Startup schema setup carried old-data repair logic even though the project has not been deployed with customer data.

Replace the migration wrapper with a direct schema sync path and remove legacy backfill and column rewrite tests.

Database startup now initializes the current target schema without carrying historical compatibility code.

Publication status aliases kept old naming alive after the API contract moved to draft, succeeded, and cancelled values.

Remove the alias constants and switch services plus tests to the canonical publication status names.

Publishing code now uses one status vocabulary without compatibility indirection.

Multiple API and worker processes can call schema initialization during the same rollout.

Wrap the clean schema sync path in a Postgres transaction advisory lock without restoring legacy data repair logic.

Fresh database startup now avoids concurrent AutoMigrate and index creation races.

…mpatibility

The non-production Redis Cluster keeps TLS client authentication enabled for app-facing traffic.

Teach backend, browser-worker, and collab-service Redis clients to load client certificate and key files, then expose the paths in Kubernetes config and docs.

Cluster cutover can keep server-side mTLS without breaking app Redis clients.

The non-production Redis Cluster backup wrote topology files while dataset files stayed on pod PVCs.

Keep the snapshot trigger, then export RDB files from primary nodes into the backup PVC and update the drill notes.

Restore drills now have Redis data artifacts alongside Cluster topology metadata.