# HANDOVER — V394 Hurdle Stretch source timing sync

## Base
V393 canonical/refined hurdle-stretch geometry.

## Human QA finding
Geometry was improved, but animation tempo did not synchronize to the packaged source film.

## Root cause
V393 still used a **linear 4.75 s phase map** with only start/end knots. The source clip is 7.423333 s and the complete two-repetition window contains non-uniform entry, long grip holds, release and travel/reset intervals. A linear map cannot reproduce those timings.

## V394 correction
- No IK change.
- No solver change.
- No bone/model geometry change from V393.
- Source timing measured at 12 fps (±0.083334 s).
- Two complete alternating repetitions use source window **1.333333–6.750000 s** = **5.416667 s**.
- Explicit phase knots preserve the observed entry, first hold, release, mid-walk, second entry, longer second hold, release and reset.
- Production remains resolved playback with **720 geometry samples and 0 live solver calls**.

## Source event anchors
- first entry: 1.416667 s
- first grip hold: 1.916667–2.916667 s
- mid neutral: 3.250000 s
- second entry: 3.750000 s
- second grip hold: 4.250000–6.083333 s
- loop return: 6.416667 s
- travel reset end: 6.750000 s

Geometry hash before/after timing patch: `28e31da35dd4f39648698078a4d804a0e9ac3e8135ba5c8b9e9c0b01c46ed60f` (identical).

## QA
Run `tools/nexprod/render_v394_hurdle_stretch_cpu_skinned_mesh_temporal_qa.mjs` for two-loop actual-Nex FBX temporal QA. Chromium/WebGL parity is not claimed unless separately run.