Engagement starts when the user performs a signal they can name and repeat—not when they merely appear

What it is

Before the system treats motion as commands, the user must be able to perform an entry signal on purpose: hold a pose, speak a wake word, press a key on a wearable. That signal is separate from later command gestures. It is explicit—afterward people can say “I just signed in.” It is active—the system does not start listening because someone walked into frame. Midas touch requires an engagement state. Entry conditions answer how the user opens that door. Timeout and leave are the other door.

Why it happens

Without an entry edge, the candidate stream and the command stream are one river. The entry signal cuts the session from “motion is seen” to “motion counts.” Executable means it can be done again when it fails. Inferred entry (“looks like they want to interact,” face-on for two seconds) false-opens when people look at the screen to talk. Explicit means the signal has its own definition in the vocabulary, and does not share a wave with “next page.” A physical key or touch strip is the hardest edge. An in-air pose as entry returns to segmentation, so the shape must sit far from living motion. Entry only opens listening. It does not submit a business command, or signing in itself false-fires a function.

Studying it

Compare: listen on entering frame, listen on facing the screen, explicit mid-air entry, explicit key entry. In tasks interleaved with living motion, measure idle false-opens, time to intentional entry, and whether people can report listening state. Wizard-of-Oz can use an experimenter's key as ideal entry to bound the gain. Log the gap between entry signal and first command: too short means people cannot tell sign-up from command. Do not measure entry in a protocol where the experimenter says “go”; that outsources the signal.

Where it stops holding

A physical key on a wearable already is active entry; stacking an in-air sign-up is unnecessary unless the key is out of reach. People who cannot perform that action need an alternate entry (other hand, voice). Booth demos whose presenter's voice is the entry make the product look as if no signal is needed. With several people in frame, entry also has to bind to a person, or one sign-up starts listening for everyone—that is attribution. The entry signal still has to exist first.

Applying it

• Specify an entry action or key separate from the command vocabulary, repeatable on failure, with no business function on entry itself.
• Turn off “listen on appearing.” Appearing may warm the sensor; listening waits for the entry signal.
• In acceptance ask “how do you make it start listening.” If they cannot answer, or can only say “walk up to it,” the entry condition fails.

Related

• Same group: C4.15.2 The entry signal must sit far from everyday motion, or it will fire by accident · C4.15.3 The entry signal itself needs immediate feedback that listening has started · C4.15.4 An entry that is too heavy cancels the contactless advantage over touch
• Adjacent: C4.02 The Midas touch problem · C4.16 Disengagement and timeout
• Search: engagement entry · clutch in · wake gesture