When an automated system loses the result of an infrastructure change, the original operation may still be underway. Sending the instruction again can create another attempt while the first remains unresolved. A useful recovery path needs to preserve the original request, account for what may already have started, and establish who can authorize the next step.
We’ve implemented and tested two connected safeguards in a bounded infrastructure experiment. Before dispatching the particular change under test, it saves the original request and records the attempt. Recovery treats that attempt as potentially submitted. Reopening the saved initial-attempt record does not authorize a fresh submission, even when a new process or a new authority decision is available.
Continuing from a verified intermediate state is a separate path. It must preserve the exact original request, establish closure of the prior invocation and use a current authorization decision. If authority has changed, that decision must explicitly adopt the same original request. Missing or inconsistent history prevents continuation; a restored record cannot silently become a new request.
The qualification combines modeled early-interruption and record-reopening cases with a controlled process-stop test later in the operation. It checks permitted continuations as well as refusals. The changes have passed review and tests on our development branch. Authorization in these tests is simulated; this is not a deployed infrastructure-management service.
These safeguards do not undo work already accepted. An operation admitted before authority changes may still finish, so declining another attempt is not the same as cancelling the first one. The experiment does not establish immediate revocation or duplicate prevention across independent systems.
For Duploflow, this is progress on the action side of automation: moving the request record ahead of the first dispatch and making continuation accountable to that same request. It gives the experiment a tested boundary for deciding what may happen next without treating an interruption as a fresh start.