Why is set-up work a better test of business competence than chat answers?

**Why not just ask a model how set-up works?**

A model can describe the process fluently and still fail to perform it. Describing is cheap. Doing requires picking the right record among near-matches, noticing that a condition is unmet and choosing a status that the downstream system will accept.

**What makes the result checkable?**

Every set-up task ends in state: a party record, an account with a status, linked products, open cases and notes. A grader can compare that state with what policy required. Two answers that read equally well can leave very different records behind.

**What kinds of mistakes does it expose?**

- Creating a duplicate party instead of linking to an existing one. - Activating before conditions are cleared. - Applying the wrong product tier or billing term. - Dropping information at a handover so the next team cannot proceed. - Treating a system rejection as success.

**Does it replace other evaluations?**

No. It complements knowledge and reasoning tests by adding multi-step action under constraints. A buyer would normally use it alongside existing suites.

**Is the material real?**

No. Everything MirrorFirm produces is generated. Ask for a sample and inspect it for grounding and disclosure risk before relying on it.