W3C PROV

PROV organizes provenance around three fundamental types:

• Entity: A physical, digital, or conceptual thing with a fixed identity (e.g., a dataset, a document version, a file).
• Activity: An action or process that occurs over time and generates or modifies entities (e.g., a data cleaning script, a human edit, a model training run).
• Agent: An entity that bears responsibility for an activity, software agent, or organization (e.g., a data engineer, a cron job, a specific microservice). These types form the backbone of any provenance assertion.

The power of PROV lies in its defined relationships that link the core structures:

• Generation: An Activity creates a new Entity.
• Usage: An Activity uses an existing Entity.
• Derivation: An Entity is transformed from, or updated from, another Entity.
• Attribution: An Entity is ascribed to an Agent.
• Association: An Agent is assigned responsibility for an Activity.
• Communication: An Activity exchanges information with another Activity. These relations form a Directed Acyclic Graph (DAG), preventing circular provenance loops.

PROV-O is the OWL2 ontology that maps the PROV data model directly to RDF (Resource Description Framework). This allows provenance graphs to be serialized as triples and integrated into existing knowledge graphs. Key benefits include:

• Semantic Reasoning: Infer implicit relationships (e.g., if A was attributed to B, B is an Agent).
• Interoperability: Merges seamlessly with other RDF-based standards like Dublin Core and FOAF.
• SPARQL Querying: Enables complex graph traversal queries to trace lineage across distributed systems.

PROV-N is a textual notation designed for human readability and illustrative examples. It provides a concise syntax to express PROV instances without the verbosity of XML or RDF/XML. For example: wasGeneratedBy(ex:report1, ex:compile_activity, 2024-05-01T10:00:00) This notation is crucial for documentation, debugging, and rapid prototyping of provenance chains before they are implemented in machine-readable formats like PROV-XML.

PROV includes advanced constructs for complex scenarios:

• Plan: An entity that represents a set of instructions or a script that an agent follows during an activity. This distinguishes between the code (Plan) and its execution (Activity).
• Bundle: A named set of provenance statements, allowing for provenance of provenance. A bundle can have its own creator and timestamp.
• Collection: An entity that groups other entities, enabling provenance tracking for aggregates like a directory of files or a database table.

Beyond the RDF-based PROV-O, the standard defines concrete serializations for different ecosystems:

• PROV-XML: A native XML schema for representing provenance in XML-based workflows and SOAP services.
• PROV-JSON: A JSON representation designed for web applications and RESTful APIs. It uses a simplified structure with keys for entity, activity, and agent objects, making it lightweight and easy to parse in JavaScript and Python environments.