# Frontend (React + Vite) – WebGL Graph Renderer

The frontend renders the snapshot from `/api/graph` using WebGL2:

- Nodes are drawn as points
- Edges are drawn as lines only when sufficiently zoomed in
- Selection + neighbor highlighting is driven by backend “selection queries”

## Run

Via Docker Compose (recommended):

```bash
docker compose up --build frontend
```

Open: `http://localhost:5173`

## Configuration

- `VITE_BACKEND_URL` controls where `/api/*` is proxied (see `frontend/vite.config.ts`).
- The right-side cosmos graph reads these `VITE_...` settings at dev-server startup:
  - `VITE_COSMOS_ENABLE_SIMULATION`
  - `VITE_COSMOS_DEBUG_LAYOUT`
  - `VITE_COSMOS_SIMULATION_REPULSION`
  - `VITE_COSMOS_SIMULATION_LINK_SPRING`
  - `VITE_COSMOS_SIMULATION_LINK_DISTANCE`
  - `VITE_COSMOS_SIMULATION_GRAVITY`
  - `VITE_COSMOS_SIMULATION_CENTER`
  - `VITE_COSMOS_SIMULATION_DECAY`
  - `VITE_COSMOS_SIMULATION_FRICTION`
  - `VITE_COSMOS_SPACE_SIZE`
  - `VITE_COSMOS_CURVED_LINKS`
  - `VITE_COSMOS_FIT_VIEW_PADDING`
- The right pane keeps a static camera after an explicit `fitViewByPointPositions(...)` from the current seed positions.
- `VITE_COSMOS_SIMULATION_CENTER` is the main knob for keeping the graph mass near the viewport center during force layout.
- `VITE_COSMOS_DEBUG_LAYOUT=true` enables a small debug overlay and `console.debug` logs for graph-space centroid/bounds, screen-space origin/centroid placement, zoom, alpha/progress, and space-boundary pressure.
- In Docker Compose, set them in the repo-root `.env` and restart the `frontend` service.

## UI

- Drag: pan
- Scroll: zoom
- Click: select/deselect nodes

Buttons:

- **Top-right:** selection query mode (controls how the backend expands “neighbors” for the current selection)
- **Bottom-right:** graph query mode (controls which SPARQL edge set the backend uses to build the graph snapshot; switching reloads the graph)

The available modes are discovered from the backend at runtime (`/api/selection_queries` and `/api/graph_queries`).

## Rendering / limits

The renderer uses a quadtree spatial index and draws only a subset when zoomed out:

- Points:
  - Per-frame cap: `MAX_DRAW = 2_000_000` (sampling over visible leaves)
- Lines:
  - Drawn only when fewer than ~20k nodes are “visible” (leaf AABB overlap with the camera frustum)

Selected and “neighbor” nodes are drawn on top using index buffers.