W3C PROV

Defines the fundamental building blocks for representing provenance as a directed graph. The model is built on three primary node types:

• Entity: A physical, digital, or conceptual thing with a fixed aspect, such as a dataset, a document, or a software version.
• Activity: An action or process that occurs over time and generates or modifies entities, like a data transformation script or a model training run.
• Agent: An entity that bears responsibility for an activity, which can be a person, an organization, or a software service. These nodes are connected by binary relations such as wasGeneratedBy, used, wasAttributedTo, and wasDerivedFrom to form a complete lineage graph.

A formal OWL2 Web Ontology Language specification that maps the PROV Data Model to a set of classes and properties for the Semantic Web. This enables:

• Interoperability: Linking provenance assertions across different systems using RDF triples.
• Reasoning: Inferring implicit provenance relationships through ontological logic.
• Serialization: Representing provenance graphs in standard formats like RDF/XML, Turtle, and JSON-LD. The ontology defines a hierarchy of classes, starting with prov:Entity, prov:Activity, and prov:Agent, and their corresponding object properties.

A human-readable, functional-style syntax designed for writing and illustrating provenance examples directly in technical specifications and documentation. It provides a concise way to express PROV statements without the verbosity of XML or JSON.

• Example: wasGeneratedBy(ex:report, ex:analysis, 2024-05-20T10:00:00Z)
• Purpose: Simplifies communication between engineers and domain experts when modeling lineage.
• Mapping: Every PROV-N expression has a direct equivalent in the PROV-O ontology, ensuring no loss of semantic fidelity.

PROV supports multiple concrete serializations to ensure integration with diverse enterprise architectures:

• PROV-XML: An XML schema for systems that rely on SOAP or XML-based message queues.
• PROV-JSON: A lightweight JSON representation optimized for web APIs and modern data pipelines.
• RDF/XML & Turtle: Standard Semantic Web formats for linking provenance to existing knowledge graphs. This multi-format approach allows a single provenance graph to be consumed by a legacy data warehouse, a cloud-native microservice, and a graph database simultaneously.

Defines a formal set of structural and event-ordering constraints that valid provenance instances must satisfy. This ensures logical consistency across the provenance graph:

• Typing Constraints: An entity cannot be generated by an activity that started after the entity's generation time.
• Ordering Constraints: The start of an activity must precede its end, and usage of an entity must occur within the activity's lifespan.
• Impossibility Constraints: Prevents circular derivations where an entity is indirectly derived from itself. These constraints enable automated validation of provenance integrity before ingestion into compliance systems.

A mechanism for packaging a set of provenance statements as a named Bundle, which is itself an entity with its own provenance. This enables:

• Provenance of Provenance: Tracking who asserted a lineage record and when, critical for auditability.
• Modularity: Combining provenance from multiple sources, such as a data lake, a training pipeline, and a human annotator, into a single coherent record.
• Attribution: A bundle can be attributed to a specific agent, establishing accountability for the provenance assertions themselves, not just the underlying data.