How is a sufficiency decision different from document extraction or classification?

**Question: Is this just extraction with a checklist on top?**

No. Extraction answers what a document says. A sufficiency decision answers whether the whole set is enough for the next step, and that depends on several things extraction does not cover.

- **Policy.** The requirement differs by product, jurisdiction or customer type, so the same file can be sufficient in one context and not in another. - **Absence.** The key evidence is often a document that is not there, which cannot be extracted. - **Validity.** A document can be present but expired, unsigned, illegible, or for the wrong entity. - **Consistency.** Two documents may each be fine but disagree on a name, address or amount. - **Action.** The output is a decision and, usually, a specific request to a named party.

A model can extract every field correctly and still decide badly, for example by treating a dated letter as current evidence. The reverse also happens: a model may judge sufficiency well from a short summary of each document.

For evaluation this means you should score the decision, the gap list and the request separately from field accuracy. A MirrorFirm task pack could include the same files with and without extracted fields supplied, so you can see which skill a result reflects.

All files in a release are generated. Ask for a sample to check how closely the policies and failure patterns match your use case.