Data provenance
Data provenance is information about the origin and production history of data. It can identify source entities, activities, and responsible parties. W3C’s PROV family provides a model for expressing such relationships. Provenance is broader than a file’s location and does not automatically establish legal permission.
How it works
Capture the source, collection context, dates, methods, and responsible systems or organizations. Link transformations and derived artifacts to their inputs where possible. A useful record distinguishes direct observations from estimates, imported information, and later corrections.
Provenance and lineage overlap: provenance emphasizes origin and responsibility, while lineage commonly emphasizes movement and transformation through systems. Both can contribute to a reliable dataset history.
Why it matters for licensing
A recipient may need to assess whether records reflect the intended activity, whether collection methods changed, and whether the provider can explain the source. Provenance can support rights due diligence, but the agreements or permissions themselves still need separate examination.
Example
Fictional example: A dataset identifies the application that generated each service event, the extraction date, and the process used to attach an outcome label. Reviewers can trace a questionable record back to its source rather than guessing from the final file.
Limitations and misconceptions
A recorded source can be incomplete or wrong. Technical signatures and metadata support particular checks, not universal truth or ownership claims. Documentation should make gaps visible rather than implying a complete chain when one is unavailable.
Questions to ask
- Can each data component be traced to a documented source and collection method?
- Which transformations and responsible parties are recorded?
- What separate evidence supports the proposed licensing rights?
Sources
- W3C — PROV Overview · Accessed
- Gebru et al. — Datasheets for Datasets · Accessed
Explore whether your business data could be a fit.
Start with a description of your systems—not a data upload.