We are the licensor, so the risk sits with us.

Field of use

Priced and granted by use, not bundled by default. Evaluation and benchmarking, training and fine-tuning, and dataset construction are separable. A narrow field of use costs less and we would rather sell you the one you need than the one that sounds bigger.

Source of record: legal/Supplier_Licence_Standard_Terms.md

Term, exclusivity and sub-licensing

Default is non-exclusive with a defined term. Exclusivity is available per corpus, per field of use, priced accordingly, and we will tell you when it is already committed. Sub-licensing is a schedule, negotiated explicitly, never assumed.

Source of record: legal/Supplier_Licence_Standard_Terms.md

Reuse and renewal

Where a corpus is licensed again, for a further field of use or a further term, the source company is paid again. That is written into the upstream agreement before any file moves, which is what makes suppliers willing to hand over a real operating record rather than a sample.

Source of record: legal/Supplier_Licence_Standard_Terms.md

What we warrant

That we hold the rights we are granting and maintain them for the term. That we validate provenance through a documented process rather than by assertion. That the de-identification standard was applied and the result was reviewed by two people. That the data was obtained consistently with the terms of service of the systems it came from. That nothing in the delivery is scraped or sourced from the public internet.

Source of record: legal/Provenance_and_Chain_of_Title_Record.md

Where liability sits

Indemnity for the rights chain is ours. We carry it because we are the party that can actually control it: we hold the supplier agreements, we run the exclusion screen and we operate the de-identification. Limits are set in the licence and matched to insurance in force. We will not publish an uncapped promise, and you should be careful of anyone who does.

Source of record: legal/Entity.md

Audit

You may audit the provenance record for any corpus you licence, once in twelve months, through an independent professional under confidentiality. The record is built to be read by someone who did not create it.

Source of record: operations/Chain_of_Custody.md

What we do not do

• We do not sell a list, a feed or a subscription. Every licence is a negotiated instrument against a specific manifest.
• We do not deliver a corpus whose rights record is incomplete, whatever the commercial pressure.
• We do not licence personal information. What we licence is de-identified before it leaves our systems.
• We do not name our suppliers to buyers, or our buyers to suppliers.