Data quality acceptance criteria
Disputes after delivery usually come from undefined quality. Agreeing an acceptance protocol before the contract is signed is cheaper than arguing afterwards.
Defect classes
- Critical
ExamplePersonal data where none was declared; content without rights
Possible agreed treatmentMay trigger rejection or quarantine under the agreed protocol
- Major
ExampleWrong label, corrupt file, wrong language
Possible agreed treatmentCounts toward threshold
- Minor
ExampleFormatting drift, missing optional field
Possible agreed treatmentLogged, fixed in next drop
Protocol
- Agree a random sample drawn by the buyer, not curated by the seller.
- Set the major-defect threshold in writing before sampling.
- Define re-delivery windows and what happens on a second failure.
- Keep the sample and findings as part of the provenance file.
Have a dataset worth discussing?
Tell us what you own, which rights you can license and the commercial scope you have in mind.