Class Graph

Every node in the graph, in discovery order.

Every connection, grouped by producing node in order. Like order(), the result is independent of how the graph was discovered.

Nodes in dependency order — every node appears after the nodes feeding it. Evaluation doesn't need this (pulling is recursive and finds its own order); it's for display and for reasoning about a graph you didn't build.

The DFS is seeded in node-id order rather than discovery order, so the result depends only on which nodes are in the graph — not on which node Graph.from happened to start from. Several topological orders are usually valid; this picks the same one every time.

Structural dump: node ids, kinds, port names, and edges.

Nodes come out in order, not discovery order, so the dump of a given graph is the same whichever node Graph.from started at — which is what makes it safe to diff two dumps against each other.

This is a description, not a serialization — there is no fromJSON. Rebuilding a graph from JSON needs a registry mapping kind to a factory, plus per-node config in the dump; neither exists yet. Use this for inspection, diffing, and rendering.

Discovers every node reachable from roots, following connections in both directions.

const graph = Graph.from(sink);
graph.order().map((node) => node.kind);
Copy