How it works

Copy-on-write isolation with shadow tables.

Flow

Schema session_{id} created

Reads: production + staged. Writes: shadow tables.

Submit for review. Writes blocked.

Merge atomically or drop schema.

First write creates _shadow_{table}. All writes go there.

Reads combine production + shadow. Agent sees consistent state.

Prevent race conditions during merge.

Detects conflicts by comparing hashes at merge time.

SELECT	Union view (production + shadow)
INSERT	Shadow table
UPDATE	Each affected row copied to shadow, then updated
DELETE	Mark deleted in shadow metadata

Row hashes are compared at merge. If production changed, you can force apply, reject, or review per-row.

1. Acquire lock→2. Verify hashes→3. Check FKs→4. Apply→5. Drop schema