Remesh debug dumps

RemeshDump.Capture runs the remesher's first five stages and keeps what each one produced: the conditioned triangles as a mesh, the feature polylines as a line set, the cross field as a cross per vertex, the singularities as points, the patch layout as one region per triangle, and the quantization as one label per arc.

It returns the data. It never writes a file.

When a remesh looks wrong, which stage is the first question, and a monolith cannot answer it. RemeshReport says which stage was slow and which one dropped something. This says what each one produced, which is the half you need in order to look at it.

The four artefacts map onto the four things that go wrong:

⚠ It returns arrays and never touches the filesystem, because Core/ is under the virtual-path rule: no System.IO.Path, no File. Turn a RemeshSegment list into an .obj, a gizmo batch or an editor overlay in the layer that is allowed to — the format that suits an editor is not the one that suits a bug report.

⚠ A patch index rather than a colour. "The layout as coloured regions" is what the design asks for, and a palette baked in here would be one more thing to disagree about while throwing away the identity you need to point at a patch in the report. Colouring an index is one modulus away.

A 4-RoSy field has no arrow. Field emits two crossed segments per vertex — the representative direction and its quarter turn — because drawing only the representative invents a sign the solver never had, and the discontinuity that draws is a rendering artefact somebody will spend an afternoon chasing as a solver bug.

⚠ The arm length is a fraction of the bounding box's diagonal and never a length. CrossArm is 0.004. A fixed number would draw nothing on a millimetre-wide part and a hairball on a kilometre-wide one, which is the mistake this repository has made three times.

Capture does the work again rather than reaching into a remesh that already happened, and it costs what a remesh costs. The alternative is a capture hook threaded through Remesher, which puts a debugging concern in the middle of a pipeline every caller pays for.

⚠ What makes that legitimate is the determinism gate. The same input and settings give the same answer at any worker count, so the artefacts captured here are the artefacts the remesh had. If that ever stopped being true, this facility would be the second thing to break and the report would be the first.

It stops before extraction, because the extraction's artefact is the mesh Remesher.Remesh already returns.

Find the stage that refused.

See where the quantizer had to spend. The cost is the squared distance between what an arc got and what the density field asked for, summed over arcs — so the arcs where the two are far apart are the ones the consistency system paid for.

Draw the layout. One region per conditioned triangle, -1 where the partition claimed nothing.

• Retopology settings and reports — what the report says, stage by stage.
• Quad remeshing — the pipeline these artefacts come out of.
• docs/plan/41 — § D1's seven stages and § R4's audit.