Say too-far and too-close out loud, and do not use the same lamp as a missed gesture

What it is

When the system already knows whether a failure is distance or gesture, it must say those as two different states. Too far / too close is stance. Seen, but not an item in the vocabulary is recognition. One “not detected” lamp for both makes people change the wrong variable. Active expression means not waiting to be asked: emit the stance signal at the moment of crossing. Progressive cues on volume walls, and how three visual encodings differ, are later feedback design. At the distance layer, at least: far, near, and misread, as three separable utterances.

Why it happens

On the system side the diagnosis is already split: joint confidence falling with distance, whether the hand fills the frame, the classifier's score on a visible trajectory. Collapse that into one Boolean “was there an event” and the diagnosis is thrown away. Repair strategies on the user side are bound to diagnosis: a stance signal should make people step in or back; a recognition signal should make them change the move or open teaching. If the signals look the same, both strategies are tried at random and task time is spent on the wrong repair. A distance signal must also be able to appear with no gesture: someone merely standing far, not yet moving, should still see “please come closer,” or the ruler is only revealed after a failure.

Studying it

Create three failures: stand past the far edge on purpose, stand inside the near edge, produce a wrong move at a comfortable distance. Pair each with three feedbacks: none, a unified “not detected,” and split stance/recognition. Measures: time to first correct repair, whether repair type matches cause, count of wrong repairs. Add a “stand far, do not gesture” cell to see whether the distance hint is understood before anyone moves. Channel (on-screen words, lamp colour, sound) as a factor, checking actual distinguishability, not three copy lines on one lamp.

Where it stops holding

When language fails or the screen is out of sight, written stance hints are nothing; a non-visual channel is required. Too-frequent “please come closer” becomes noise, people mute it, and the edge is invisible again. With several people, the stance hint must say who is too far, or the near person crowds nearer. In a headset, “hand left the field of view” and “hand too close and soft” are both near-edge, with different repairs (bring it back vs pull it away), needing more than “too close.” That is a headset special case; room cameras usually need only far/near.

Applying it

• Prepare three different copy-and-signal sets for far edge, near edge, and recognition miss. Ban a shared “not detected” lamp.
• When someone has crossed interaction distance without yet moving, emit the stance hint. Do not wait for a failure.
• Accept with three injected failures and check that the repair matches the cause. If it does not, change the signal, not the vocabulary.

Related

• Same group: C4.10.1 Interaction distance should sit inside capture range · C4.10.2 The coordinate anchor decides consistency after movement · C4.10.3 Distance edges are invisible, so leaving them and a miss look the same
• Adjacent: C4.22 Perceiving interaction-volume boundaries · C4.11 Substitutes for contact feedback
• Search: range feedback · too-far indicator · error differentiation