Independent runtime & integration validation with real human interaction —
the step synthetic vectors cannot replace. Everything runs 100% on your device;
nothing is transmitted automatically. The report is generated locally and leaves
your device only if you download and submit it yourself.
Protocol: verification/FIELD-TEST.md
How it works
1. Consent. The test records a behavioral session. Explicit opt-in is required — the seal is attribution, not consent.
2. Interact naturally for ~90 s. Read this page, scroll, move the mouse, type a few sentences in the box below — exactly as you would normally. Three short micro-tasks will appear (tap · press & hold · guided move) — the v1.3 challenge-response channel that tests motor execution. They take a few seconds each and never block the rest of the session.
3. Get a report. A structured JSON report (classification curve, tempo, pauses, context, challenge trust, device class) is generated on-device. Download it and attach it to a GitHub issue — or keep it for your own evaluation.
No keystroke content, no pointer coordinates, no raw biometrics are ever recorded or exported — only on-device computed, quantized values (see the limits field of the report).
Live session
Interact naturally — read, scroll, type in the box below, move the mouse.
Micro-task
TAP
Elapsed
0s
Confidence
—
Classification
—
Events
0
Pauses >2s
0
Tempo trend
—
Challenges
0 / 3
Goal: 90 s of natural interaction. You can finish early after 30 s.
The report contains only quantized, on-device computed values plus coarse device class (screen size, touch capability, user agent). Review it before submitting.