13 — Final Summary

Macro: PRE_BIRTH_PILOT_DOT_TOOLS_PERMIT_AND_COMPOSITE_UNIQUE_ROLLBACK_REHEARSAL Status: PASS. Live verification complete; all five rehearsals completed inside reversible transactions; 0 live mutation; entry state == exit state.

What was asked vs. what was proven

Workstream	Question	Result
A	Can an additive, reversible birth_admission_permit table enforce the invariants?	PROVEN — 7-state model, CHECK, idempotency unique, single-active partial-unique, expiry, consumed-reuse all tested
B	Can (entity_code, collection_name) replace UNIQUE(entity_code) safely?	COMPOSITE_READY — 0 violations across 1.12M rows, 0 FK deps, 0 null collections; defect & fix both reproduced
C	Can fn_birth_registry_auto's conflict target be patched without breaking 166 triggers?	PROVEN — CREATE OR REPLACE keeps OID 39232; 166 triggers stay bound; rollback restores exact md5
D	Can a dot_tools permit-consuming BEFORE gate block birth without a permit?	PROVEN — no-permit/expired blocked, valid succeeds + consumes; no fn_birth_gate change, no prod lock
E	Can finalize-at-commit confirm birth only if the row survives the transaction?	PROVEN — DEFERRABLE constraint trigger (live layer2 pattern), CONSUMED→FINALIZED at boundary, vanished row fails
F	Sequential DOT entrypoint to remove operator memory burden	SPEC COMPLETE — dot-birth-admit state machine, idempotency, retry, break-glass, stop-conditions
G	Keep governance decoupled but integrable after BORN	DESIGN CLEAR — cursor-tail CDC, 0 new tables, capture-only until OSPA ≥ 1, no-lost-handoff
H	Superuser bypass reality	DESIGN COMPLETE, DETECT_ONLY — 5 baselines + break-glass ledger, fail-closed via gate_drift BLOCK dim
I	Apply packet	COMPLETE — ordered, reversible, owner-approval checklist, one-way-door flagged

Headline findings

1. The model is technically ready to apply for the dot_tools pilot. No engineering blocker remains; every blocker is authority (owner DDL, registrar creds, scheduler, OSPA).
2. Composite unique is trivially clean — already satisfied by 100% of live data; the only nuance is that dropping UNIQUE(entity_code) becomes a one-way door once a second-collection code is born (sequence the pivot births separately).
3. The 166-trigger risk is not real — CREATE OR REPLACE is OID-stable; it's a one-line body swap, not a trigger rewrite.
4. The finalize pattern already exists in production (trg_iu_birth_gate_layer2) and was reused verbatim in shape.
5. A DDL-guard event trigger already fires in prod — the drift-monitor substrate partly exists.
6. The hardcoded ^[A-Z]+-[0-9]+$ code rule (in fn_pre_birth_check) must become registry-driven, or dot_tools keeps a 47% rejection rate — this is on the apply checklist.

Honesty / forbidden-action compliance

• No DDL/DML committed; birth_registry, dot_tools, all functions byte-identical (doc 11).
• No composite unique added live; no fn_birth_gate flip; no permit table created; no trigger permanently attached; no production DOT registered; dot-pivot-update not executed; RP cleanup not continued (NO-GO); OSPA not faked (still 0).
• Birth-first is NOT claimed achieved. This is rehearsal only.

Live state at exit (unchanged from entry)

birth_registry 1,126,728 rows · dot_tools 309 (unborn 0) · BLOCK dims 59 / 6 / 16 / 1 · OSPA 0 · handoff.object_born active=false · bypass GUC unset.

Next

PRE_BIRTH_PILOT_DOT_TOOLS_APPLY_PERMIT_AND_COMPOSITE_OWNER_GATED — apply the doc-10 packet under owner approval, with a hard STOP before dropping the single-column unique.