# CLAUDE.md

Tips for future agents working in this repo.

## Architecture

```
matlab/
  solve_template.m the whole solve as ONE script — mip load -> import ->
                   resample -> surfaceop -> write result.json — with
                   {{PLACEHOLDER}} tokens for the parameters. The engine's
                   buildSolveScript fills them per solve, and the UI offers
                   the filled script for download any time a mesh is loaded
                   (before solving too — a solve might fail or hang). The
                   same script runs in desktop MATLAB with the converted
                   .msh alongside: via mip, or by commenting out the mip
                   line and putting surfacefun on the path.
src/mesh/          Pyodide + meshio upload pipeline (bridge.py runs in Pyodide)
src/engine/        run-per-solve wrapper over numbl/browser's
                   createNumblSession: solve() fills the template and boots
                   a fresh session with the script + mesh staged, reads
                   result.json back via session.readFile, and disposes the
                   worker. numbl owns the worker, VFS, mip bootstrap, and
                   IndexedDB package persistence; prewarm() at page load
                   triggers the one-time package download.
src/pde/presets.ts PDE definitions, presets, slow-mesh warning threshold
src/render/        three.js SurfaceView (mesh preview / solution) + parula
scripts/engine-test.mjs  headless Node check of the whole MATLAB pipeline
```

## Key gotchas

- **numbl >= 0.4.12 from npm.** Needs `NumblSession.readFile` (0.4.10),
  enumeration-class support and the 1×1-tensor broadcast-assignment fix
  (0.4.11 — surfacefun's `surfacemesh.patchtype` / `dealm` idiom depend on
  both), and the 1×1-tensor gather-orientation fix (0.4.12 — surfacefun's
  `trianglepts` recursion breaks without it; see numbl's
  `test_scalar_tensor_vector_index_shape.m`). To develop against a local
  numbl checkout, point package.json at
  `file:../../numbl` and run `npm run build:lib && npm run build:browser`
  there after source changes; when switching back to a `^` range,
  `rm -rf node_modules package-lock.json && npm install` (else `npm ci` fails
  on the stale `file:` link).
- **surfacemesh.import needs MSH 4.1, one cell type.** surfacefun's gmsh
  reader (`+surfacemesh/+import/gmsh.m`) parses MSH 4.1 node-entity blocks,
  not 2.2, and accepts triangle (type 2) or quad (type 3) elements — but
  errors on meshes containing both. `src/mesh/bridge.py` and
  `scripts/make_samples.py` write the canonical 4.1 form (one surface entity
  block, sequential 1-based ids); the bridge splits quads into triangles
  when an upload mixes the two kinds. Uploaded .msh files in other layouts
  pass through meshio and get rewritten to 4.1.
- **Triangle patches are flat vectors, not grids.** An order-p tri patch
  holds n(n+1)/2 points (n = p+1) in surfacefun's `trianglepts(n)` ordering;
  quad patches are column-major n-by-n grids. `solve_template.m` reports
  `ptype` plus points-per-edge `n`, and SurfaceView triangulates tri patches
  with a JS port of surfacefun's `trilattice.m`.
- **Solve errors reject the solve() promise** with the MATLAB error message
  (a failed script run is a numbl bootError). Each solve is a fresh session,
  so nothing needs to stay alive across failures.
- **jsonencode collapses 1-element vectors to scalars.** Patch arrays are
  at least (p+1)(p+2)/2 >= 6 long so it never bites here, but remember it
  when adding payload fields.
- Package caching: numbl/browser persists /system (mip + installed
  packages) in IndexedDB, wiped after 30 min of inactivity (numbl's default;
  lowered from 24 h so a rebuilt surfacefun package refreshes without a manual
  clear) — the prewarm session at page load re-downloads ~28 MB after a wipe;
  solves after that only pay a per-run `mip load` (~1 s). Delete the
  `numbl-embed-system` IndexedDB database to force a cold boot.

## Testing

- `npm run engine-test` — full headless solve in Node against the local
  numbl build (dist-lib), including a quantitative eigenfunction check. It
  fills and runs matlab/solve_template.m standalone per solve (as the
  browser does; the fill logic mirrors engine.ts's buildSolveScript), sharing
  one VFS across solves as the stand-in for IndexedDB persistence, and
  passes the mip search path explicitly, exercising the same
  searchPaths-scan behavior the numbl/browser session relies on. Downloads
  are cached in `.cache/` keyed by URL; delete the cache to test fresh
  installs.
- `python3 <venv>/bin/python` with meshio 5.3.5 can exercise
  `src/mesh/bridge.py` outside Pyodide (redirect its `/work` constant).
- Browser verification (Pyodide upload path, session boot, IndexedDB
  persistence across reloads, 3D view) is manual: `npm run dev`.