Module graph

Structs§

EdgeId: A graph edge’s identity. Content-addressed over its endpoints and type, so the same relationship is one edge however many times it is observed.
GraphEdge: One edge: its id, endpoints, type, weight, attributes, and an optional valid-time window for bitemporal facts.
GraphNeighbors: A one-hop neighbor read: the cheap, common traversal. depth follows the same hop repeatedly.
GraphNode: One node: its id, labels, attributes, optional embedding, and optional source.
GraphQuery: A graph traversal: start, hop spec, optional node and edge filters, and what to return. Reuses the query Filter predicate language, so there is one filter grammar across query and graph.
GraphResult: The data a graph traversal returns. Which fields are populated depends on the query’s GraphReturn.
GraphUpsert: Upsert nodes and edges into a graph. The projector path: idempotent on content-addressed ids, so re-applying the same extraction is a no-op.
Hop: One traversal step: follow edges of an optional type in dir, up to max hops at this step.
NodeId: A graph node’s identity. Content-addressed (the hash of the node’s label and canonical value), so the same entity extracted from different messages converges on one node. Minted SDK- or projector-side.
Path: One path through the graph: parallel node and edge id sequences.

EdgeDir: Which way a hop follows edges from the current frontier.
GraphError: Why a graph operation failed.
GraphReply: The result of a graph operation: Ok with the traversal data, or Err with a structured failure.
GraphReturn: What a graph query returns.
GraphStart: Where a traversal starts: explicit node ids, the nodes matching a predicate, or the nodes nearest an embedding (vector-seeded traversal).
SourceRef: Where a graph element was last observed: the source record an extraction came from, so a reader can navigate back to its origin. On an edge this is the record that asserted the relationship (the meaningful provenance, kept last-writer since the edge is rewritten on each observation to maintain its validity window). On a node it is the first record the entity was seen in (first-writer, so a re-observed node’s stored bytes stay stable). Excluded from the content-addressed id, so it never affects identity or idempotent upsert. The complete history is the source log, which the projector can replay. Absent on the wire when unknown.

validate_graph_name: The canonical graph-name rule, shared by the SDK client edge and the serving plane: non-empty, at most MAX_GRAPH_NAME_BYTES bytes, no ASCII control characters.