"""
Graphgen — what a graph is, and where graphs come from.

Two halves, and they are independent:

    graph.py     the model. Nodes, edges, groups, and the meaning carried on
                 them. No colour, no font, no layout engine.

    schema.py    the first source. A `schema.json` or a modelgen `schema/`
                 folder becomes `{models, relationships, source}`, which
                 `api.py` serves at /station/tools/graphgen/api/schema and the
                 browser viewer draws.

`{models, relationships, source}` is a **published contract**, not an internal
shape: modelgen emits it (`generator/jsonschema.py`), datagen exposes it
(`base.py::schema`), shuntgen generates it, and `cfg/amar` reads it off disk.
Two modelgen tests assert it. It does not change to suit anything here.

The senses of "graph" parked for later — video pipeline processing graphs, local
computer-vision graphs, Supabase-style schema diagrams — are more *sources*.
They belong beside `schema.py`, feeding the one model, rather than each growing
its own drawing code.

## Drawing one is not this tool's job

`docgen` takes a graph and produces DOT, SVG, and documents, with the styling
supplied as a data profile. The dependency runs one way: docgen reads this
model structurally and imports nothing from here, so either folder works with
the other absent. `docgen/shape.py` is that contract, written down.

    from graphgen import Graph
    from docgen import Profile, emit, render     # the other half, if present

## Importing this is cheap

`api.py` is deliberately not imported here. Pulling in the model must not pull
in FastAPI — `run.py` imports `station.tools.graphgen.api` explicitly when it
mounts the router, and nothing else should have to pay for that.
"""

from .graph import CLASSES, Edge, Graph, Group, Node

__all__ = ["Graph", "Node", "Edge", "Group", "CLASSES"]