Stage 1 — 03 fn_birth_policy_decision Foundation (Supertrack C)

Function

fn_birth_policy_decision(p_collection text) RETURNS jsonb — LANGUAGE plpgsql, STABLE (no write). Returns {decision, reason, source_policy, coverage_status, governance_role, no_go}.

Decision map (single definition of the exempt/defer policy, replacing logic implicit in multiple scripts):

coverage_status	decision	no_go
BIRTH_EXEMPT%	BIRTH_EXEMPT (skip)	false
BIRTH_REQUIRED	BIRTH_REQUIRED (birth)	false
BIRTH_DEFERRED_NEEDS_REVIEW	BIRTH_DEFERRED (defer)	false
not registered / null / other	UNKNOWN_BLOCK	true

Deterministic, no-write, fail-closed for unknown/unclassified. Preserves the existing live BIRTH_REQUIRED (birth) and BIRTH_EXEMPT (skip) behaviour exactly.

Source-of-truth alignment with the live gateway

The live fn_birth_registry_auto skips only coverage_status LIKE 'BIRTH_EXEMPT%'; everything else (incl. DEFERRED and unregistered) it births. The Stage 1 policy is therefore identical for EXEMPT and REQUIRED and stricter for DEFERRED (defer) and UNKNOWN (block) — never more births than live. See report 06/07.

Test matrix — v_birth_policy_decision_test_matrix

Samples per class + named cases (entity_labels → BIRTH_EXEMPT, registry_changelog → BIRTH_EXEMPT) + __nonexistent_collection__ → UNKNOWN_BLOCK. Returns the live decision/reason/no_go per collection.

Regression guard — v_birth_policy_decision_regression_guard (6/6 PASS, live)

Test	Pass
exempt_returns_skip (ALL 36 EXEMPT)	t
required_returns_birth (ALL 74 REQUIRED)	t
deferred_returns_defer (ALL 58 DEFERRED)	t
unknown_blocks_failclosed	t
null_blocks_failclosed	t
deterministic_same_twice	t

These run the function over every collection in each class — not samples — so the proof is exhaustive over the live registry (168 collections).