How can a late internal task create a meaningful agent evaluation?

A retail stock controller owns count task TASK-2841. The count is due before a replenishment planner can release order ORD-6138, but the task remains open after the policy threshold. A scheduling assistant sees the late status and prepares a reminder.

The task history contains a reassignment: the stock controller's shift has ended, and responsibility now sits with the duty supervisor. A handover note also records that one storage area is inaccessible. The policy requires the current owner to receive a request for either completion or an exception decision.

The agent links the blocked order to the count task, contacts the duty supervisor and asks for the missing decision. The supervisor records an approved partial count. TASK-2841 changes from overdue to exception accepted, allowing ORD-6138 to proceed with the affected stock lines excluded.

This scenario teaches that useful follow-up addresses the reason work is blocked and the authority needed to unblock it. MirrorFirm could generate this dependency chain as a task pack. Evaluation teams can assess whether the agent finds the current owner, preserves the unresolved stock restriction and avoids treating a reminder alone as completion.