What it can do
Every one of these is measured from consented video, on-device by default, and reported with the uncertainty attached.
Aggregate attention, energy, and synchrony
From video of a room, the engine estimates how much of the group is oriented forward, how the collective energy moves over time, and how synchronized the room is. These are aggregate, group-level reads, not scores about any one person.
Behaviour and geometry correlates over time
It tracks facial landmark geometry and body posture frame by frame, and reports patterns (attention arcs, stability, movement) as behavioural correlates. It reports what the geometry did, not what a person felt.
A room count that is an honest lower bound
At crowd scale the engine reports the number of people it could detect and says plainly that this is a lower bound, not a certified headcount. When the room is too dense or the faces too small to read reliably, it reports a count only, or abstains.
Single-subject facial and body signals, with confidence
For a single person, it reports facial-behaviour and body-language signals with a confidence level attached to each. When a signal cannot be measured reliably, it abstains rather than guessing.
What it does not do
These are not omissions to fix later. They are deliberate limits, enforced in the product and in the terms.
It does not read emotions or minds
GRW does not read emotions off a face. It measures geometry. There is no established scientific basis for inferring a specific emotion from facial biometrics, and the product does not claim to. In EU and UK workplace and education settings, any emotion-framed output is gated off entirely (EU AI Act Article 5(1)(f)).
It does not diagnose
Nothing the engine produces is a medical, psychological, or psychiatric diagnosis. It is not a clinical instrument and must not be used as one.
It does not identify or track individuals at crowd scale
The crowd read is aggregate only. It does not name, re-identify, or build a per-person profile of anyone in a large room. There is no per-person score at crowd scale.
It is not surveillance, and not a lie detector
GRW is a professional assessment tool for consented footage. It is not a monitoring system, an attendance tool, a productivity tracker, or a deception detector, and must not be deployed as one.
It does not rank you against a population it has not measured
The engine does not surface a percentile or "elite average" benchmark unless it has a validated reference cohort to stand behind it. Where no such cohort exists yet, it withholds the comparison rather than fabricate one.
It is not for employment decisions in gated settings
GRW outputs must not be used for hiring, discipline, performance management, or surveillance in EU and UK workplace and education contexts. Those uses are gated off in the product and prohibited in the terms.
When it is not sure, it says so
The engine attaches a confidence level to what it reports and abstains when a signal cannot be measured reliably. A blank or withheld result is the honest answer, not a failure. For the legal basis behind the gated uses above, see our privacy and terms.