Data type guide

Prepare datasets and environments for AI licensing

Repeatable tasks with outcomes that can be checked.

Sources and guidance · Reviewed October 2026

What makes this collection useful

An evaluation environment is more than a pile of examples. It needs an initial state, permitted actions, success criteria and a way to reproduce the result.

Prepare before approaching buyers

Document task construction, ground truth, leakage controls and third-party dependencies. Separate train and evaluation splits by entities or time where appropriate.

Possible evaluation contexts

Agent evaluation; reinforcement learning; specialist benchmarks. These are possible uses, not statements of current buyer demand.

Where preparation breaks down

Contamination, unstable scoring, inaccessible dependencies and subjective success labels.

What to put in the datasheet

Task count; scoring method; reset instructions; version; known failure modes. State how each measure was calculated and which records were excluded. Keep the sample aligned with the collection described.

A practical next step

Create a one-page inventory, separate rights-cleared material from unresolved records and use the readiness checklist. Only then prepare a sample for a verified counterparty under appropriate evaluation terms.

Common questions

Does owning the files mean I can license them?

No. Ownership of the storage or account does not settle copyright, privacy, confidentiality or third-party restrictions. Review the rights in the content and the intended use separately.

Can I get a price without sending the full collection?

A high-level datasheet can support an initial discussion. A buyer may later request a reviewed sample and diligence; avoid sending raw archives at the first contact.

Use this as a preparation framework. It is not legal advice, a valuation or a promise that a buyer will accept your data.