- Engagement: the person’s attention level to the interaction at a given moment, expressed as engaged, disengaged, or neutral
- Signals: specific social behaviors detected in the interaction, such as hesitation, agreement, confusion, etc.
Which model reports which signals
Every model reports engagement. Which signals it reports depends on what it reads:
Inter-2 Audio draws every conclusion from what it hears, such as tone, pace, pauses, and word choice, so it does not report the four signals whose evidence is mostly visual:
confusion, disagreement, skepticism, and stress. The descriptions below mention visual cues such as eye contact or posture where they apply; Inter-2 Audio judges the same states from their audible cues only. Choose the model with the model field on POST /v2/upload/analyze or WS /v2/stream/analyze.
Probability and Rationale
Alongside the signal label and its timestamps, each detection (an upload signal, or asignal.detected or signal.updated message) can carry probability and rationale. Either one can be absent.
probabilityis a simple likelihood rating (low|medium|high). It helps you judge how reliable a detection is and decide how to use it (for example, you might only act onhighsignals, or showlowsignals as a “maybe”).rationaleis a short explanation of what the model noticed and how it connects to the signal. An Inter-2 Audio rationale cites what was audible. It’s designed to be human-readable, so you can quickly sanity-check the detection and understand what drove it.
Engagement state
These values are the engagement levels in the table above (for example,engagement_status on a V2 upload window) and are separate from the signal types.