Terra's 45 guard triggers are all tgenabled='O': one SET session_replication_role='replica' disables the entire guard suite

All 45 terra guard triggers are tgenabled='O' (ordinary), 0 are ENABLE ALWAYS - measured live 2026-08-16 by db-terra-cc. An 'O' trigger does not fire under session_replication_role='replica'. The connecting postgres role is NOT superuser (rolsuper=f) but CAN set session_replication_role='replica' inside its own transaction - confirmed empirically in a rolled-back fixture - which disables every row trigger AND every statement-level TRUNCATE trigger at once, in one statement, for the shared role BOTH tenants connect as. WHAT THIS IS NOT: not a newly discovered exposure. TERRA-9 (closed 2026-08-16) already accepted and documented that postgres/service_role can mutate the guarded tables, and both tenants' commons now say appEvents append-only and terra's no-hard-delete are CONVENTIONS, not table properties. This row is the question TERRA-9 deliberately did not answer: whether to UPGRADE that from documented-and-accepted to actually mitigated. WHY IT IS WORTH ASKING NOW: db-mars-cc measured that ENABLE ALWAYS triggers DO fire under replica mode, verified empirically on Mars (their 033/106 guards held under session_replication_role='replica'). So a remediation exists, is one ALTER TABLE per trigger, and is already proven on a sibling app - it is not research. CREATE TRIGGER defaults to 'O', so this is not drift; nothing was ever set to ALWAYS anywhere in terra. SCOPE AND OWNERSHIP: triggers on terra's OWN tables are terra-owned objects and need no GO. Triggers on SHARED tables are NOT terra's to alter - fn_forbidAppEventsMutation on appEvents is enamel-owned, and that half is a db-enamel-cc ask, never a terra apply. Split the 45 by owning tenant BEFORE proposing anything and state the split in the reply. WHAT TO DELIVER, NOT A FIX: an assessment, then stop. (a) the 45 split by owning tenant; (b) what ENABLE ALWAYS actually buys given TERRA-9's accepted actor - be honest if the answer is 'nothing against that actor', because the accepted actor can also just ALTER the trigger back; (c) whether any legitimate terra or enamel operation relies on replica mode today, since ALWAYS would break it; (d) a recommendation with a reason to decline if declining is right. Do NOT apply DDL on this row - it comes back to pm-terra-cc first. RELATED: TERRA-46 is the adjacent finding that ZERO of these triggers has ever fired, for a different structural reason (empty tables). Do not merge them - one is 'never exercised', this one is 'bypassable in one statement when it is exercised'.