Permission and usefulness belong in the same review.
These are the requirements we scope for a licensing engagement. They are not a certification of every proposed source or a claim that a dataset has already passed review.
Rights and provenance
Record the source, collection period, authority to license, previous licensing and relevant third-party restrictions. A source business’s approval does not automatically cover its customers’ documents, contractor work or licensed content.
Privacy and confidentiality
Identify personal, sensitive and commercially confidential information before access or preparation. Establish the lawful processing arrangements and appropriate notices. Removing names alone does not establish effective anonymisation; pseudonymised records may remain personal data.
Connected, checkable cases
Assess whether the original task, contemporaneous context, actions and actual outcome can be linked. Distinguish observed evidence, human annotations and generated material. Keep future outcomes out of the starting context.
Measured quality
Agree completeness, accuracy, duplication, coverage and acceptance criteria with the buyer. Report what was checked, how and with what limitations. A large collection is not automatically useful or unique.
Training and evaluation separation
Where required, separate evaluation cases from training deliveries by appropriate source, task or time boundaries. Document overlap checks and avoid claiming that an untested collection is contamination-free.
Scoped handling and delivery
Initial enquiries contain descriptions only. Before any dataset moves, agree authorised access, preparation responsibilities, transfer method, retention, incident handling and deletion obligations. Controls and any certifications must be evidenced for the specific delivery arrangement.
The website collects qualification enquiries; it is not a dataset upload portal. Read the website privacy notice for how enquiry information is handled. A dataset engagement requires separate terms.