What licence questions matter before using generated material in a benchmark?

**Does access to a generated sample establish all usage rights?**

Do not assume it does. Ask which uses the proposed licence would cover and obtain the applicable terms before moving beyond the agreed sample use. No particular permission should be inferred from the material being generated.

**Which uses should evaluation teams describe?**

Describe internal evaluation, model training, agent development and any external benchmarking. Explain whether contractors or collaborators need access, whether records will be supplied to hosted model services and whether the task pack will be embedded in a reusable evaluation product.

**What about derivative artefacts?**

Ask how the proposed terms address modified tasks, translated records, annotations, execution traces and trained model artefacts. Evaluation traces may reproduce substantial source content even when the published benchmark report contains only aggregate findings.

**What should be clarified before publication?**

Specify whether you intend to publish prompts, record excerpts, full tasks, answer keys or downloadable packages. Ask about attribution, redistribution and any restrictions relevant to those outputs. These are questions for the licensing discussion, not statements of offered terms. A clear intended-use description helps identify mismatches before the pilot becomes part of a wider research or product workflow.