RFD 1138: Where a range of motion comes from
Problem
A fit of 17 detected points to 104 unconstrained rotations folds bones through each other and reports 0.022% of stature. Nothing in the loop sees it, because the residual measures reprojection and so does the referee. A joint bound would, and a bound needs a range of motion.
The only such data here is addbio_*_rom in the kusudama project. CLAUDE.md blocklists AddBiomechanics .b3d as an IDENTITY source, for a narrow and inequitable population. Bounding a joint angle is a different use, and whether the objection follows the data into it has never been decided. This asks rather than answers: no prior decision was found, and inventing a citation would be worse than an open RFD.
Decision
Open. Three candidate sources and what each costs.
Constructed, from our own assets. Sweep ANNY’s pose library or a licence-clean motion set through the rig and measure the envelope. True by construction, reproducible, and narrower than a body: it measures what the rig was authored to do. AddBiomechanics ROM. Real bodies, already in the kusudama datasets. The blocklist row is about identity and this is a joint angle, a distinction to decide rather than assume.
Our own photographs. Circular: recovering 3D angles from single uncalibrated views needs the limits the range would define. The blinded holdout cannot source it and the cosplay library is validation only.