Data Provenance Standards
By Gareth HoyleLast updated How we research
The Data Provenance Standards are a cross-industry metadata standard from the Data & Trust Alliance for recording where a dataset came from and how it may be used.
The Data Provenance Standards are a set of metadata standards published by the Data & Trust Alliance to describe the origin, lineage and permitted use of datasets, including AI training data. Version 1.0.0 was announced on July 9, 2024, after work that began in November 2023 with 19 member organizations.
What do the standards cover?
The standards have three parts and 22 metadata fields in total:
- Source, with 6 fields on who created the data and where it came from
- Provenance, with 8 fields on how it was collected, processed and changed
- Use, with 8 fields on legal rights, restrictions and intended use
The Alliance says future updates are to be managed by the standards body OASIS.
Should a data seller use the standards?
A seller can use the standards as a checklist when building a data manifest or dataset card, because buyers' diligence asks for the same facts. Following a published standard also makes a dataset easier to compare. See data provenance explained.