Proof of Presence
Pulse under the skin, a randomized screen-reflection challenge, and lip–voice alignment — scored on the weakest signal, not the average.
Three physical signals, weakest-link scored
Remote photoplethysmography recovers a blood-volume pulse from subtle skin-colour change using the CHROM method, band-limited to 0.7–4.0 Hz with a periodogram SNR driving confidence.
The active challenge paints a randomized colour sequence issued by the server and correlates the observed skin reflectance against the emitted sequence across plausible frame lags. A replayed video cannot reflect a sequence it has never seen.
Cross-modal sync correlates audio energy against mouth-region motion, and honestly reports itself unmeasurable when there is too little speech to judge.
The overall score is the minimum of the contributing signals rather than their average, so a strong signal cannot mask a failing one.
Check it yourself
The artifacts behind this page
- Signal values are visible live in the demo's sideband readout.Watch the signals →
What this page does not claim
Limits
- Synthetic-voice detection is not implemented. The field exists in the contract and the policy threshold is defined, but the value is reported as a neutral constant rather than a measurement — a number nobody measured must not be allowed to pass a gate.
- Blink and micro-saccade analysis are not implemented.
- Frame-cadence analysis is a weak signal in a browser and is treated as corroboration, never as proof.