Knowledge Graph

A packet can carry knowledge_nodes[] — entities and concepts extracted from its sources, with typed relationships between them. This is the raw material the Registry turns into first-class edges for graph-aware retrieval, consensus, and authority scoring.

KnowledgeNode

Field	Type	Notes
id	string	Required
entity_name	string	Required
entity_type	enum (extensible)	Required — see below
description	string
source_ids	string[]	Sources that evidence this entity
relationships	object[]	Typed edges — see below
domain_data	object	Domain-specific attributes

Entity types (extensible — x-/x_ for novel values):

compound · supplement · drug · condition · algorithm · parameter ·
concept · organism · gene · biomarker · food · other

Relationships

Each entry in relationships[]:

Field	Type	Notes
target_id	string	Required — the other entity
relationship_type	enum (extensible)	Required — see below
strength	number 0–1	Edge weight
evidence_grade	enum	Grade of the evidence for this edge

Relationship types (extensible — x-/x_ for novel values):

synergistic · antagonistic · prerequisite · contraindicated ·
metabolizes · inhibits · activates · similar_to

How the registry uses the graph

When a packet is published, the registry persists these typed/weighted relationships as cross-packet edges and resolves entities to canonical identities. That graph then powers:

• /entities/:name/relations — typed edges for an entity
• /consensus/:name — grade-weighted, empirically-derived contested signal
• /entities/:name/authority — global PageRank authority
• /answer — graph-aware citation chains (entity → relation → entity → sources)

See Query the registry and Trust & safety.

→ Next: JSON Schema & Types · Actions & Agents (Layer E)