Graph fact engine

The Graph Fact Engine verifies claims by checking them against detailed knowledge graph (KG) structures, rather than just unstructured text. This provides structured fact-checking for complex relationships.

How it works

It decomposes claims into Subject-Predicate-Object (SPO) triples and queries a connected knowledge graph (like Neo4j or NetworkX) to verify the relationship exists.

Triple extraction: Extract (Paris, is_capital_of, France) from “Paris is the capital of France”.
Pathfinding: Search the KG for a path between Paris and France with edge is_capital_of.
Transitive verification: Verify implicit relationships (e.g., if A is in B, and B is in C, is A in C?).

Usage

response = client.verify_graph(
    claim="Elon Musk is the CEO of Tesla",
    graph_id="tech_leaders_kg"
)

Result contract

The Graph Fact Engine returns a DiagnosticResult. A claim only resolves to VERIFIED when every material triple has full graph support — partial matches do not verify:

Path	`DiagnosticResult.status`	Notes
All triples fully supported	`VERIFIED`	Emits a `proof_ref` bound to the matching graph paths.
One or more triples unsupported	`UNVERIFIABLE`	NLI fallback results are recorded in `advisory_checks` only.

The NLI fallback is advisory only — it can never promote an unsupported claim to VERIFIED.

When to use

Complex relationships: Family trees, corporate hierarchies, supply chains.
Multi-hop reasoning: “Is the CEO of the acquisition target verified?”
Structured data: When your source of truth is a database or graph, not a document.

Overview

Core Engines

Analysis Engines

Specialized Engines

How it works

Usage

Result contract

When to use

​How it works

​Usage

​Result contract

​When to use

How it works

Usage

Result contract

When to use