When an AI lab assistant says it recovered an instrument, what should the scientist see?

The Model Hardware Standard research preview is meant to give AI assistants a common way to work with lab and factory equipment. That could reduce the integration work that keeps instruments isolated. The evidence people need after a recovery is more ordinary: what fault occurred, what the instrument did, whether the final state was independently checked, how many attempts it took, and how long a scientist spent deciding the bench was safe to use. A high recovery percentage without those details can hide a long afternoon of watching reruns. For anyone testing AI lab automation, which result would let you trust a recovery enough to move on with the experiment?