The knowledge graph is a sidecar derived from reviewed pages and source records. Canonical human-readable pages remain primary; graph traversal, JSON-LD, RDF, property graphs, search indexes, and GraphRAG are secondary views over reviewed evidence.
Graph components
| Node or edge | Use | Rule |
|---|---|---|
| Stable IDs | Identify pages, sections, entities, claims, sources, contradictions, and review events. | Do not use titles as permanent IDs. |
| Claim nodes | Store factual assertions separately from prose when precision matters. | Attach source spans, status, sensitivity, reviewer, and freshness rule. |
| Typed edges | Represent depends_on, part_of, supersedes, contradicts, source_for, and related_to. | Use controlled relation names. |
| Contradiction edges | Preserve source conflicts and current resolution state. | Contradicted claims need a contradiction relation and review state. |
| Exports | Publish JSON-LD, RDF, property-graph, or retrieval-cache views. | Exports are read-only derived artifacts from reviewed content. |
Validation
- Every exported node links back to a canonical page or source record.
- Every claim has source status and review state.
- Contradicted pages include contradiction relations.
- Private, blocked, and draft nodes are filtered or labeled before retrieval.
- Graph export does not imply hosted graph database, public API, public SPARQL endpoint, certification, or automatic sync.