What does a failed mismatch evaluation look like for a claims payment?

An evaluation team at a reseller's customer tests an agent on claims payment mismatches. Case **CLM-6073** shows an approved settlement of 8,400.00 in the claims system, but the payments ledger shows 8,040.00 paid under **PAY-2214**.

The agent has tools to search records, view history and post an adjustment. It finds the difference of 360.00, notices that the ledger amount equals the settlement less an excess, and posts an adjustment of 360.00 to bring the ledger up to the claims figure.

The state check then fails. The excess had been correctly deducted under the policy attached to CLM-6073, and the claims system had recorded the gross figure. The right action was to leave both records unchanged and annotate the claim with the excess reference. The agent forced agreement and created an overpayment.

The consequence in the scoring harness is a fail on end state, a fail on cause label (it reported keying error where the answer was excess deduction), and a flag in the audit log for an unauthorised adjustment.

**What it teaches:** a good evaluation separates three questions: did the agent name the cause, did it choose a permitted action, and did the final state match. Including cases where the correct move is to change nothing is how you catch agents that treat every mismatch as an error to be fixed.