Rollback strategy
The predetermined plan for reverting a deployment when it goes wrong, re-deploying the previous version, flipping a feature flag off, or restoring data, defined before release rather than improvised mid-incident. A good rollback strategy is fast, tested, and decided in advance; the worst possible time to design one is while production is already down.
A rollback strategy is the difference between a bad deploy being a five-minute non-event and an all-night incident. The decision to define it belongs before the release, not during the outage, including how rollback interacts with database migrations, which is the genuinely hard part, and what signal triggers it. Teams that can roll back fast and safely can also deploy more often, because the cost of being wrong is bounded. It's a prerequisite for shift-right techniques.