Every score is explained in-product — with the caveats that keep it fair.
Opaque scores create grievances and churn. So every number here is published, explainable, and visible to the person it describes. And each one carries the same honest warning: it measures interaction, not value.
Activity level
A 0–100 score measuring input density in a rolling 10-minute window.
Published formula
raw = 0.6 × min(keys / 600, 1) + 0.4 × min(mouse / 400, 1) · activity_level = round(100 × raw)- 0No input — idle or away
- 1–29Low — reading, watching, thinking, on a call
- 30–59Moderate — mixed reading and writing
- 60–100High — sustained typing or heavy mouse work
The mandatory caveat, shown wherever this score appears
It measures interaction, not value. A designer, a developer reading code, and a person on a video call all legitimately produce low activity levels — so it must never be the sole basis for a performance decision, and the product says so on the report.
Productivity score
The share of active time spent in apps and sites categorized productive for that role.
Published formula
productivity = 100 × productive_seconds / (productive + unproductive)The mandatory caveat, shown wherever this score appears
Neutral is excluded from the denominator, never counted as unproductive — counting Slack or email as unproductive is the classic mistake that makes these tools distrusted. Below 60% coverage the score renders greyed as 'insufficient classification' rather than presented as fact. Categories resolve per role, not globally.
Screen productivity
The score-only alternative: productive foreground seconds over total foreground seconds — with no image captured.
Published formula
screen_productivity = 100 × productive_foreground_seconds / total_foreground_secondsThe mandatory caveat, shown wherever this score appears
The reason we can deploy where every screenshot vendor is banned. Reported with an hourly breakdown and top productive / unproductive apps — but no image is ever captured, so there is nothing to blur, redact, subpoena or leak. Ships on every plan, including Free.
“Worth a look” — never “violation detected”.
Data-quality flags surface a pattern for a human to look at. The product makes the conversation easy, not the accusation — it never silently trusts a signal, and never silently discounts one.
Automation tooling
Perfectly periodic mouse events, zero keyboard, static screenshots.
Ghost hours
Tracked with near-zero activity for over two hours.
Time anomaly
Wall-clock and monotonic clock diverge on a device.
Attribution gap
A high unassigned ratio for a member over several days.
Capture failure
Screenshots or app sampling failing on a device.
Self-report spike
A sudden rise in manual offline entries — usually a broken agent.
What we deliberately do not ship:
a single ranked “employee score”, activity level as a payment gate, silent scores the member can't see, named-individual job-search alerts, or screenshot OCR. Each one is statistically meaningless, unfair, or an invasion of privacy — so it stays out.
And every report a manager can run about a member, that member can run about themselves.