Why do model labs need conditional approvals rather than plain approve or reject data?

**Why not just use approve or reject data?**

Binary decisions hide the part that causes errors in deployment. In a business, most approvals are not clean. They arrive with requirements attached, and the requirements are checked later, by different people, against different evidence. A model trained only on outcomes learns that 'approved' leads to 'proceed'. A model that has seen conditional states learns that approved work may still be blocked.

**What capabilities does it exercise?**

- Reading a decision together with its attached obligations. - Tracking state across time, since conditions are satisfied at different moments. - Judging whether evidence meets a condition, not just whether a document exists. - Recognising authority: who may waive, extend or vary a condition. - Writing a handover that tells the next person what remains open.

**Is this only for finance?**

No. The same structure appears in insurance underwriting, utility network connections, healthcare administration, logistics and professional services. A dataset can vary the domain while keeping the underlying gating logic, which helps a lab test whether behaviour transfers.

**Is the data real?**

No. MirrorFirm's output is generated, informed by observation of how businesses operate. Buyers should inspect how well it is grounded and check for disclosure risk as part of evaluation.