4,200 characters in 2 seconds. Zero human signal.

A memo arrives in your doc tool. It has no typing cadence, no mouse jitter, no idle pauses - just instant output. Manav flags low human entropy and offers a one-click relabel as "Human-Supervised AI" if a person stands behind it.

Before: LOW After: MEDIUM

The catchAI-only output is fine. Calling it human work is not.

Human pane · Asha 0:00

Heartbeats0

Wpm0

Entropy0.00

AI pane · @minashah 0:00

Heartbeats0

Wpm0

Entropy0.00

Time compressed 60× for the demo. Human side simulates 4 hours of organic typing in 4 seconds.

Human entropy score0%

heartbeats × cadence × mouse entropy × pauses

AI entropy score0%

heartbeats × cadence × mouse entropy × pauses

LOW_HUMAN_ENTROPY · 76% confidence. The 4,200-char memo on the right has no heartbeats, no mouse entropy, and a 2-second creation window. The detection rule fires - the memo lands in the dashboard with an Unverified label, not a Verified one.

Detection JSON

{
  "type":        "LOW_HUMAN_ENTROPY",
  "severity":    "medium",
  "confidence":  0.76,
  "evidence": {
    "textLength":             4200,
    "createdDurationSeconds": 2,
    "heartbeatCount":         0,
    "mouseEntropy":           0.02
  }
}

The relabel path

A human can stand behind the AI output - phrase challenge + 3-second hold. The artifact is re-stamped as Human-Supervised AI and the anomaly closes.

Human-Supervised AI

Relabeled. The proof page now shows Human-Supervised AI and the anomaly is closed.

Security review · 4 min

Keep the tour going.

Next: Worker timeline - Every minute of one worker's day, audit-ready.

Continue tour