What licensing questions and limits apply to a generated approval environment?

Licensing is agreed per engagement, so this answer lists what to ask rather than terms.

**Questions to put to MirrorFirm**

- Which uses does the licence cover: internal evaluation, model training, fine-tuning, publication of results, or redistribution of the data? - Can the data be used to train models that you then ship to customers? - Does the licence cover derived artefacts, such as graders built from the answer key? - What happens to your rights if you stop using the material? - Can you receive updated releases, and do earlier versions stay usable?

**Limits to expect**

- The output is generated and synthetic. It is not real business records, and MirrorFirm does not deliver any. - Generated data does not guarantee anonymity, privacy compliance or regulatory approval. If your agents will touch regulated processes, your own review still applies. - A generated authority matrix reflects a modelled organisation. It will not match your policies in every detail, and it should not substitute for your actual delegation instruments. - Coverage is limited to the domains and patterns scoped. Ask what is out of scope, in writing.

**Disclosure risk**

Ask how the generator guards against reproducing identifiable organisational detail, and what you can run yourself to look for it, such as searching for rare strings or unusual role titles.

Prices, timelines and service levels are set in the engagement and are not described here.