IU Core Runtime Activation E2E — 06 Runtime behaviour proof

All proofs ran inside BEGIN … ROLLBACK transactions — nothing committed. Gate-open and dry-run states were created inside the transaction and discarded. A session GUC app.canonical_writer='fn_iu_create' was set LOCAL to satisfy fn_iu_gateway_write_guard for the rolled-back probe unit_version inserts.

SECTION 1 — runtime trigger, master gate CLOSED — 3/3 PASS

A real unit_version INSERT fired trg_iu_out_version.

• 1.1 gate closed → trg_iu_out_version emitted 0 event_outbox rows.
• 1.2 gate closed → 0 iu_route_attempt rows.
• 1.3 parallel path live → trg_aa_iu_notif_version wrote 1 iu_notification_event row — the two sinks are independent.

SECTION 2 — runtime trigger, master gate OPEN (end-to-end emit) — 4/4 PASS

Gate opened inside the txn, then a real unit_version INSERT:

• 2.1 fn_iu_core_routes_enabled() reads OPEN inside the txn.
• 2.2 trg_iu_out_version emitted exactly 1 event_outbox row.
• 2.3 emitted row registry-matched: type=version_applied, stream=update, lane=immediate — passes trg_event_outbox_type_validate.
• 2.4 payload_classification='safe_metadata' — the H3 fix proven on the live trigger path, end to end.

SECTION 3 — inbound route, disabled vs dry-run — 3/3 PASS

Gate open; inbound trigger trg_iu_sql_in_iu_sql_link fired by iu_sql_link INSERT.

• 3.1 route disabled (as seeded) → capture found no enabled route → 0 iu_route_attempt rows.
• 3.2 route set enabled=true (dry_run stays true) → INSERT produced exactly 1 iu_route_attempt.
• 3.3 the attempt row carries status='dry_run' — logged, never delivered.

Verdict

Gate-closed inertness, gate-open emit, disabled-route silence, and dry-run logging are all proven on the real attached triggers against production — with zero committed rows.