Audit my JSON/JSONB columns and schemaless documents for structure drift. Find every column or field that stores arbitrary JSON and check what's actually been written into it. For each: sample the stored values and report the shape variations found (missing keys, type drift like string-or-number, renamed fields across versions); define the expected shape and validate it at write time going forward, either with database-level validation or a single shared validation function; add size limits so a JSON field can't grow into a megabyte blob that slows every row read; and check whether anything queries into the JSON, because queried keys may need to become real columns or get a proper index. Report any data already stored in the wrong shape and write a cleanup migration. Don't delete variant data you can't interpret; report it. Don't convert JSON fields to columns wholesale; only where querying justifies it. Keep the fix proportional: validate at write, index what you query, cap the size. Deliver: the drift report per JSON field, the validators added, the cleanup migration for existing bad rows, any indexes added, and verification: attempt writes with wrong shapes and confirm rejection.