The "98% backend coverage, lint, PR review, runtime harness all green, and the column had no writer" line is the one every team running agent-driven delivery should tape to the wall. Coverage measures code that exists, and a schema with no producer is invisible to every check you listed, which is why write-path assertions (integration tests that insert then read back the exact new column) are the only thing that would have caught this in May. Do you now block a migration PR until a proof-of-write test lands with it, or handle it further downstream?