11000x · 09 — Regression matrix + lessons + carry-forward

Regression matrix (post-apply, before commit)

Invariant	Pre-baseline	Post-11000x	Verdict
Repo HEAD	d88d08b	d88d08b (working tree dirty, commit pending)	✓
Tests collected	1305	1324 (+19 new)	✓ +19
Tests passing	1305	1324	✓
DOT D9 inventory total	156	163 (+7)	✓ +7, all ok
D8 drift unregistered objects	0	0	✓
public.tables count	263	263	✓ unchanged
public.functions count	486	489 (+3 fns from 029)	✓ as designed
event_type_registry rows	25	25	✓ unchanged
dot_iu_command_catalog rows	20	20	✓ unchanged
iu_vector_sync_point total	152	152	✓ unchanged
- indexed	149	149	✓ unchanged
- dryrun	3	3	✓ unchanged
iu_lifecycle_log rows	146 (all enact)	146	✓ unchanged
information_unit lifecycle_status='enacted'	146	146	✓ unchanged
event_outbox total rows	121,289	121,289	✓ unchanged
event_outbox event_domain='piece'	0	0	✓ bounded proof clean
iu_core.routes_master_enabled	true	true	✓ unchanged
iu_core.retention_enabled	false	false	✓ unchanged
iu_core.three_axis_auto_refresh_enabled	false	false	✓ unchanged
piece_event_runtime.emit_enabled	(absent)	false	✓ default closed
piece_event_runtime.dry_run_only	(absent)	true	✓ safe default
Qdrant collection iu_core_iu_chunks	(not queried live)	(not touched by macro)	✓ no driver run
production_documents collection	(not queried live)	(not touched)	✓ NEVER touched
PR #669	OPEN (last known)	not modified by macro	✓
Mac cron pilot	installed	installed; healthcheck.py updated to 8 surfaces	✓ will pick up 8th surface on next fire
pg_dump pre-mutation backup	—	79.4 MB sha c62ed38a…	✓

All invariants pass.

Lessons written to memory

feedback-event-outbox-check-vs-registry-drift.md (NEW)

Registering an event_domain in event_type_registry is necessary but not sufficient — event_outbox.event_domain has a separate CHECK constraint that must also accept the new value.

Why: 10000x seeded 6 piece.* event types in the registry; the column CHECK on event_outbox still rejected 'piece'. The defect lay dormant because no piece event was ever emitted. 11000x surfaced it in Phase B of the bounded BEGIN/ROLLBACK proof and patched it via migration 030.

How to apply: any future macro that adds a new event_domain or event_stream value must include in its verification pack at least one BEGIN; INSERT INTO event_outbox …; ROLLBACK; round-trip.

feedback-trigger-emit-over-fn-rewrite.md (NEW)

When wiring side-effect emission to existing lifecycle functions, prefer a TRIGGER on the canonical audit log table over CREATE OR REPLACE rewrites of the lifecycle function body.

Why: lifecycle functions (fn_iu_supersede, fn_iu_piece_split, fn_iu_piece_merge) are 100+ lines of plpgsql with FSM gates, advisory locks, post-write read-back assertions, and governance FK probes. Rewriting them risks subtle bugs and complicates rollback. A trigger on iu_lifecycle_log gives 100% coverage of successful transitions with a ~40-line standalone fn whose drop is byte-clean.

How to apply: new emit-event hooks for lifecycle transitions go via a new trigger fn on iu_lifecycle_log. Edits to fn_iu_* in sql/iu-core/<num>_*.sql should be RESERVED for true bug fixes in the lifecycle behaviour itself.

feedback-pinning-tests-bump-per-macro.md (RE-CONFIRMED)

The 11000x macro hit the trip-wire tests/test_iu_core_ddl.py::EXPECTED_COUNTS on first run after the migration. The error pointed to the SUM=163 assertion, which led to discovery of 13 other pinning test files. All 13 were bumped in the SAME working tree (this commit), not in a follow-on, in line with the existing rule.

Carry-forward

For macro	Item
Operator	Phase 1 dry-run rehearsal: UPDATE dot_config SET value='true' WHERE key='piece_event_runtime.emit_enabled' — wait one supersede/split/merge call — inspect v_piece_event_outbox — UPDATE … value='false'. Runbook in ops/piece-event-runtime-package-11000x/05-operator-runbook.md.
11500x	Templates + Instances: fn_iu_template_from_collection, fn_iu_instance_from_template, fn_iu_validate_piece_graph, +3 DOT commands
12000x	Wire route-worker subscribers for event_domain='piece' — only after the operator has run Phase 1 dry-run rehearsal in production
Open external	PR #669 (frontend/DevOps owner) — not touched by 11000x
Open external	Retention enable timing: id-18 eligible ~2026-06-22, command_run eligible ~2026-08-21+ — not enabled by this macro

What's NOT in scope (and stays out)

• Production_documents collection: NEVER touched.
• PR #669 merge: NOT done.
• Nuxt deploy: NOT triggered.
• Qdrant collection reindex / new collection: NONE.
• Retention enable: NOT done.
• Cross-IU vector boundary: still healthy.
• Lark / S177 Cowork: unrelated, untouched.