Core answer
Start here. This is the direct spoken answer to practice first.
Why this question matters
Bad releases are where release theory becomes real. This drill matters because the production behavior is shaped by platform configuration, code boundaries, and the operational path the team will use during a real incident.
I design the App Service release so rollback is a real option before an incident. The exact artifact is deployed to a staging slot, warmed, smoke-tested, and swapped with known slot-specific configuration. Database and message changes remain backward compatible while old and new versions may overlap. A feature flag can separate deployment from exposure, but it does not make an incompatible schema or external side effect reversible.