Will a release include the documents themselves, or only metadata about them?

**Can a release include document content?**

Yes, it can be scoped that way. A buyer can ask for metadata only, for generated text content, for structured renderings such as key-value fields, or for a combination. All of it is generated, not copied from real records.

**Which should I choose?**

Match the representation to what your agent reads in production.

- If the agent receives output from an upstream extraction step, structured fields with confidence values and legibility flags may be enough. - If the agent reads raw text, ask for generated text with realistic faults such as truncated pages, repeated headers and mixed document types in one file. - If it works from images or PDFs, discuss in a pilot what format is feasible, because file rendering is a separate scoping question.

**Does metadata-only data still test sufficiency?**

Partly. It tests presence, age and type checks well. It tests consistency checks poorly, because a name or amount mismatch needs content to compare.

**What should I inspect?**

Read a sample of generated documents for anything that looks too specific to a real source, such as unusual combinations of details. Generated output does not remove the need to review grounding and disclosure risk, and it does not stand in for real records.