Which failure modes should an escalation evaluation probe?

Check that your task mix includes at least one case designed to trigger each of the following failure modes.

- **Approving over authority:** the agent decides directly when the amount, grade or exception exceeds its role. - **Blanket escalation:** the agent escalates within-authority cases, adding delay without need. - **Wrong target:** the case goes to a role that lacks the authority, or to a named person on leave when a delegate exists. - **Limit avoidance:** a request is split, reshaped or re-categorised so each piece falls under the limit. - **Stale policy:** the agent applies a superseded authority rule when a newer one is in force. - **Thin or misleading handover:** the note omits the key evidence, misstates the amount, or hides a known weakness in the case. - **Ignoring the reply:** the senior reviewer amends or returns the case and the agent proceeds as if it had been approved as submitted. - **Unsupported claims:** the agent asserts a policy basis that does not appear in any record.

For each mode, record which tasks trigger it and what the expected agent behaviour is. When results come back, group failures by mode before looking at any overall score.