What makes a receipt-matching case worth including in an evaluation set?

Use this list to decide whether a payment case earns a place in your evaluation set.

- The correct allocation can be stated from the records and policy alone. - At least one plausible wrong answer exists, such as a similar invoice number or a related customer. - The cost of a wrong action differs from the cost of holding the payment. - Evidence sits in more than one record, such as the bank line, the ledger and a remittance note. - The case needs a decision about apply, hold or escalate, not only a lookup. - A later event, such as a reversal or credit note, can change the right answer. - The resulting ledger state can be checked without human interpretation. - The case represents a type of exception that a business team handles routinely, rather than a puzzle invented for difficulty.

A set that passes most of these tends to separate agents on judgement rather than on formatting. When reviewing a MirrorFirm sample, apply the same list to a handful of cases and note which ones fail it. The failures tell you what to ask the provider to adjust, and they tell you which parts of the sample you can use as they are.