The run cycles anchor on their feet, and stride at a real cadence

Oliver: 'the motion seems very flashy and unnatural'. Two causes, both
measured: drawChar centres a frame by its total width while the racket's
reach varies across the cycle, so the BODY teleported sideways up to 28
source px on a frame swap (the female player was worst) — every frame is now
padded until its lower-third centroid sits at the frame centre (residuals
±0.2 px, the normalisation lives in _runroll.py's cut). And the hardcoded
4.5 cycles/s was tuned on 2-4 near-duplicate frames; on eight real phases it
was a strobe — rolled sets stride at TUNE.runHz (2.4), old art keeps 4.5.

In-pane: body wobble down from ~36 device px per swap to 3.2.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

assets/_runroll.py

@@ -176,6 +176,24 @@ def cut(k, install=False, n=0):
     os.makedirs(outdir, exist_ok=True)
     outs = [os.path.join(outdir, f'{pre}{i}{suf}.png') for i in range(1, got + 1)]
     _fine.main(sheet, outs, TARGET_H, [], -1)
+    # ANCHOR EVERY FRAME ON ITS FEET. drawChar centres a frame by its total width, and the racket's
+    # reach varies wildly across a cycle (the cut female frames measured -28..+15 px of feet drift
+    # from frame centre) — so the BODY teleported sideways every frame, which is exactly Oliver's
+    # "very flashy and unnatural". Pad each frame with transparency until the lower-third centroid
+    # (the same surface the stride gate reads) sits at the frame's horizontal centre.
+    import numpy as np
+    for o in outs:
+        im = Image.open(o).convert('RGBA')
+        a = np.array(im)
+        op = a[..., 3] > 40
+        ys, xs = np.nonzero(op[int(op.shape[0] * 0.66):])
+        if not len(xs): continue
+        cx = xs.mean()
+        off = round(2 * (cx - im.width / 2))   # feet right of centre -> pad RIGHT, and vice versa:
+        if off == 0: continue                   # new centre (w+|off|)/2 then lands on the feet
+        pad = Image.new('RGBA', (im.width + abs(off), im.height), (0, 0, 0, 0))
+        pad.paste(im, (0 if off > 0 else abs(off), 0))
+        pad.save(o)
     # THE STRIDE GATE, on what was actually cut: every neighbouring pair (wrapping — it loops) must
     # differ in the legs, or the cycle holds a duplicated pose. Threshold calibrated against
     # _famref's (30% = a real stride pair; 12 catches only true near-duplicates).