Glossary

Under the closed-world assumption, a statement absent from a defined complete dataset is treated as false within that dataset's scope.

Data fusion combines observations or datasets to produce a consolidated representation for a defined task.

Information extraction turns statements in source material into structured candidate entities, attributes or relations.

Knowledge fusion combines claims from multiple sources into a coherent account while preserving uncertainty and origin.

A knowledge graph represents entities and their relationships as connected, typed statements.

Knowledge-base population adds candidate entities, attributes and relations to a structured body of knowledge.

An ontology specifies concepts, relationships and constraints used to interpret statements in a domain.

Under the open-world assumption, a missing statement does not imply that the statement is false.

Schema alignment maps fields and relationships across sources while accounting for differences in meaning and scope.

Slot filling finds a value for a specified field of a specified entity from available evidence.

Attribution identifies the source or actor responsible for a statement, dataset or derived work.

Chain of custody documents the possession, transfer and handling of evidence from acquisition through use.

Content addressing identifies a digital object by a digest of its bytes rather than solely by its storage location.

Data lineage traces data from source inputs through transformations to downstream outputs.

An evidence span is the specific passage or region of a source used to support a claim.

Link rot is the loss of access to cited web content when a URL stops resolving or no longer serves the referenced material.

Provenance records the origins, transformations and responsible actors associated with information.

A retained source copy is a preserved version of material used as evidence for a claim or decision.

A sentence-level citation ties a particular sentence or claim to the exact source material offered as support.

Chunking divides source material into units that can be indexed, retrieved, and cited.

Context engineering is the design of what information and instructions a model receives for a particular task.

A context window is the bounded input span a model can process in a single inference interaction.

GraphRAG uses graph structure to organize or retrieve context for a generated answer.

Grounding ties an output claim to information that can be inspected outside the claim itself.

A hallucination is generated content that is unsupported by its relevant source or conflicts with established facts.

Reranking reorders retrieved candidates using an additional relevance assessment.

Retrieval-augmented generation supplies retrieved material to a generative model for a particular response.

Vector search retrieves items by comparing numerical representations of their content or meaning.

An audit trail is a reviewable record of actions and changes affecting a defined process or information asset.

Data residency specifies where particular data is stored or processed under a defined policy or obligation.

Need-to-know access limits information to people or processes whose assigned task requires it.

A retention schedule states how long defined records are kept and what happens when that period ends.

A single source of truth is a governed arrangement in which consumers use one agreed account for a defined fact.

A system of record is the designated authoritative record for a defined class of information and decisions.