How Sgraal's MemCube v4.1 provenance chain aligns with the W3C PROV data model. JSON-LD examples and EU AI Act Article 12 mapping.
W3C PROV is a family of specifications for representing and exchanging provenance information on the web. Published as a W3C Recommendation, PROV defines a data model (PROV-DM), serializations (PROV-JSON, PROV-O in RDF), and constraints for describing the lineage of data: who created it, how it was derived, and where it came from.
The core concepts are Entity (a thing with provenance), Activity (a process that transforms entities), and Agent (an actor responsible for activities). Relationships like wasDerivedFrom, wasGeneratedBy, and wasAttributedTo link these together into a directed acyclic provenance graph.
Every provenance-relevant field in MemCube v4.1 has a direct counterpart in the W3C PROV data model. This alignment is intentional — it means Sgraal provenance data can be exported to any PROV-compatible system without lossy transformation.
| MemCube v4.1 Field | W3C PROV Concept | Description |
|---|---|---|
| provenance_chain | prov:wasDerivedFrom | Ordered list of agents that touched this memory, mapping to a chain of derivation relationships |
| source_trust | prov:qualifiedAssociation | Trust score as a qualified relationship attribute between agent and entity |
| memory_location | prov:atLocation | Where the memory resides, mapping to PROV's location qualifier |
| timestamp_age_days | prov:generatedAtTime | Age since creation, inverse of PROV's absolute generation timestamp |
A single MemCube v4.1 memory entry can be losslessly represented in W3C PROV JSON-LD. The following example shows a semantic memory that passed through three agents:
{
"@context": {
"prov": "http://www.w3.org/ns/prov#",
"sgraal": "https://sgraal.com/ns/memcube#"
},
"@id": "sgraal:mem_trade_001",
"@type": "prov:Entity",
"prov:generatedAtTime": "2026-04-11T14:30:00Z",
"prov:atLocation": { "@id": "sgraal:vector-store-primary" },
"prov:wasDerivedFrom": [
{ "@id": "sgraal:mem_trade_001_v0", "prov:wasAttributedTo": "sgraal:agent-planner-01" },
{ "@id": "sgraal:mem_trade_001_v1", "prov:wasAttributedTo": "sgraal:agent-summarizer-03" }
],
"prov:wasAttributedTo": {
"@id": "sgraal:agent-executor-07",
"prov:qualifiedAssociation": {
"sgraal:source_trust": 0.91
}
},
"sgraal:content": "Trade execution approved for portfolio rebalancing.",
"sgraal:type": "semantic",
"sgraal:source_conflict": 0.03,
"sgraal:downstream_count": 5
}
Aligning with a W3C Recommendation gives Sgraal provenance data immediate credibility in enterprise and regulatory contexts. Auditors and compliance teams already know PROV.
PROV-compatible systems — knowledge graphs, data lineage tools, audit platforms — can ingest Sgraal provenance data without custom adapters or transformation logic.
Article 12 requires high-risk AI systems to maintain traceability logs. W3C PROV alignment provides a standards-backed audit trail aligned with this requirement. Whether the alignment satisfies your specific Article 12 obligation is a determination for your DPO or counsel.
As provenance standards evolve (PROV 2.0, IEEE P7013), Sgraal's alignment enables forward compatibility with minimal schema changes.