What should we ask about licensing, limits and intended use before starting?

**Can we use the data for both training and evaluation?** Ask this directly. Intended use affects how the release is structured, because evaluation tasks should be kept separate from anything used to train or tune.

**How are holdouts handled?** Agree which tasks are held out, who can see expected snapshots and labels, and how you will stop them leaking into training data or public documents.

**Can we share the data or results?** Clarify whether internal teams, contractors and external reviewers can access it, and whether you can publish aggregate results. Licence terms are set per engagement, and this page does not state them.

**Are there limits on what can be generated?** Output is generated, so it should not be treated as real business records. It does not guarantee anonymity, privacy compliance or regulatory approval, and you remain responsible for your own assessment of fitness for purpose.

**What about price, timing and support?** These are discussed during scoping. Do not assume them from a sample.

**What is a sensible first step?** Request a small sample with documentation, review it against your own checks, and use that review to define the pilot. Keep questions about versioning and updates on the list, so a benchmark does not change under you unnoticed.