docs/research/QVM_POSE_LEAN.md

Pose pitch publication and lateral scratch arithmetic

RE-663. Reproduce with python3 scripts/analyze_qvm_pose_lean.py; check with python3 scripts/test_qvm_pose_lean.py. The machine report retains source identity and bounded fixtures.

Actual32797–33167 uses half-pitch only when flag96 is active, subtracts360 once above180, adds15 for IDs20/21 or5 otherwise, and publishes through argument40. Velocity is freshly copied and normalized with floating1/length, unlike the earlier integer reciprocal scratch. Nonzero velocity scales length by0.1, builds a basis from local60/64/68 and computes a lateral dot-product scalar at local168. The remaining32254 instructions contain no LOAD of that scalar, so it is documented as an unconsumed calculation rather than a proven visible lean effect. Zero velocity skips basis/native trig calls. Finite sqrt/trig models and original legal-input/library limitations remain explicit.

No original executable/full QVM/native collision/physics/rendering executes. Native responses and legal input domains are supplied or unverified where stated; full runtime/macOS/Linux Rust/Bevy/Solari gameplay remains open.