id; split/merge require review_decision_id; allow_no_review_decision NEVER flipped.
Điều 35 (DOT)	all new commands are catalog pairs with correct mutating flag + audit.
Điều 36 (Collection)	test collections born via registered path; collection law stays paper where DRAFT — no runtime collection-create invented.
Điều 37 (Governance/Factory)	no factory built; governance registries untouched; no 4 Mothers.
Điều 38/39 (IU/KG/cut/vector)	IU body single-owner respected; Axis-A reconstruct preserved through split/merge; iu_kg_edge derivation noted as iu_relation (drift, U11).
Điều 45 (Event/Queue)	register-before-emit for iu.*; idempotency + DLQ classification + heartbeat respected; route-scoped delivery preferred.

4. No-double-ownership

Each concern maps to exactly one owner law (verified live in survey §11). This pack introduces no new owner: read tests (b/c/f) sit under Điều 38/39; gate protocol approval under Điều 32; DOT under Điều 35; event/route under Điều 45. No concern is owned twice. The only NEW concern remains the 4 Mothers application layer (future Điều XX), explicitly deferred. ✔

5. No vector pollution

iu_core.vector_sync_enabled=false is respected and never flipped by any prompt; event payloads use refs-only (MP-D8 deny-list includes vector/embedding/body/secret); 1 IU → ≥1 point invariant unaffected (no vector writes proposed). ✔

6. No hardcode / no hidden second SoT

• All identifiers resolved from live registries/config; tag keys validated against iu_metadata_tag_registry.
• No second source of truth introduced: filters read iu_three_axis_envelope (derived from IU), not a parallel store; harness evidence is reporting, not authority. ✔

7. No gate left open

The bounded gate protocol (08) mandates fail-closed + verify-close; every mutating prompt's acceptance requires a final dot_config snapshot identical to baseline. allow_no_review_decision and vector_sync_enabled are pinned closed. ✔

8. No mutation in this macro

Confirmed: only read-only MCP query_pg (AST-validated read-only role; write-denied) + KB document uploads (the deliverables). No PG/Directus/Qdrant write, no migration, no DOT run, no gate change, no event registration, no schema/table/template/collection creation, no law enactment. ✔

9. No missing b–f path

Every test b–f has a dedicated deliverable with substrate, contract, DOT wrapper, gate map, assertions, evidence, failure modes, PASS criteria, and an implementing macro. Test a is acknowledged proven. ✔

10. No underload

12/12 deliverables; 5 author-ready prompts (+ a 6th flagged); drift table covers all §2.2 artifacts; gate matrix covers all 8 listed gates; DOT matrix covers all required proposed commands; backlog U1–U11 with parallelization; full law/risk review. Substance matches a large planning macro. ✔

11. Residual risks (carried forward)

• R-A: cutter_governance.review_decision shape unconfirmed by RO role → U3 must privileged-read first (BLOCKED-able).
• R-B: delivery gate may not support route-scoping → test e may need a global flip with all 15 routes proven inert (higher risk; mitigated by pre-open inventory + post-check).
• R-C: cut state-machine wedge (no mark_verified→mark_rejected) → U5 prerequisite for d's failure path.
• R-D: running 4 Mothers before b–f close → R1–R6 from survey (duplicate fields, industrial đẻ-rơi, ungoverned mutation). Held by the no-go.

12. Verdict

PASS — document-only command pack complete, constitutionally compliant, no mutation, no gate left open, every b–f path covered, sequencing and author-ready prompts provided. Implementation remains correctly BLOCKED behind the named macros and their approvals.