Key takeaways
- Provenance records where data came from and how it became the value you are using.
- A source URL alone may not explain matching, transformations or later corrections.
- Preserve enough lineage to investigate a claim without exposing unnecessary sensitive data.
Overview
For enriched records, provenance often belongs at field level because different values can come from different sources. A useful record includes the source identifier, observation date, collection method and relevant transformations. Provenance supports audit and debugging, but it does not independently certify that the source is true. Users still need to evaluate the evidence.
How it works
Capture the source and observation context when information enters the system.
Preserve links through transformations, merges and derived values.
Expose enough evidence for a reviewer to reproduce or challenge the result.
Follow the value from source to destination
A company-size value may originate in a public document, pass through a provider estimate, be mapped into a size band and then appear in a CRM. Each stage changes what the value means or how it is represented. Provenance connects the final field to those stages so a reviewer can understand the evidence rather than assuming the displayed number was directly observed.
W3C’s PROV model provides a formal way to describe entities, activities and responsible agents. An operational workflow can start with a simpler record of source, observation time, transformation and destination. The useful depth depends on the questions the team must answer when a value is challenged or an integration behaves unexpectedly.
| Stage | Information to retain | Question it answers |
|---|---|---|
| Source | Reference, source record and observation date | Where did the claim originate? |
| Match | Input identifier and accepted entity | Why was it attached to this account? |
| Transformation | Rule and version used | How did the original become this value? |
| Delivery | Destination record and applied change | Where did the value affect another system? |
Source material: W3C — PROV overview ↓
Make conflicting evidence inspectable
When two sources disagree, preserve both observations and the reason for accepting one. The difference may reflect time, scope or method rather than a simple true-versus-false conflict. A reviewer checking company revenue needs to know the period and entity boundary; a reviewer checking employment needs the relevant observation date.
Avoid retaining only the winning value if later decisions depend on explaining it. At the same time, do not collect entire sensitive documents when a controlled reference and minimal supporting excerpt are sufficient. Apply access and retention rules to provenance records because logs and evidence stores can contain information as sensitive as the primary dataset.
Use lineage to repair affected records
Suppose an illustrative import used an incorrect country mapping. If each transformed value records the mapping version, the team can identify the affected records and reprocess only those. Without that lineage, the team may need to inspect every account or rerun expensive enrichment that was never the source of the defect.
Test whether a reviewer can trace a representative destination field back to its evidence and transformation. Also test whether a correction propagates without losing the earlier audit trail. Provenance should support real recovery and accountability, not merely add a source column that breaks when links change or records are merged.
What this looks like in practice
A company’s industry comes from a registry while its technology signal comes from a website observation. The record retains separate citations and dates instead of labeling both fields with one provider name.
Examples explain the concept; they are not reported customer results.What to check
Check whether an exported field can be traced to a specific source and transformation. Include derived scores and merged records in the audit sample.
Common mistake
Adding a general homepage link as a citation when it does not support the particular claim or show when the information was observed.
Data provenance vs. Data accuracy
Provenance explains origin and processing. Accuracy evaluates truth. Traceability makes accuracy checks easier but does not replace them.
Read the Data accuracy definition →Questions answered
What is Data provenance?
Data provenance records where information came from and the activities, transformations or responsible parties involved in producing it, allowing users to trace a result back to supporting evidence.
Does every field need a separate source?
When fields come from different evidence, field-level provenance is more useful. Shared source metadata can be reused where it accurately describes all included values.
What should happen when a source disappears?
Retain permitted identifiers, dates and relevant audit context. Mark the source as unavailable rather than silently replacing the citation with unrelated evidence.
Is a citation enough for provenance?
A citation is useful source evidence, but it may not explain which entity was matched, when the information was observed or how the value was transformed. Include the additional lineage required by the workflow. For consequential updates, the destination change and rule version can be as important as the original URL.
Should provenance be stored for every field?
Use a level of detail that fits the field’s consequence and the need for investigation. Important identity, qualification and delivery fields often benefit from field-level evidence. A single record-level source can be misleading when different attributes came from different providers or were updated at different times.
References and further reading
Primary documentation and source material for this topic. Sources checked September 14, 2026; provider requirements can change.
Continue reading on the blog
Explore all articles and guides →Put the concept to work.
Explore the relevant AstroFabric workflow and see how the pieces connect.
Help keep this guide useful. Suggest a correction or browse the full glossary.