Continuous Deployment

Continuous deployment releases every change that passes automated validation, without a human release step. Continuous delivery is the weaker sibling: every change is made releasable, and a person chooses when to press the button.

The practice depends entirely on the confidence of the validation, which is why it tends to arrive together with strong testing, Feature Flags, and quick rollback. Flags matter most: they decouple deploying code from enabling behaviour, so shipping unfinished work becomes safe and releasing becomes a configuration change rather than a deployment.

Deployment is also where a whole category of environment-specific failure lives, and it is the category least covered by ordinary checks. A database migration that runs in a fresh environment and fails in the deployed one; a required secret that exists locally and not in production; a platform-provided variable under a name the application does not read; a certificate that has to be provisioned separately. None of this is visible to a type checker or a test suite.

The only honest verification is to reproduce the deployment path end to end — reset, migrate, build, start in production mode, and request the real thing. See Database Migration and Ground Truth.