07 — Implementation Backlog (NOT executed in this macro)

Status: Backlog only. No implementation occurs in this macro. Each item below is a discrete unit of work for a future macro, scoped to "honest PARTIAL > fake PASS" (40000x lesson). The user explicitly approved design-only scope here.

The backlog is ordered so each step's preconditions are met by an earlier step. Items are tagged with the source design doc they implement.

---

P0 — Substrate (migration NNN)

Source: 03-…-architecture.md, 05-…-birth-protocol-plan.md.

#	Item	Acceptance
P0.1	Author migration NNN_iu_staging_substrate.sql (DDL only)	File present; lints clean
P0.2	Author paired NNN_iu_staging_substrate_rollback.sql (REFUSED-guarded)	File present; rollback exits clean on empty state
P0.3	Add 2 tables, 2 views, 8 functions, 2 triggers, 3 configs, 5 event types per §5 of 05-…	DDL count matches
P0.4	All functions: SECURITY DEFINER, STRICT, owned by workflow_admin, REVOKE FROM PUBLIC, GRANT EXECUTE TO directus	privilege audit clean
P0.5	CHECK constraints from 03-…-architecture.md §2 + §3 in place	\d shows them
P0.6	Birth-registry triggers on both tables	trigger introspection shows row count = 2

Apply path. Owner-applies via pg_hba local trust socket as workflow_admin (70000x lesson feedback-pg-hba-local-trust-unblocks-role-channel). Do not borrow superuser; do not run half-as-owner (50000x feedback-honest-channel-block-beats-partial-trigger).

---

P1 — Birth registration (no DDL — registry writes)

Source: 05-…-birth-protocol-plan.md §2.

#	Item	Acceptance
P1.1	Create species SPE-NVS via dot-species-create with metadata.vector_eligible = false	row exists; metadata correct
P1.2	Register COL-IUS-001 iu_staging_record via dot-collection-register (purpose set; vector_eligible = false)	registry row + Nuxt page lists it
P1.3	Register COL-IUS-002 iu_staging_payload similarly	same
P1.4	Map both via dot-species-map to SPE-NVS	species_collection_map shows 2 new rows

Order. P1 runs after P0 lands so the trigger-driven birth rows exist for the registry sanity check.

---

P2 — DOT modules

Source: 05-…-birth-protocol-plan.md §4.

#	Item	Acceptance
P2.1	dot_iu_staging_create (Python entrypoint + SQL fn wrapper)	R280 probe + first happy-path run dry_run
P2.2	dot_iu_staging_approve	same
P2.3	dot_iu_staging_consume	same
P2.4	dot_iu_staging_reject	same
P2.5	dot_iu_staging_cleanup	gated by staging_cleanup_enabled = false; dry-run reports rows_eligible
P2.6	dot_iu_staging_unregister	REFUSED-guarded on non-cleaned rows
P2.7	Catalog rows in dot_iu_command_catalog; R280 governance probe PASS (expected_count = baseline + 6, all_resolvable = t)	probe green

---

P3 — Vector-exclusion enforcement

Source: 04-…-vector-exclusion-contract.md.

#	Item	Acceptance
P3.1	Author sidecar collection_registry_vector_policy (PK=FK to collection_registry.id) with vector_eligible boolean DEFAULT true NOT NULL and semantic_search_eligible boolean DEFAULT true NOT NULL; author read-side view v_collection_vector_eligibility joining parent + sidecar with COALESCE(crvp.vector_eligible, true) default. No ALTER TABLE on collection_registry.	sidecar table + view exist; per Q5 A-as-sidecar freeze (08-review-questions.md)
P3.2	Update vector connector to read vector_eligible + species metadata.vector_eligible	code review + unit test
P3.3	Add static skip-list with iu_staging_* prefix to connector	grep finds entry
P3.4	Wire panic gate iu_core.no_vector_staging_excluded: default true; refuse startup on false	startup test
P3.5	(Optional) extend Agent Data search filter (~20 LOC per 20A report) for explicit metadata exclude	tested

---

P4 — Healthcheck

Source: 04-…-vector-exclusion-contract.md §4.

