Eleven merges, and the hero never moved

The mobs stop walking like the player

Four bodies with distinct silhouettes, corrupt mass and asymmetric heads all played Walk_Loop — the hero’s own clip. A gait is now a row in the profile table, three mechanisms deep.

The brief was half wrong and the gate said so. A layer of pure constant offsets is decoration: the mob stands differently and moves identically.

Offsets alone score 0.0215° dynamic against a measured instrument floor of 0.0269° — literally zero. Arm-into-body contact 37 → 0 on the Breaker.

hero’s walk

own gait

The neck anchor was innocent

I ruled that build_neck anchored to nothing and ordered the fix, bounded at 6 mm with a per-region falsifier. The falsifier earned its keep.

Over budget and ineffective. The mobs’ neck bases were already on the torso axis — there was nothing to re-anchor.

The anchored neck moves the hero +7.180 mm — over its own 6 mm budget — and closes 0.038 mm of BasicZ1’s 83.21 mm escape.

Every hair shell was fitted to the hero’s skull

The mobs were bald against a 352 KB hair system. Hair is a row now, costing 2.8–4.9% of the tri budget and displacing zero growth triangles — the corruption cap was never binding.

Underneath was one default: _head_bvhs lofted from FRAME_PRESETS unconditionally, so every clamp, cull, lift and sink query answered about the wrong head.

Median error −35.1 mm on the Crier — hair floating off the head — to +52.2 mm on the Fused: skull straight through the hair.

The suit bay was touching the van

Both complaints were real and both of my diagnoses were wrong. The 3.76 m I kept quoting is an interact anchor, and RIG-01’s anchor is its side door.

And SUBJECT_FRAME_FRAC cancels for a standing subject — standing_span multiplies by it and frame_width divides. Tuning it would have moved nothing while looking like a fix.

Measured against bodywork the rack sat 0.03 m off the van. Framing 59–62% → 65–71%, with the printer, vat and gunsmith bit-identical.

0.03 m

1.34 m

Four clips refuse the phase lag, on their own evidence

All 53 clips ship the gait on all four mobs. Roll, Jump, Jump_Land and Vault take the offsets and swing scale without the lag — measured at bake time, not branched.

The lag breaks on one-shots, not on violent clips, which is not what I predicted when I flagged Roll as the risk.

Loop seams run 0.00–0.649° on 208 of 212 clip-instances, then 29.3 / 50.6 / 53.4 / 85.5. A lag on Vault produced 99.2° of arm travel in one frame.

Three mobs had a ribbon where a ribcage should be

_no_undercut measures overhang in world Y. On a leaning frame the spine carries the chest front forward faster than any depth can, so the clamp is unsatisfiable and returns its own floor — reporting success.

The priced fix would have taken the hero’s chest down 17.64 mm. Scoping it structurally — the datum is the running maximum of the spine curve — made it a no-op on the hero by construction.

Minimum chest-ring depth 0.100 mm on three of four mobs. Hero displacement after the fix: 0.0000 mm in every region of both frames.

0.20 mm

151.62 mm

The motion gate graded a clip that does not ship

The gait was baked into all four GLBs and proved from the bytes. t9_motion went on grading the unlayered source, so six of thirteen failures were measured on geometry the game does not contain.

It reported the Breaker’s pelvis × elbow_L at 66.1 mm where the gait pass measured 17.1. Body-only contact 86 pairs → 13.

A ladder brackets a crossing; it does not measure one

S_ARM_OUT_DEG exists so the constants cannot quietly stop being what their comments say. It caught exactly that — the arm clearance had been derived against mobs whose chests were 0.1 mm deep.

I read the coarse ladder’s next rung as the answer and reported a 25% widening. Swept properly in degrees, the real move is 0.45%.

Zero arm pairs first at 22.1°, not the 27.5° the six-point ladder implied. The residue at 22.0 was one pair at 0.239 mm.

The material rename was measured, then declined

Mob hair wearing M_Hero_Hair would put every zombie in the player’s runtime tint table. The concern was well-formed — the table is name-keyed on resource_name.

Nothing leaks. The tint walks the player’s own subtree, no mob source references the appearance layer, the write is on a duplicate under two regression tests, and the two GLBs are separate imports so equal names are unequal resources.

A rename would move MobRigCache’s merged-surface count, which tests assert. It would cost something and buy nothing.

The gash was two bugs

Facial hair rode SKULL/FACE from FRAME_MASC unscaled — the same defect as the hair one, in the half nobody had wired. Fixed first, wired second.

Then patch dropped whole columns. Making the unit a cell exposed the second half: the hole radius was derived from a width and applied in two dimensions.

21.7 mm against a 20 mm row pitch takes a column top to bottom — the same gash by another route, with every number green.

The extents gate could not see head, hair or beard

It was written when a mob was a body, and three subsystems landed on top of it. For hair the envelope is not inside the head — hair extends past the skull. It is a bounded allowance around the surface it grows from, anchored on the same float that brakes _lift_clear.

It went red on its first run, and reported rather than fixed: Mob_Fused_Head reaches 2.5 mm above the bind-pose crown. Not the hair — the skull.

Positive control caught at 5.2×: with the brake removed the crown reads 203.8 mm against a 39.2 mm allowance.