Implicit H3 Compute Globe
The ~692 million H3 resolution-8 cells are an address space, not a table. This globe never builds them. A single triangle is rasterised; every pixel asks the shader for its cell.
The grid is a function. The terrain is a fractal. The hex is the political boundary.
Ray-hit the sphere, find the icosahedron face, hex-quantize at the current resolution, draw the cell with a signed-distance hexagon. Cost is pixels, not cells. deck.gl instances thousands; this evaluates hundreds of millions by never creating them.
Mapspinner-style FBM lives inside the hex: natural geography under a statistical boundary. Voxel drum extrudes observed HL-IRRI net risk as hexagonal prisms — a 3D bar chart whose 691 million columns exist only as an SDF.
Slide 0 → 8. The shader blends parent and child hex SDFs, so 122 cells fracture toward 691,776,122 without a tile request. That is H3's hierarchy as a visual function, not a layer swap.
Full latLngToCell in GLSL is the expensive port (icosahedron base cells, aperture-7, pentagon distortion). This globe uses an H3-topology DGGS in the shader and CPU picking. Exact Uber H3 bits stay in h3o on the platform. Observed HL-IRRI sites are overlaid by geodesic proximity, which does not need a 26k×26k lookup texture.
What this is, and what it is not
Analytic indexes are ihf:res:face:q:r, not H3 index strings. Pentagon count is twelve at every resolution, matching H3. Cell counts match the published table. Empty ocean cells still exist in the address space — they are not scored safe. The sixty glowing sites are the v0.1 pilot book; everything else is unmaterialised.
HL-IRRI methodology → · OCML / G4 fabric → · SEI coupling index →