The body hides skin under clothing by splitting primitives, so the visible/hidden border follows triangle edges

[gidich]

The body hides the skin under clothing by splitting the mesh into eight primitives and making seven
of them fully transparent. The cut follows triangle edges, so the boundary between visible skin and
hidden skin is a zigzag rather than a line.

Measured 2026-08-27 on mpfb-peds-parent-aisha.motion-bind.glb

mpfb_ob_patient_aisha_body carries eight primitives:

material	alpha	cutoff	tris	baseColorFactor	texture
mpfb_skin_ob_patient_aisha	OPAQUE	0.5	20,242	[1,1,1,1]	819,496 B PNG
openclinxr_hidden_upper_…	MASK	0.5	2,664	[0,0,0,0]	none
openclinxr_hidden_foot_…	MASK	0.5	2,760	[0,0,0,0]	none
openclinxr_hidden_lower_…	MASK	0.5	506	[0,0,0,0]	none
openclinxr_hidden_upper_….001	MASK	0.5	218	[0,0,0,0]	none
openclinxr_hidden_foot_….001	MASK	0.5	140	[0,0,0,0]	none
openclinxr_hidden_orphan_…	MASK	0.5	138	[0,0,0,0]	none
openclinxr_hidden_lower_….001	MASK	0.5	88	[0,0,0,0]	none

20,242 + 6,514 = 26,756, the body's total.

Every hidden primitive has alpha 0 with a MASK cutoff of 0.5, so all of its fragments are discarded.
The regions vanish cleanly. What does not vanish cleanly is the border: a primitive split is a
partition of triangles, so the edge between the visible skin and the discarded skin runs along
triangle edges and cannot follow a smooth line.

The observation this explains, stated as an inference

I have graded "the neck ends in a jagged sawtooth collar at the garment" across several head-framed
captures of this actor. A triangle-granularity visibility boundary at the neckline is a mechanism
that produces exactly that silhouette.

I have not proven those are the same boundary. I measured the mechanism, not the correspondence.
Verifying it needs the visible/hidden triangle border projected into the graded frame, which I did
not do.

A second, separate finding in the same table

The hair material is baseColorFactor = [0.16, 0.11, 0.09, 1], OPAQUE, no texture. Flat dark
brown across 4,975 triangles. That is consistent with the featureless cap I have graded, and with
part of the tone break at the hairline where an untextured flat colour meets a skin primitive
carrying an 819 KB PNG. Also unproven as correspondence.

The fix shape, and what is not decided

A texture-space alpha mask cuts at texel granularity and can follow a smooth neckline; a primitive
split cannot. Whether that is worth the change here is not determined — it costs a mask texture per
body and the current approach is free at runtime.

Also open: whether the eight-way split is why the body is eight primitives at all, or whether
something else needs them separate.

Not measured

Whether the other ten shipped humanoids use the same split; only the parent was read. Whether the
.001 duplicates and the hidden_orphan primitive are intentional or residue — seven hidden
primitives for three named regions is more than the naming suggests.

factory_step: clothing_consume

[gidich]

Correspondence evidence. This card measured a triangle-granularity visibility border and said plainly
that it had not shown the border was the jagged edge I graded. Three assets now show that edge at
exactly the places the border runs.

MY GRADES, front lit, native resolution, on merged main

Captured with pnpm asset:model-vetting:glb-grade. All three are rest poses.

asset	what the garment/skin boundaries look like
mpfb-clinical-physician-adult	dark speckled fragments around the throat where the lab coat collar meets the neck; ragged edge where the teal scrub shows through; scrub V-neck torn at the shoulders
mpfb-peds-patient-child	dark notches at the shoulders where the collar meets the neck, and at both sleeve cuffs; a hard horizontal seam at the waist where the top's hem meets the trousers
mpfb-clinical-nurse-adult	a jagged teal band across the throat above the V-neck with a dark notch in it; light jagged slivers at both sleeve cuffs; the same hard horizontal band at the waist

Three of three. The defect appears at the neck, the cuffs and the waist — which is the complete
set of places a garment ends and skin begins on these assets.

Why that is the border this card describes

The hidden primitives are named openclinxr_hidden_upper, openclinxr_hidden_lower and
openclinxr_hidden_foot. Upper covers the torso under the shirt, lower the legs under the trousers.
Their boundaries with the visible skin primitive fall at the neckline, the sleeve ends and the
waist — the three locations above.

Still not a projection. I have not put the visible/hidden triangle border into image space and
matched it pixel for pixel, which is what would close this. What I have is three assets showing a
jagged boundary at three predicted locations, which is stronger than the single observation this card
was filed on and is not the same as proof.

What it does not say

Nothing here measures the other eight humanoids, and nothing measures whether a texture-space alpha
mask would look better — only that the current boundary is visibly jagged where the mechanism says
it must be.

Not measured

Whether the waist band is the same mechanism as the neck and cuffs. It could be a garment hem
against a garment waistband rather than skin against a hidden region, and I have not read which
meshes meet there.

[gidich]

CORRECTION — my correspondence evidence is withdrawn, and this card's premise with it

I posted that three of three graded assets show a jagged boundary at exactly the places the
hidden-primitive border runs, and called that correspondence evidence. That inference is wrong, and
the repo already measured why.

tools/openclinxr/evidence/the-hide-mask-boundary-is-not-a-sawtooth.test.ts carries a withdrawn
headline dated 2026-08-13 that says it plainly:

Every openclinxr_hidden_* material ships alphaMode=MASK, alphaCutoff=0.5,
baseColorFactor=[0,0,0,0], no texture. glTF discards any fragment whose alpha is below the cutoff…
The mask is discarded, not drawn. Corroborating from pixels: if it drew it would be a black band
around the torso, and no grade capture shows one.

I read those exact material properties myself with a NodeIO dump and filed this card on them. I did
not ask whether an alpha-zero MASK primitive draws anything. It does not. A border that is never
rasterised cannot be the jagged edge I graded on three actors.

That file also names what the visible zigzag actually is, measured: the cargo-pants waistband top
ring, HF residual p95 18.96 / 10.79 / 12.28 mm against the same body's shirt hem at 2.01 / 0.47 /
1.47 mm, tracked by the-waistband-is-as-smooth-as-the-hem.test.ts (#373).

What survives of this card

The measurement, which was correct and is not in dispute: the body is eight primitives, 20,242 skin
triangles plus 6,514 across seven fully-discarded MASK primitives with no texture. The live question
is the one that file already identifies and this card did not — dead geometry: seven discarded
primitives per actor, roughly 4,570 vertices on aisha, costing draw calls and vertex shading every
frame for nothing.

That is a cost question, not an appearance one. This card should be re-aimed at it or closed in
favour of whatever already tracks it. I am not re-aiming it in the same breath as withdrawing the
premise — measure first.

A live RED on main, and it is not from #739

Running that contract on main at 50a1e993: 2 of 3 clauses fail, both on one actor.

(2) mpfb-peds-nurse-kevin: mask ring 40 verts < floor 42
(3) mpfb-peds-nurse-kevin: mask Y extent 538mm < 98% of 571mm

Attribution, measured rather than assumed — kevin's hidden_upper primitive at three commits:

tree	verts	Y extent
f82e1cc2 (before #739)	1719	538 mm
de57fc4d (#739)	1719	538 mm
main (after #740)	1719	538 mm

Identical at all three. Neither #739 nor #740 caused it — the contract's 571 mm reference predates
both. It is an older regression or a stale floor, and it is red on main now.

Not measured: which commit did move it, and whether 571 mm is still the right reference.