40000x · 05 — Regression matrix

Compares the live post-mutation state at the close of 40000x against the 25000x PASS baseline.

Table-by-table

Surface	25000x	40000x	Δ	Verdict
information_unit	175	175	0	invariant ✓
iu_collection_template_registry	4	4	0	invariant ✓
iu_collection_template_version	4	4	0	invariant ✓
iu_collection_template_instance_lineage	17	17	0	invariant ✓
iu_auto_instantiate_event_log	13	13	0	invariant ✓ — macro is dry-run-only on this table
iu_piece_collection	24	24	0	invariant ✓
iu_piece_membership	118	118	0	invariant ✓
iu_lifecycle_log	146	146	0	invariant ✓
iu_vector_sync_point	152	152	0	invariant ✓
event_outbox total	123460	123460	0	invariant ✓
event_outbox (event_domain='piece')	0	0	0	invariant ✓
event_outbox (event_domain='template')	0	0	0	invariant ✓
event_type_registry	26	26	0	invariant ✓
dot_iu_command_catalog	24	26	+2	expected (mig 034)
iu_core_retention_policy	3	4	+1	expected (mig 035)
cutter_governance.review_decision	(unchanged)	(unchanged)	0	invariant ✓
cutter_governance.cut_change_set	(unchanged)	(unchanged)	0	invariant ✓

DIEU envelope axes

Axis	25000x	40000x	Δ	Verdict
DIEU-28	27	27	0	✓
DIEU-32	23	23	0	✓
DIEU-35	36	36	0	✓

Live query: select axis_a_doc_code, count(*) from iu_three_axis_envelope where axis_a_doc_code like 'DIEU%' group by axis_a_doc_code order by axis_a_doc_code.

Qdrant

Collection	25000x	40000x	Δ	Verdict
iu_core_iu_chunks	149	149	0	invariant ✓
production_documents	9456 (memory)	9499 (live)	+43	external drift, not caused by 40000x

The +43 on production_documents is between 25000x close (2026-05-25 ~04:33Z) and 40000x baseline (~05:38Z) — an hour of Directus production activity. This macro performed zero Qdrant writes (no credential staging; agent-data container was not invoked; the only Qdrant interaction was a one-shot read via alpine/curl on docker_incomex for regression counts).

Confirmation: iu_core_iu_chunks (the IU Core collection) is unchanged at 149 — the macro's only Qdrant-adjacent invariant.

Gates

11 IU Core write gates inert, both at start and at end. Infrastructure gates iu_core.routes_master_enabled=true and iu_core.route_worker_enabled=true unchanged.

Tests

Suite	25000x	40000x	Δ
Full pytest	1324 passed	1324 passed	0 (3 pinning updated, 0 added, 0 removed)
test_iu_core_540x_operator_runtime.py	green	green	one literal 24 → 26 + comment
test_iu_core_500x_operator_surface.py	green	green	one literal 24 → 26 + comment
test_iu_core_ddl.py	green	green	EXPECTED_COUNTS unchanged (D9 = 181)

D9 conformance

181 objects, 7/7 PASS — unchanged. Migrations 034 and 035 are both data-only (catalog + policy seeds); no schema-class object was added, modified, or removed.

production_documents — explicit non-mutation proof

The macro:

• Did not load any embedding model or run any embedder.
• Did not call any Qdrant write API.
• Did not start, stop, or docker exec into incomex-agent-data for write work.
• The single Qdrant interaction was a read of two collection metadata blobs via docker run --rm --network docker_incomex alpine/curl -s -H api-key … http://incomex-qdrant:6333/collections/{iu_core_iu_chunks,production_documents}.

The +43 row delta on production_documents is therefore attributable to the live Directus production system (or another macro/agent that owns production_documents), not to 40000x.

Verdict

REGRESSION_PASS. Every documented invariant from 25000x is preserved. The two expected deltas (dot_iu_command_catalog +2, iu_core_retention_policy +1) are the intentional outcomes of migrations 034 and 035 respectively.