feat(microscope-slide-hyperobject): explicit scale_comp parameters behind the legacy material injection

[aldoruizluna]

What changed

Digital Twins MES, Phase 2 wave 2, lane W2-MAT. Owner ruling: "flag now, flip later". This PR is the commons side that lets yantra4d turn RENDER_MATERIAL_INJECTION off without losing function.

• slide.scad: new explicit inputs scale_comp_x, scale_comp_y and scale_comp_z (unitless, default 1.0) drive the scale() around the slide body. The legacy mat_shrinkage_* and mat_clear_slide are now undef unless the platform injects them with -D. An injected value wins (_legacy_or()), so every render made with the flag on is unchanged. The legacy branch is commented REMOVE AT THE FLIP.
• The library function slide_slot_width() gets the same precedence for fit_clearance_slide (mm, default 0) over mat_clear_slide. This cartridge renders no slot, so its manifest does not declare fit_clearance_slide. microscope-slide-holder declares it in its own PR.
• mat_clear_press and mat_clear_loose were declared here but never read, so they are no longer declared.
• slide.py (CadQuery twin) reads scale_comp_*, so both kernels agree for any explicit value. It never read mat_* and still does not, so its output with injection on is unchanged.
• project.json: declares scale_comp_x, scale_comp_y and scale_comp_z (slider 0.9 to 1.1, step 0.001, labels in en/es/fr/pt, advanced visibility). README.md documents them.

GOC-1 mapping

• The new ids pass the GOC-1 variables denylist; I checked them against generator-output.schema.json (exact list plus the prefixes mat_shrinkage_, mat_clear_ and thermo_). After the flip, shrinkage reaches this cartridge only as a geometry-named number in variables, and nothing is recorded in legacy_physical_inputs.
• The manifest defaults equal the source literals: scale_comp_* = 1.0. With RENDER_INJECT_FULL_PARAMS on, the injected defaults leave the geometry unchanged (row A4 below).

Evidence

I rendered locally with OpenSCAD 2026.02.13 and --backend=Manifold. Geometry was compared as volume, AABB, body count and the sha256 of the sorted unique vertex set rounded to 1e-4, never as STL bytes. The rows cover slide_standard 0 to 4 (all 5 values; the cartridge has no presets):

Row	Render	Expected	Result
A0	after, no material	= before, no material	5/5 identical
A4	after, explicit defaults (scale_comp_*=1.0)	= before, no material	5/5 identical
A1	after, bambu-tpu-95a injected (the full mat_*/thermo_* set, as yantra4d sends it)	= before, TPU injected	5/5 identical
A5	after, sinterit-pa12-smooth injected	= before, PA12 injected	5/5 identical
A2	after, explicit scale_comp=1.012/1.012/1.018 with no injection	= before, TPU injected	5/5 identical (the explicit parameter carries the function)
A3	after, TPU injected plus explicit scale_comp=0.95	= before, TPU injected	5/5 identical (the injected value wins)

CadQuery twin, before vs after: identical at all 5 standards, with and without injected mat_*. The explicit scale_comp in CQ matches the SCAD TPU render (for example, standard 0 gives a volume of 2060.1353 mm³ and a bbox of ±38.456 × ±13.156 × ±0.509 mm in both).

y4d-spec check ./microscope-slide-hyperobject --render --require-openscad --parity --openscad-path libs --openscad-path .:

• at the CI pin 3ff3736: cartridges=1 failures=0 notes=0 geometry=verified renders=2 presets=0 skipped=0 parity=1/1 ok
• at the P2-S head (510efba, with the default-drift rule): the same, notes=0

CI status

Green on Linux CI (own runners) at the time of writing: manifest conformance (all cartridges) passed in 45s, render scope passed, and render (cartridges changed in this PR: microscope-slide-hyperobject) passed in 1m22s.

Deviations / notes

• target_material is untouched; this cartridge never declared it. Removing it is the flip, which is the owner's decision.
• Finding, out of scope here: the vendored copy of this library in microscope-slide-holder/libs/ already carries a single-polyhedron rib fix (a coincident-face union under Manifold) that this upstream slide_retention_rib() lacks.

🤖 Generated with Claude Code