Composable scroll case components and versioned styles (educelab.v1, villa.2026-10)

[csparker247]

Summary

Case design features are now composable components, and a style is a named, immutable preset over them. Two styles ship:

• educelab.v1 (default, permanent): today's honeycomb case, unchanged.
• villa.2026-10: ScrollPrize/villa's scrollcase at 7d6a82c. No shell, curved split, square bolted caps, kinematic disc.

Component	Options
[split]	plane, curve (villa's S, spanning the whole case; amplitude, flip)
[shell]	honeycomb, solid, none
[ends]	shell (floor + lid), caps (villa's square bolted caps)
[mount]	generic-112.5, generic-65, kinematic, none

Components mix freely, e.g. villa's caps and kinematic disc around a honeycomb shell on a curved split. Combinations that can't be built are rejected when the config loads.

Commits

1. Restructure config into components and presets. Typed sections, presets resolved as defaults → preset → user values. Golden files in tests/presets/ freeze each released preset. The config schema changes (no compatibility shim); examples/generic.toml is updated.
2. Curved split. split.py defines every split as a profile y = f(x), shared by the mesh and B-rep stages. That keeps undercut removal along ±Y valid. The curve spans the case through the shell wall, so no flat seam lines up with the beam during CT. The divider follows it, the honeycomb leaves it solid, and nubs sit at (x, f(x), z). Nub positions are now checked against the fitted layout.
3. Villa components and villa.2026-10. Caps are ported from upstream's builder code. New rule: the stand requires shell ends.
4. Upstream comparison. scripts/villa_reference.py runs upstream's code in its pinned environment; test_villa_reference.py requires our bodies to match its volumes and bounding boxes.
5. Docs site. MkDocs Material built from docs/ and deployed to GitHub Pages from main by .github/workflows/docs.yml, with a strict build check on PRs. The villa comparison stays in the repo but is left off the site. The README is rewritten to match the repo.
6. Config template. examples/template.toml, shown on the "Creating a scroll case" page. test_examples_load checks that every example config loads.
7. Review fixes.
   • With no mount disc, the left label is engraved on the left half's own floor. Before this fix it was silently lost.
   • Marker rings are built on solid shells too. With shell.type = "none" they're ignored instead of rejected, so switching a preset to no shell doesn't also require turning them off. This is documented under [shell].
   • Caps around a shell wider than the mount leave a wall_thickness margin so the shell isn't tangent to the cap sides. Caps without a shell, including villa.2026-10, are unchanged.
   • The build fails if the bottom cap's mounting holes don't fit on the cap, e.g. a small scroll with caps and no mount.

Smaller commits: config errors are reported without tracebacks, and a CLAUDE.md is added.

Verification

• No geometry change for existing configs. educelab.v1 builds of a generic case, a fully featured case, a solid-shell/65 mm case and PHercParis3 have identical volume, bounding box and face count before and after each commit.
• Exact match with upstream. At four scroll sizes, the symmetric difference between our villa.2026-10 case bodies and upstream's is 0.000 mm³.
• End to end on PHercParis3/4 (docs/villa-2026-10-comparison.md):
  • Ours are closed single bodies; upstream's PHercParis3 halves have 26 holes and dozens of fragments.
  • Clearance is 1.95 mm for ours and 1.8 mm for upstream's, against a 2 mm target.
  • No blocked slide-out paths in either.
  • Our scrolls sit 180° from upstream's, because the wide axis's sign is arbitrary in both.
• 68 tests pass (~80 s).

Test plan

• uv run pytest
• Baseline rebuild comparison for educelab.v1
• Upstream reference regenerates byte-for-byte
• mkdocs build --strict (CI)
• Set Pages source to GitHub Actions after merge
• Print and scan a curved-split honeycomb case before making the curve the educelab default

🤖 Generated with Claude Code