How long an indicator persists after input stops determines its accuracy

What it is

A presence timeout is the decay window during which “editing” remains visible after the last event. Too short a window disappears during thought or reading, creating false negatives; too long leaves a ghost occupant after departure, creating false positives. Accuracy is therefore an estimate of continuing intent, not simply fast disappearance.

Why it happens

Systems observe input, focus, and connection, not intent. Timeout converts intermittent events into a continuous state and acts like a lease: activity renews it, explicit exit ends it, and disconnection lets it expire. Natural pause distributions differ between chat and complex writing, so one interval creates different biases. A multi-signal state machine is often more faithful than keystroke timing alone.

Studying it

Extract inter-input intervals, focus changes, commits, and returns from real sessions by task type. Replay candidate windows to compute false-positive duration, missed activity, and flicker, then test whether errors change yielding and conflict. Ground truth about continuing intent requires self-report or replay interview rather than circularly defining it by the next keystroke.

Where it stops holding

No timeout fits every device and task. Mobile backgrounding, assistive input, dictation, and long reflection alter event rhythms. Using presence decay to infer labor is an invalid secondary purpose. Disconnection should remain distinct from voluntary stopping.

Applying it

• Maintain an activity lease from input, focus, commit, explicit exit, and connection signals.
• Calibrate by task type and prevent flicker across ordinary pauses.
• Distinguish typing, possibly continuing, and disconnected instead of claiming binary certainty.
• Compute false positives, misses, and flicker from replay, then validate impact on conflicts and unnecessary waiting.

Related

• Same group: V2.06.1 Cues support yielding rather than blocking · V2.06.2 Cues must appear when editing starts · V2.06.3 Cues need local scope · V2.06.5 False positives destroy trust
• Adjacent: V2.01 Presence awareness · V2.08 Availability and willingness to be interrupted
• Search terms: presence timeout · activity lease · presence decay