04 — Objective C: Axis Assignment Coverage

Verdict: PASS — committed (integrated active state). SQL: sql/C_axis_assignment_coverage_clone_COMMIT.sql. 16/16 checks PASS including 5 fail-closed negatives.

What was committed — the integrated two-axis state

The topic axis was activated and persisted, and topic owners seeded in the parallel axis_object_ownership table, producing the project's first integrated multi-axis coverage state:

axis	status	required	gap	reading
responsibility	active	210	0	production-equivalent; reads existing v_object_effective_owner
topic	active	6	2	honest residual (designed-in for detection)
total		216	2

The 5 topic required cells (after adding a stale tasks assignment) are owned for 4 and intentionally left uncovered for 1 (modules), and a 6th cell appears from the quarantine test — yielding the residual 2.

Confidence / zone (proven)

Every topic assignment carries confidence, match_score (kept separate — never collapsed), zone, provenance, and assigned_kind. The committed object→topic picture:

object	topic	zone	confidence	kind	quality_flag
agents	ai-ops	candidate	0.92	human	ok
governance_registry	registry-kg	candidate	0.85	agent	ok
workflows	workflow	candidate	0.96	human	ok
modules	workflow	candidate	0.55	agent	low_confidence
tasks	ai-ops	candidate	0.80	agent	stale (assigned 60d ago, rule staleness 30d)
iu_staging_record	ai-ops	quarantine	null	agent	quarantine (null provenance)

Detection (all four required signals proven)

v_axis_assignment_quality computes a single quality_flag per active assignment from the coverage rule:

• low_confidence — modules (0.55 < 0.70 threshold).
• stale — tasks (assigned_at older than staleness_days).
• quarantine / missing_provenance — iu_staging_record (null provenance forced to quarantine).
• missing assignment — v_axis_missing_assignment lists inventory objects with no topic assignment (29 after this run's assignments) — bounded by inventory.
• missing owner — v_axis_coverage_gap for topic (modules + the quarantine cell).

Invalid assignment — fail-closed (5 negatives, all rejected)

Each was attempted in a savepoint-isolated subtransaction and must error:

#	attempt	rejected by	SQLSTATE
N1	assignment with NULL provenance, zone≠quarantine	chk_provenance_or_quarantine	23514 (check)
N2	assignment to a non-existent axis value	FK → axis_value	23503 (fk)
N3	duplicate active assignment (same object/axis/value)	uq_axis_assignment_active	23505 (unique)
N4	owner referencing a non-existent value	FK → axis_value	23503 (fk)
N5	duplicate accountable owner for one cell	uq_axis_obj_accountable	23505 (unique)

Positive control: a NULL-provenance assignment is allowed only when zone='quarantine' — proving the provenance-or-quarantine invariant is a gate, not a blanket ban. That quarantine cell then correctly counts as an uncovered required cell (it still needs an owner), bumping the topic gap as expected.

Significance

This is the first time the project has run two governed axes at once with honest, differentiated coverage: responsibility fully covered (the proven model), topic partially covered with every quality signal — low confidence, stale, quarantine, missing owner, missing assignment — surfaced rather than hidden. The fail-closed negatives show the substrate cannot be driven into an invalid state.