Recovery Matters as Much as Execution
Infrastructure automation has to survive interrupted commands, partial completion, stale state, and asynchronous work without blindly doing things twice.
One of the easiest ways for automation to damage an environment is to repeat an operation simply because it did not receive the response it expected. A command may have started a package installation successfully, the connection may have dropped, and the automation system may still be tempted to run the same operation again.
STRATUM tracks persistent Missions and dependency graphs so interrupted work can be investigated before it is repeated. The AI can observe the current guest or infrastructure state and determine whether the previous action completed, partially completed, failed, or is still progressing. That state becomes new evidence for the next decision.
This becomes essential when independent parts of a Mission execute concurrently. A failure on one branch should not automatically destroy progress on unrelated branches. STRATUM can keep successful work intact, isolate the failed path, repair what actually needs repair, and then continue from the current state instead of replaying the entire mission from the beginning.
The same discipline applies at completion. STRATUM separately verifies the requested outcome instead of equating a completed command with a completed mission. Recovery and verification are both part of the execution model because real infrastructure is rarely a perfectly linear sequence of commands.