Nought Point Three Per Cent Of Rows Disagree
Decide what you would do from the brief alone, including whether you would change anything at all. Everything below it is available, but the exercise stops working if you open it first.
An expand-and-contract migration is moving appointments to a new schema. Dual-write has been running for three weeks. A sampling job reports that about 0.3% of rows differ between old and new — always in the fee fields, always for appointments touched by the admin console.
Backfilling the differing rows and cutting over — 0.3% is small, the backfill is one script, and the divergence stops by definition once the old table is gone. The backfill genuinely reconciles the rows and the cutover genuinely succeeds; the admin console starts failing four days later on the path that writes directly, by which time the old table has been dropped and the rollback plan assumed it existed. The reasoning error is treating a residual difference as data quality when it is a directory of unknown writers, and the specific danger is that the trap works right up to the point where it cannot be undone.
Read this even if you are confident. It is here rather than behind a button because it is the answer most teams actually ship, it passes review, and the cost of it does not arrive until the change after this one.