Attention scoring

How OnlookAI computes a 0–100 attention score per session.

Overview

The scoring engine processes a rolling window of attention signals and outputs a single score between 0 and 100. The score is updated in real time as new signals arrive.

Signal types

Signal	Effect	Points
TAB_HIDDEN	User switched away from tab	−20
WINDOW_BLUR	App window lost focus	−20
IDLE	No input for 30s	−20
NO_FACE	Face not detected in camera	−25
TAB_VISIBLE	User returned to tab	+10
WINDOW_FOCUS	App window regained focus	+10
ACTIVITY_RESUMED	Input detected after idle	+10
FACE_DETECTED	Face detected in camera	+10

Rolling window

Scores are computed over the last 10 events using a ring buffer. This makes the score responsive to recent behavior without overreacting to momentary distractions.

Thresholds

The default intervention threshold is 40. You can configure this per session:

const session = track(userId, { threshold: 60 })

When the score drops below the threshold and no intervention is currently active, the drop event fires.

Score ranges

Range	Meaning
75–100	High attention
50–74	Moderate — monitor
0–49	Low — consider intervening