# Sweep31: support-axis and whole-body motion audit

The owner questions whether Sweep29 primarily turns the whole trunk around the
non-holding-side foot, with only weak internal torso twist. Reassess Sweep30's
motion architecture and evidence from first principles. Analysis only: preserve
all native/source files and do not silently implement or adopt a new candidate.

Inspect every original front frame, source foot/trunk/head crops, and the exact
current A/B media. Distinguish image-left non-holding foot from viewer-right
holding-side foot. Track observable contours and facial/tool features with
occlusion flags; these are not joint positions or force measurements.

Separate five candidate causes: common body orientation, contact-relative root
translation/orbit, torso-internal differential rotation, neck counterrotation,
and local arm movement. Audit actual code/control receipts for which causes
Sweep30 can represent and which its A/B comparison keeps fixed.

Assess the owner's off-center support-axis reading against symmetric contact,
central rotation plus translation, and torso-twist readings. State ambiguity
from a single uncalibrated view. Recommend a minimal falsifiable next experiment
instead of adding deformation merely because it is technically available.

Deliver a bound evidence report, feature overlays/full-frame sheets, mechanism
comparison, and a portable public analysis page alongside unchanged references.