Not to mention your deployment code itself can be buggy and is very difficult to write tests for. I actually got bitten by this recently. An automated deployment that I wrote years ago had an edge case race condition that could cause multiple deployments running from containers on the same docker host to collide, where the package from the first deployment would be pushed to the target of the second deployment. That deployment worked reliably for years, until one day it didn't. It was... a very stressful day.