What should an evaluation team clarify about scope and licensing before starting?

Start with the evaluation decision the team needs to make. Comparing candidate agents, training a model and publishing a benchmark involve different uses of the same generated material. State which activities are intended before discussing a release or licence.

A scope request should distinguish internal evaluation, training, modification, redistribution and public reporting. Clarify whether contractors or external model services will process the material, whether prompts or task excerpts will appear in reports, and whether derived task variants will be shared. These are matters to agree explicitly, not permissions to infer from access to a sample.

The technical scope should identify the business domain, policy boundaries, interaction mode and expected evaluation outputs. A static dataset may support decision classification, while a simulated environment is needed to assess actions that change records.

MirrorFirm could prepare generated material against that scope, subject to agreed terms. Ask which licence conditions apply to samples and subsequent releases, and how those conditions travel with exports. A useful first step is a written use-case brief pairing each intended activity with the artefacts and rights it requires. That makes both the evaluation design and the commercial discussion concrete.