Initial curated RAGcore source from c16667f0c9397e2bfc8dd0255b114fba23c8d05e

runbooks/RESTORE.md

@@ -0,0 +1,23 @@
+# Restore Runbook
+
+Executable via `RestoreGateService`/`scripts/restore_test.py` as of M6-05
+(steps 1-8; step 9's scheduled trigger and step 10's operational traffic
+cutover remain manual/operational). See `artifacts/evidence/M6-05/` for a
+known, confirmed external blocker on this development host: Qdrant's own
+snapshot-recovery mechanism currently fails here for reasons unrelated to
+this code (root-caused in detail there) -- the code is complete and
+unit-tested, and expected to pass real-infra validation unchanged once
+that environment issue is resolved.
+
+1. Select a completed verified backup and validate checksums.
+2. Create an empty isolated restore environment with matching supported software.
+3. Restore configuration references/secrets separately.
+4. Restore PostgreSQL and apply only documented compatible migrations.
+5. Restore immutable blobs and parsed artifacts; verify sampled hashes.
+6. Restore Qdrant snapshots and collection aliases.
+7. Run reconciliation for missing/orphan points and artifacts.
+8. Run database integrity, counts, citations, golden retrieval and security denial tests.
+9. Record evidence and only then allow traffic.
+10. Keep the failed/current environment intact until restore acceptance.
+
+A production incident restore follows the same proof, with an explicit incident commander and rollback decision.