Mesh3D

Indexed triangle mesh: three parallel buffers (positions, normals, indices) sharing one index space. Vertex3D { position, normal } is the interleaved GPU layout (24 bytes; layout-tested).

Flat per-face shading

Mesh3D::compute_vertex_normals() writes flat per-face normals — for each triangle, every vertex in the triangle gets the same face normal. Matches THREE.BufferGeometry::computeVertexNormals() for non-indexed geometry.

The trade-off vs. shared-vertex meshes: more vertex memory, but sharp dihedrals without a geometry shader. For low-poly geometry like the 404 pyramid (18 vertices) the cost is negligible.

The pyramid constructor

Mesh3D::pyramid(apex_y, base_half) produces the engmanager.xyz 404 layout:

flowchart TB
  apex["apex (0, apex_y, 0)"]
  base["square base at y = -1.05<br/>nw / ne / se / sw at ±base_half"]
  apex --> base
  apex -->|"4 side faces"| sides["12 vertices"]
  base -->|"2 base triangles"| bases["6 vertices"]

Vertex layout is 18 positions / 18 indices = 0..18, deliberately not shared so each face owns its normal. Tests assert 5 unique face normals (4 sides + 1 base — base triangles share a normal since they're coplanar).

```admonish note title="pyramid() is the reference constructor" cube, ico_sphere, etc. follow the same pattern: hand-laid positions per face, compute_vertex_normals() at the end, indices: (0..N).collect(). The constructor exists to make the 404 port mechanical; bring-your-own meshes are a normal use of Mesh3D { positions, normals, indices }.