Recover
Preserve a credible path to resume, restore, roll back, or reconstruct after interruption or failure.
Identification
Diagnostic questions
- What state must survive an interruption?
- Can a partial operation be resumed safely, or must it be restarted?
- Can a change be reversed without guesswork?
- What evidence is required to reconstruct what happened?
In practice
docmend keeps a manifest of what a run intended and what it achieved, which is what makes restore and resume possible rather than aspirational. Star Trek: Alter Course migrates saves forward across versions instead of rejecting them. Project Standards treats upgrading a repository that has fallen behind as a supported path rather than as a reason to start over.
Recovery is a design property, not a subsequent addition. Whether a partial operation can be resumed is decided by how state was written, long before anything goes wrong.
Background threads
Casualty control is recovery under time pressure: stabilise, establish what state you are actually in, and restore in an order that does not create the next casualty. What transfers is the expectation that abnormal conditions are normal enough to have been planned for.
Common failure
A backup nobody has restored from. The recovery path exists on paper and has never been executed, so its first execution is during the incident.