Timeout length is a trade: cut people mid-thought, or keep listening after they have walked away

What it is

If automatic leave uses a timeout, the duration has no correct value, only a trade. Too short: people pause to think, read a subtitle, answer someone beside them, the session dies, the next command has to sign in first—an unexpected cut. Too long: they have walked away, the system still treats passing motion as commands, camera and recognizer still run, privacy and false fires are not released. Duration is the product's operating point between those two failures. It must be stated per task. It cannot be read off a gesture definition.

Why it happens

The timeout clock measures “time that did not count as input,” not wall-clock. Continuous adjustment, aiming, watching feedback, may produce no new token and still be inside the session. If the clock counts those as idle, it cuts mid-operation. If it only runs on total stillness, living fidgets in a room keep it from ever filling. On the resource side, listening on means capture on; duration is the capture window. Public devices lean short, to hand the session to the next person. Private living rooms lean long, so someone watching a film can pinch once in a while. One duration cannot serve both rooms.

Studying it

Sweep timeout from seconds to minutes. Tasks include thinking pauses, watching content, walking away to pour water. Measure mid-task cut rate, time still listening after walking away, cost of re-entry, idle false fires. Split by scene: browsing, watching, public terminal. Let people set the duration and look at their operating point, but do not take self-set as the only evidence—people will tune until they are no longer annoyed, not until privacy is optimal. Log how long the capture indicator stays on, as an objective of resource release.

Where it stops holding

Law or venue rules that capture must stop the moment nobody is there shrink the trade. Continuous drag should pause the clock, or any duration will be misread during a long drag. Network delay makes the timestamp of “last input” unreliable and the clock skews. Children leave longer gaps; an adult duration will cut them. Timeout is not an entry condition, and must not be used as a single gesture's offset.

Applying it

• Nail an operating point per scene: short for public, longer for private watching, and say whether the clock pauses during aiming and continuous control.
• Accept on both columns: “thinking pause was cut” and “still listening after walking away.” Optimising only one is not a pass.
• Log actual capture-window length. If it is routinely far longer than the task, shorten it; do not add teaching.

Related

• Same group: C4.16.1 Leave can be explicit or automatic · C4.16.3 Warn before an automatic timeout so people are not caught mid-move · C4.16.4 Explicit leave should be symmetric with entry, for memory
• Adjacent: C4.03 Gesture onset and offset detection · C4.29 Visibility of capture state
• Search: session timeout · idle timeout · privacy window