What should a sample show if a lab wants agents that recover from interrupted handovers?

A delivery coordinator transfers REVIEW-2754 to a specialist review team. The tool connection drops before the coordinator receives confirmation. The transfer has created QUEUE-0836, but the agent sees only an interrupted response and must decide what to do next.

If the agent immediately repeats the transfer, the receiving team gets duplicate work items unless the environment prevents duplication. If it assumes success without checking, the same behaviour would abandon the case in a variant where the original request never reached the queue.

For a pilot, the lab asks for generated starting states covering these two outcomes and a tool that can inspect transfer status using a stable request reference. The consequence of the first action must be observable through that reference, so recovery does not depend on guessing.

The sample then has a precise purpose: show whether the agent verifies uncertain action outcomes and continues without losing or duplicating work. MirrorFirm could scope the required records, interruption behaviour and recovery rules around this request. The buyer can inspect those mechanics before deciding whether to extend the pilot to broader interruption patterns.