Glossary
Provenance
Provenance records the origins, transformations and responsible actors associated with information.
Provenance and citation
Provenance explains how a piece of information came to be. It can identify source material, activities that transformed it and agents responsible for those activities. W3C PROV provides a common model for describing entities, activities and agents, but a useful provenance record need not use one particular serialization. The central question is whether a reviewer can follow a claim back through its material history.
In practice, provenance may connect a displayed value to an extracted sentence, a source document, its version and later review decisions. Some links express direct quotation; others express derivation or interpretation. These should remain distinguishable. A URL alone is weak provenance because content can move or change, and a citation to a whole document may not identify the passage that supports the claim.
Organizations defending decisions need provenance to explain both what was known and how it became an accepted value. Data lineage describes a related path through processing steps; attribution identifies a particular source or creator; chain of custody focuses on possession and integrity. Provenance covers the broader account without making the underlying source automatically reliable.
Example
A Halden Therapeutics AG (fictional) status traces from a displayed field to a reviewed claim, a dated excerpt and the retained notice containing it.
The PROV model in brief
The W3C provenance family, introduced in PROV-Overview, gives a shared vocabulary for describing where something came from. Its data model has three core types: entities, which are things such as a document, a dataset or a value; activities, which are things that happen over time, such as extracting, merging or reviewing; and agents, which are people, organisations or software responsible for an activity. These are related by statements such as an entity being generated by an activity, used by an activity, derived from another entity, or attributed to an agent. The PROV-O ontology expresses the model in OWL so that provenance can be held as RDF alongside the data it describes. The NIST glossary defines provenance in the security context in compatible terms.
The model's value is mainly in agreeing on what to record. A system that can answer which entity a value was derived from, by which activity, using what inputs, under whose responsibility and when, has the facts needed for most audit and dispute questions. The article on W3C PROV for LLM context provenance proposes a testable profile for applying this to the material given to a language model, and the one on minimum provenance requirements lists what a value needs in order to be defended.
Granularity and cost
Provenance can be recorded for a whole dataset, a table, a row, a field or a single statement. Finer granularity answers finer questions and costs more to store and maintain. A workable approach is to record fine-grained provenance for values that feed decisions, and coarse provenance for the rest, and to be explicit about which is which. Provenance about statements in a graph has been awkward in RDF for years; the article on RDF 1.2 provenance explains how the newer specification handles statements about statements.
Provenance is itself data and can be wrong, forged or lost. It should be created at the time of the activity, by the system performing it, and not reconstructed later. It should be protected against alteration, which is the subject of append-only logs and chain of custody. And it should be checked: sample a set of values, follow their provenance to the sources, and confirm that the sources say what the record claims.
A worked example
A dashboard shows that a trial of Halden Therapeutics AG (fictional) began on 4 March. The provenance of that value is a short chain. The value is an entity derived from a registry page entity, retrieved on 12 May by a collection activity run by a named job, and attributed to the registry as agent. A review activity on 13 May by a named analyst approved the value. The page itself is held as a retained copy with a digest. A reader who doubts the date can follow the chain, open the copy, read the passage and see who approved. If the registry later changes the date, the chain still shows what was read on 12 May, so the dashboard's earlier figure can be explained and not silently replaced.
Related terms and boundaries
Provenance is the general idea; data lineage is its application to data pipelines, attribution is its focus on responsible parties, a retained source copy is its evidence, and chain of custody is its integrity over time. An evidence span pinpoints the part of a source that matters. Provenance supports, but does not establish, the reliability of a source.
Common misreadings
- Treating a URL as a complete evidence record.
- Equating provenance with source credibility.
- Losing transformations between source and final value.
Appears in
- Data catalog vs system of record
A catalog tells users what data exists and how to find it; a system of record governs accepted values within a defined scope.
- Document citation vs sentence citation
A document citation identifies an artifact; a sentence citation locates the specific statement supporting a claim.
- Golden record vs evidence-carrying record
A golden record presents preferred values; an evidence-carrying record keeps the dated source sentence with each value.
- Knowledge graph vs document store
A graph represents entities and relations explicitly; a document store preserves documents as units of content.
- Retrieval-augmented generation vs a system of record
Retrieval supplies context for an answer; a system of record governs which values may be treated as settled.
- Citing at the sentence
Bind each assertion to the exact source passage, its locator, reading date and retained copy.
- Declaring unknowns
Distinguish not found, blocked by licence and not applicable, with a dated search boundary.
- Reporting source conflicts
Preserve competing values and their grounds until a scoped decision can be defended.
- Resolving identity by rule
Make each same-entity or different-entity decision explicit, evidenced, reviewed and reversible.
- Retaining source copies
Preserve the material actually read, subject to rights, and bind its citation with a content digest.
- Scoping a first record
Start with one contested entity and a bounded decision, then test whether its evidence can be defended.
- Separating reader clearance
Specify who may inspect each value and its evidence, including passages that disclose restricted material.
- Central Index Key (CIK)
The SEC's identifier for a filer in its EDGAR submission system.
- ClinicalTrials.gov identifier (NCT number)
A stable identifier for a clinical study record on ClinicalTrials.gov.
- Common Procurement Vocabulary (CPV)
A controlled classification for the subject matter of public procurement in Europe.
- Data Catalog Vocabulary (DCAT)
An RDF vocabulary for describing datasets and catalog distributions.
- Digital Object Identifier (DOI)
A persistent identifier for a registered digital object or other referent.
- Economic Operators Registration and Identification (EORI)
An EU customs identifier for economic operators and, where relevant, other persons.
- European Case Law Identifier (ECLI)
A common identifier format for European judicial decisions.
- Global Trade Item Number (GTIN)
A GS1 identifier for trade items at a defined packaging or presentation level.
- Harmonized System (HS) code
A hierarchical commodity classification used as the basis for customs tariffs.
- JSON-LD
A JSON-based serialization for linked data and RDF-compatible statements.
- Legal Entity Identifier (LEI)
A globally issued identifier for a legal entity participating in financial transactions.
- Resource Description Framework (RDF)
A graph data model for statements about identified resources.
- Schema.org
A shared vocabulary for structured descriptions of things on the web.
- Unique Ingredient Identifier (UNII)
An FDA-assigned identifier for a substance based on its defined identity.
- VAT identification number
A jurisdiction-specific tax registration identifier used in value-added tax administration.
- W3C PROV
A family of specifications for representing provenance of entities, activities, and agents.
- Wikidata item identifier (QID)
An identifier for a Wikidata item in a collaboratively maintained knowledge base.
Found an error on this page? See corrections for what counts as one, how to report it and the dated log.
Enquiry
Where does "Provenance" bear on a decision of yours?
This entry sits under provenance and citation. If a decision in your organization turns on it, send the entity concerned, the sources you fuse to describe it and who may read which of them. We reply with what a first record would cover.
Or write to [email protected] in your own words.