Wire-Plane Sketchpad

Draw a curve. Watch a dot retrace it at the speed your motor cortex would actually use — slowing through tight curls, racing through gentle arcs — then bend the whole plane and watch the rhythm survive.

Demonstrating: “Does the motor cortex draw on a wire plane?” — Patrick Iglesias-Zemmour, arXiv:2603.03337. The paper argues the classic two-thirds power law of human handwriting/drawing (speed ∝ curvature^−1/3, i.e. tangential speed is geometrically equivalent to constant equi-affine speed) stops being a coordinate accident once the plane is equipped with a “wire diffeology”: curves — not 2‑D patches — become the fundamental objects, and the equi-affine arc-length becomes a genuine covariant 3‑tensor under the full diffeomorphism group. Below, the left panel plays out the power law on whatever you draw (v ∝ κ^−1/3) against a constant-speed control dot. In Warp mode, dragging the handles applies a smooth deformation of the plane — a stand‑in for that diffeomorphism — and the speed profile is recomputed from scratch on the warped geometry. If the paper's claim holds, the slow/fast rhythm should still line up with the original, even though every curvature value on the page has changed.

Sketchpad

draw a curve with your mouse / finger

power-law dot (v ∝ κ⁻¹ᐟ³) constant-speed control dot path colored by speed (violet=slow → amber=fast)

Playback speed1.0×

show deformation grid in Warp mode

Draw any loopy or straight shape (loop, S-curve, zigzag all work well) and playback starts automatically. Then switch to Warp and drag the six handles to bend the canvas.

Speed Profile invariant ✓

original curve recomputed after warp

—

pattern correlation between original & warped speed profiles (same material points along the curve, freshly re-measured curvature)

Curvature is measured with the discrete osculating-circle (3-point circumradius) estimator, smoothed slightly to tame mouse jitter, then mapped to speed via v = κ^−1/3 — the geometric heart of the two-thirds power law. The x-axis of this chart is fractional arc length along the curve (the same "material point" index in both the original and warped embeddings), so a high correlation here is a direct, feelable check of the paper's covariance claim, not a coincidence of the picture.