loop_windows moves beside loop_window and takes the model; a test pins it on a real take with the two skeletons

The worker's window path loaded the 8B model at import, so no test reached it and swapped atan2 arguments or
the 30-joint skeleton went unnoticed. loop_windows(output, model) needs only model.output_skeleton and
model.fps. The test uses a committed 240-frame somaskel77 take (fixtures/take.npz, joints from a live Vili BVH
by forward kinematics) and stand-ins that carry the 77- and 30-joint skeletons' names and parents.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

scripts/loop_window.py

@@ -3,9 +3,13 @@
 A seam is the jump from frame b back to frame a. Its cost is the pose difference plus 0.1 s times the velocity
 difference (Kovar, Gleicher and Pighin 2002, motion-graph transitions), both as RMS over joints, with each frame
 taken relative to its own Hips and turned to face a fixed way, so where the take wanders and which way it faces
-do not count. Numpy only.
+do not count. loop_window is numpy only; loop_windows needs torch and kimodo's heading, and no model weights.
 """
+import math
+
 import numpy as np
+import torch
+from kimodo.constraints import compute_global_heading
 
 VELOCITY_WEIGHT_SECONDS = 0.1
 
@@ -39,3 +43,18 @@ def loop_window(joints, heading, fps, min_frames, hips=0):
     best = np.flatnonzero(cost == cost.min())
     pick = best[np.lexsort((firsts[best], -(lasts[best] - firsts[best])))[0]]
     return int(firsts[pick]), int(lasts[pick]), float(pose_cost[pick]), float(velocity_cost[pick])
+
+
+def loop_windows(output, model):
+    """One loop window per sample of a Kimodo output, as the worker answers them: {first, last, poseCost, velocityCost},
+    frame indices into the take, last inclusive. The joints are posed_joints, always in model.output_skeleton (the
+    somaskel77, not model.skeleton), and the heading is Kimodo's own from those joints."""
+    skeleton = model.output_skeleton
+    hips = skeleton.bone_index["Hips"]
+    windows = []
+    for joints in output["posed_joints"]:
+        heading = compute_global_heading(torch.from_numpy(joints[None]), skeleton)[0]
+        angle = torch.atan2(heading[:, 1], heading[:, 0]).numpy()
+        first, last, pose_cost, velocity_cost = loop_window(joints, angle, model.fps, math.ceil(len(joints) / 2), hips)
+        windows.append({"first": first, "last": last, "poseCost": pose_cost, "velocityCost": velocity_cost})
+    return windows