Why would our customers want mismatch datasets, and what should we tell them?

**Why does this workflow suit onward licensing?** It is common across finance, insurance, utilities, logistics and retail, so one dataset family can serve many of your customers' use cases.

**What do their teams do with it?** Typically three things: tune or prompt models that triage exceptions, build evaluation sets to compare model versions, and populate test environments for agent tools that read ledgers and post adjustments.

**Is it real data?** No. MirrorFirm generates the records. The generation is informed by direct observation of how businesses handle mismatches, but the output is synthetic. Your customers should still inspect grounding and disclosure risk, and you should not describe the material as real records or as cleared for any regulatory purpose.

**Why not just use clean transaction data?** Clean data contains no disagreement. A useful mismatch dataset carries the pair of records, the cause, the evidence trail, the applicable policy and the permitted resolution. Without those, a model can produce plausible answers but cannot be scored on correctness.

**What should we ask first?** Ask which mismatch causes are covered, whether root-cause labels are included, and whether the licence permits sub-licensing, derivative evaluation sets and use in model training. Confirm each in writing as part of scoping.