What should happen when a sample exposes disagreement about task completion?

An evaluation lead reviews a generated professional services enquiry, CASE-1286. The request asks for specialist advice and includes enough information to identify the relevant practice team. The sample's target outcome assigns the enquiry to that team for assessment.

A domain reviewer marks the outcome incomplete because no one has accepted the prospective engagement. The evaluation lead and reviewer inspect the workflow boundary: intake staff can select a practice team, but only an authorised reviewer can decide whether the organisation accepts the work.

The pilot specification changes from “resolve incoming enquiries” to “establish the responsible assessment team”. The task's terminal state becomes `awaiting_practice_review`, with engagement acceptance outside the agent's authority. The sample and rubric are updated to reflect that boundary.

The consequence is a benchmark that measures an action the agent is actually permitted to complete. Without the correction, capable routing behaviour would receive a failure label for not performing a later decision. This sample review teaches evaluation teams to resolve disagreements about completion state before using generated answers as scoring targets.