Contracts & knowledge · Provenance model
W3C PROV
Describe the entities, activities, and agents involved in producing an artifact.
What it helps you do
Let someone inspect how a result relates to inputs, transformations, and responsible actors.
Starting knowledge: No specialist knowledge is required.
- Identify the artifact and its source entities.
- Record the activity that used inputs and generated an output.
- Identify responsible agents and preserve version or revision references.
The problem it came from
The W3C PROV family was published in 2013 to support interoperable provenance information. It includes a conceptual model, ontology, and serializations.
The overview distinguishes information about a thing’s production from an automatic verdict about its trustworthiness.
Work through an example
Entity: workshop-policy revision 7
Activity: summarize-policy run 12
Agent: reviewed authoring process
Activity used: workshop-policy revision 7
Activity generated: participant-summary revision 2
Open check: confirm that cancellation exceptions survived the summary.- The source has a revision rather than an ambiguous freshness claim.
- The transformation and output are identified.
- The unresolved meaning-preservation check remains separate from lineage.
Where it helps. Where it stops.
Use it when
- A report combines data from several sources or transformations.
- An AI-generated result needs a traceable connection to supplied evidence.
Potential benefits
- Provenance helps a reviewer reconstruct dependencies and origin.
- Versioned inputs can reveal when an artifact may need revisiting.
Limits and common mistakes
- A provenance record can be incomplete or false.
- Knowing where a claim came from does not prove the source or transformation was correct.
The prompt below asks for this structure. It does not guarantee accuracy, complete coverage, better learning, or conformance to a standard. Treat those outcomes as separate checks.
Practise the judgment
Try a choice, inspect a boundary, then explain a case yourself. Hints and answers are available whenever you need them.
Which provenance statement is more inspectable?
Hint
Can another person identify the same input?
Explained answer
Best answer for this case: Used policy revision 7 to generate summary revision 2
A specific revision supports tracing and reproduction. “Latest” can change meaning immediately.
Does an accurate lineage record prove the summary preserved every exception?
Hint
An identifiable transformation can still lose information.
Explained answer
Best answer for this case: No; preservation needs separate checking
Lineage and semantic adequacy are separate properties.
An input changes after a report was generated. What should the provenance let a reader discover?
Hint
What would a reviewer need before reusing the report?
Explained answer & self-review rubric
Which revision the report actually used, which parts depend on that input, and whether any review against the new revision has occurred. A source change is a reason to inspect freshness, not proof that every statement became false.
- Retain the old input revision
- Expose dependency
- Avoid equating change with certain invalidity
Take it into your own work
Manual template
Use this structure yourself, with a colleague, or in your preferred writing tool.
# Provenance record
Output entity and revision:
Input entities and revisions:
Activity or transformation:
Responsible agent:
Observed time, if actually recorded:
Evidence locations:
Review performed:
Unverified claims:
Review the result
- Can the exact inputs be found?
- Is observed provenance distinguished from inferred lineage?
- What remains unverified despite the record?
Use this approach with your own AI tool
Help me apply W3C PROV to the material below.
Reader and existing knowledge: [describe]
Task and desired outcome: [describe]
Use these principles: Identify the artifact and its source entities.; Record the activity that used inputs and generated an output.; Identify responsible agents and preserve version or revision references..
Preserve source facts, identifiers, quantities, conditions, exceptions, and uncertainty.
Separate supplied facts from assumptions. Ask for missing information instead of inventing it.
Explain your representation choices and provide a review checklist. Do not claim formal compliance or correctness from formatting alone.
Material: [paste only information you are permitted to share]Go to the source
History and the underlying approach are grounded in these references. Examples, prompts, and practice cases here are original educational adaptations. Edition-specific details should be checked against the referenced publication.
Last editorial review: 2026-10-24. An overview is not certification, and this guide is not a reproduction of a standard.