Entity Provenance

A chronological, auditable log of every computational operation applied to a data point from ingestion to its current state. This includes ETL steps, entity reconciliation passes, and normalization functions.

• Captures data provenance as defined by the W3C PROV standard
• Records each agent, activity, and timestamp in the pipeline
• Enables rollback and debugging of corrupted assertions

Example: A name string undergoes Unicode normalization, then fuzzy matching against Wikidata, with each step logged as a prov:Activity.

A numerical weight assigned to each property assertion or edge representing the system's certainty in its accuracy. Scores are derived from extraction method reliability, source authority, and cross-validation.

• Probabilistic assertions in systems like Google's Knowledge Vault
• Low-confidence edges flagged for human review
• Enables fact verification pipelines to prioritize claims

Example: A birth date extracted via a 99% confidence regex parser receives a higher score than one inferred from a free-text biography.

Metadata that scopes a fact's truth to a specific time interval or validity window. This distinguishes current assertions from historical records and prevents temporal contradictions in the graph.

• Uses prov:generatedAtTime and prov:invalidatedAtTime properties
• Essential for tracking entity evolution over time
• Prevents AI models from citing outdated facts as current

Example: A person's schema:affiliation to a company includes start and end dates, preventing the graph from asserting a former CEO as the current one.

Cryptographic or hash-based mechanisms that ensure the integrity of the provenance metadata itself. This prevents tampering with source records and establishes a verifiable chain of custody.

• Merkle tree structures for efficient verification
• Digital signatures on provenance assertions
• Enables non-repudiation of data origin

Example: Each transformation step produces a SHA-256 hash, creating an immutable chain that auditors can traverse to validate no unauthorized modification occurred.

A permanent audit entry documenting when two distinct URIs are linked via an owl:sameAs assertion, including the matching algorithm, confidence threshold, and human validation status.

• Prevents identity fragmentation across datasets
• Records the entity reconciliation decision process
• Supports rollback if a false match is later discovered

Example: A local customer record is matched to a Wikidata Q-Node with a 0.97 confidence score, and the reconciliation API response is stored as provenance evidence.