OK. some more info on my workflow. I've written some automated tests for brush meshes and I'm satisfied they are constructed closely enough for now - this probably resolved a bunch of issues. Next I'm generating screenshots from both open brush and the three.js code using exactly the same strokes and camera positions and i'm working through the discrepancies in batches. Starting with unlit brushes (as that excludes differences in lighting and a lot of pbr shading differences)