What licence questions should a model lab settle before using generated material?

**Does access to a sample establish permission for every downstream use?**

A buyer should check the applicable licence rather than infer rights from the delivery method. A downloadable file or a response from a public MCP server can be technically accessible without answering every question about training, redistribution or publication.

**Which uses should a lab describe?**

Describe whether the material will support model training, internal evaluation, shared benchmarks, dataset transformation or contractor access. Ask how the licence treats derived task packs, adapted records and publication of evaluation examples. If generated content will enter a reusable internal corpus, include that use in the discussion.

**What should be confirmed about the deliverable?**

Confirm which artifacts the licence covers: records, schemas, task instructions, evaluator material and environment components may require explicit treatment. Ask about retention and permitted use of existing copies if access arrangements change.

These are scoping questions, not statements of MirrorFirm's terms. A useful first step is to provide a concise description of the intended data flow and request a licence that clearly addresses it before the material enters production training or evaluation pipelines.