#	Item	Acceptance
P4.1	fn_iu_staging_healthcheck() returning V1..V8 invariants	row-per-invariant
P4.2	Integrate into existing IU-Core healthcheck runner → 9/9 GREEN	runner output
P4.3	V6 Qdrant probe (scroll with payload.staging_record_id IS NOT NULL → expect 0)	external probe

---

P5 — Retention / cleanup

Source: 03-…-architecture.md §10, 05-…-birth-protocol-plan.md §6.

#	Item	Acceptance
P5.1	iu_core_retention_policy row per staging_kind (default 14 d for mark_manifest, 7 d for agent_intermediate, 30 d for review_package, …)	rows present
P5.2	First cleanup rehearsal in dry-run mode (cleanup_enabled = false)	rows_eligible = 0 on empty
P5.3	First live cleanup after non-zero rows exist (later macro)	counts move; orphan-vector probe = 0

---

P6 — Tests

Source: 04-…-vector-exclusion-contract.md §5.

#	Item	Acceptance
P6.1	8 regression tests from §5 of 04-…	pytest -k staging green
P6.2	Lifecycle state-machine tests (pending→approved→consumed; pending→rejected; expired path)	green
P6.3	Idempotency-key UNIQUE replay test	green
P6.4	Pinning tests bumped (tables, views, fns, configs, event_types)	numbers match SSOT

---

P7 — Operator runbook

Source: new doc to be written in implementation macro.

#	Item	Acceptance
P7.1	OPERATOR_RUNBOOK.md addendum: create / approve / consume / reject / cleanup commands with examples	reviewed
P7.2	Failure-mode catalogue + recovery steps	reviewed

---

P8 — MARK workflow integration

Source: 06-…-mark-workflow-integration.md.

#	Item	Acceptance
P8.1	Agent MARK entrypoint switches from file-path output to dot_iu_staging_create	first MARK run lands a staging row
P8.2	dot_iu_cut_from_manifest gate-check refuses non-approved staging rows	refuse test
P8.3	First end-to-end MARK→REVIEW→CUT→VERIFY through staging zone (Điều 37)	digest match; lifecycle progression observable

---

P9 — Documentation lift

#	Item	Acceptance
P9.1	Promote this addendum from design/ to enacted form (under Điều 36 v5.x ENACTED)	KB entry under enacted laws
P9.2	Update Điều 36 main law text to reference the addendum	one-line cross-ref
P9.3	Update 80000x package to point at staging-zone artefacts	replaced filepath references

---

P10 — Out-of-scope (separate addenda)

These were raised in the design but are explicitly out of scope of any single implementation macro:

• Staging-inclusive search mode (PG text-search over staging payloads) — separate addendum.
• Promotion path from staging to a new permanent collection (e.g. promote a reviewed import preview into a permanent table) — separate addendum if ever needed.
• A Directus UI page for staging records (Option B of 05-… §3) — optional.
• Hard delete (vs cleaned) — only if retention policy demands.

---

11. Suggested macro packaging

Group the backlog into ≤3 macros:

macro_A:   # substrate + birth + DOTs (highest blast radius)
  contains: P0, P1, P2, P3, P4, P6
  exit_criterion: 9/9 healthcheck green, all 8 regression tests pass, 0 production rows

macro_B:   # MARK integration
  contains: P5 dry-run only, P7, P8
  exit_criterion: first end-to-end MARK→REVIEW→CUT→VERIFY through staging

macro_C:   # cleanup live + retention enabled
  contains: P5 live, P9
  exit_criterion: first cleanup live with 0 orphan vectors; addendum promoted

This packaging matches the 80000x carry-forward philosophy: one substrate macro, one integration macro, one retention/closeout macro.

---

12. STOP — review gate before any of the above runs

Per the macro instructions: route to GPT/User for design review before any implementation. See 08-review-questions-before-implementation.md for the open questions that must be answered before P0 is authored.