The return stroke is part of the gesture, not leftover motion

What it is

After the hand goes out, it comes back. The return stroke (retraction) is not a private habit. It is a segment of the gesture definition: leave rest, complete the stroke, come back to a pose from which the next token can start. If the definition says only “wave right,” the recognizer will harvest the leftward return as another command, or decide there is no offset while the hand still hangs at the far point. A teaching image that shows only the end pose leaves people unsure whether to reset. In the field you then get the pair of failures: “I finished but my hand is still in the air” and “reset was taken as the opposite gesture.”

Why it happens

Open-chain motion has to close. After shoulder and elbow send the hand out, gravity and joint limits pull the limb toward a cheaper posture. That return path is often a mirror of the outbound path, and fast enough to match another template. If the vocabulary holds both “next” as right and “previous” as left, the return is a perfect lure for the opposite command. Writing the return into the definition tells the recognizer: after a successful outbound match, the following mirror stretch is the tail of the same token and must not fire again. Commit may sit at the end of the outbound stroke or at the end of the return; either way the return still belongs to this token and is not new input.

Studying it

Build two gold standards for the same action: outbound-only and outbound-plus-return. Count extra class emissions during the return. Independent variables include whether the vocabulary contains mirror pairs, and whether commit sits at outbound end or return end. Dependent measures are mirror false-fire rate, whether a next token can start immediately after reset, and whether people report “it turned the page again on the way back.” Have people deliberately not reset (hand parked at the far point) and deliberately snap back, to separate hangs from double fires. Whether teaching materials draw the return should be its own factor, not a later complaint that users “did it wrong.”

Where it stops holding

One-way, no-reset actions exist: drag a virtual object to a new place and leave the hand there. Then the return is not part of this token; it is preparation for the next, and the definition must say “do not reset.” For continuous control (pinch-and-drag) the return happens after release and should count as leaving the adjustment, not as a new swipe. On a headset, pinch-release is a contact edge, not a spatial return; do not write the two tails as one. Making a mirror vocabulary asymmetric (keep right, give left a different shape) weakens the lure, but the reset path still has to be stated or people will not know where to put the hand.

Applying it

• Every gesture description includes outbound path, return path, and rest pose, with commit nailed to either outbound end or return end.
• If the vocabulary has mirror pairs, regression must include “outbound then stop” and “full reset,” and count reverse false fires on purpose.
• Run teaching animation until the hand is back at rest; do not cut at the highest or farthest point.

Related

• Same group: C4.03.1 Every gesture needs a decidable start and end · C4.03.3 Using dwell as an onset/offset signal costs waiting · C4.03.4 Onset/offset failure appears as swallowed or repeated actions
• Adjacent: C4.07 Mid-air swipe · C4.08 Flip and rotation gestures
• Search: return stroke · gesture retraction · mirror false trigger