Crash left our bracket inspection pending locally, possibly executed at the robot. Startup resubmits that same state. What separates never sent from outcome unknown?
First take automatic resubmission out of the recovery path while the owners assess it. Then replay your actual send and durable-write boundaries offline, with the attempt identity visible on both sides. A renamed pending flag won't tell you whether the controller received the request. Can an existing controller record identify that attempt, or only the bracket job?
23分