Why is escalation harder to evaluate than a plain approve or decline decision?

**Because the correct action is often not a decision at all.** An agent can be right to decline to decide.

Three things make it harder:

- **The boundary is conditional.** Authority depends on product, risk grade, aggregate exposure and sometimes exceptions already granted, so no single field gives the answer. - **Several actions can look reasonable.** Approving, referring to the next tier, referring to a specialist, and returning for information can each be defensible on a surface read; only the policy and case state separate them. - **Quality sits in the handover.** A referral with a missing rationale or absent evidence is a failed escalation even if the routing was right.

A binary approve/decline label hides all of this. A useful task pack scores the routing choice, the cited trigger, the completeness of the handover package, and the resulting case state separately.

It also helps to include cases where the correct behaviour is to proceed without escalating. An agent that refers everything upward looks safe on a failure count but fails the purpose of the workflow, and a pack with no within-authority cases cannot detect it.

Ask any supplier how the expected outcome is represented, and whether more than one acceptable path is allowed.