How should we think about scope and licensing when starting with an intake release?

Scope follows from what you want to learn, so start from your evaluation question and work back.

**Pick a narrow slice.** One domain, a small set of case types and a handful of queues give you something you can inspect by hand. Breadth can follow once the first slice behaves as you expect.

**Choose the form.** A dataset suits offline analysis and fine-tuning experiments. A task pack suits repeatable evaluation with answer keys. A simulated environment suits testing actions and state changes. They can be combined, but each adds review effort.

**State intended use.** Say whether you plan to evaluate only, to train or fine-tune, or both, and whether you intend to publish results, share the material with partners or embed it in a product. These drive licence terms.

**Agree licence terms in writing.** Terms, pricing and timelines are settled per engagement, so ask for them to be stated explicitly, including permitted use, redistribution, derived models and publication of results. Do not assume terms from other releases.

**Know the limits.** The output is generated, not real business records. It is useful for testing decisions and workflows, but it does not replace validation on your own data, and it carries no guarantee of privacy or regulatory outcomes. Inspect grounding and disclosure risk in the sample.