vyuuhNo biometric template
Nothing in the product creates, stores or compares a face template. There is no face recognition, no face matching and no liveness check.
Trust
Independent witnesses.
One Proven Event.
A Proven mark means separate people, with nothing connecting their phones, were filming the same Event.
That is what the rest of this page explains, including how it is decided and what it does not claim.
What witness verification means.
Two phones at the same event do not know about each other. They are not paired, they share no network, and neither is told what the other is looking at. What they do share is the physical world in front of them.
Each device works out an identity for what it is observing, from its own position and what it can measure about the scene. When two devices independently arrive at the same identity, that agreement is evidence they are watching the same thing. There is no coordination between the devices at capture — no pairing and no shared local network is required, though they may happen to be on one. That absence of coordination is what makes the agreement worth something.
Two or more independent viewpoints, together with a convergence signal or overlapping fields from co-located cameras, mark an Event Sealed. Proven is a separate finding on the record, and corroboration alone never reaches it. Grouping alone is never Proven, and neither mark says that what the footage shows is true.
Witness verification is evidence of co-witnessing, meaning these cameras were observing the same Event. It does not make what they show true, settle any claim about it, or confer rights over the footage.
The published work
The method is written down and citable.
Coordination-Free Deterministic Event Identity for Independent Mobile Observers.
Preprint, CC BY 4.0, Zenodo, 2026.
doi.org/10.5281/zenodo.21694604
That is the concept DOI. It resolves to the current version, not to one frozen revision.
A preprint describes research. It is not a certification of any product feature, and the supported verification workflow should be demonstrated before it is relied upon.
Privacy
What the product does with faces, and what it does not.
The identity a device derives comes from what the device itself measures about the scene. It does not use faces, and there is no facial field anywhere in that derivation.
Face detection exists in exactly two places in the app: a camera effect, and cropping a profile picture the user chose themselves. Both run on the device, through the operating system's own vision framework, and both ask only for a bounding box.
No facial geometry
Landmark extraction was removed in September 2026. What remains is a rectangle and a quality score, held in memory for the frame being processed.
No face data leaves the device
No face-derived value is persisted, logged or transmitted. This is about face PROCESSING, not about video: a broadcast obviously transmits pictures of people, and a profile picture a user chose is uploaded as an ordinary image. Neither carries anything derived from face detection.
No third-party processor
There is no cloud face service and no third-party biometric SDK in the product, so there is no sub-processor to disclose for face data.
The language we hold ourselves to
Three rules we do not bend.
Grouping alone is never Proven. An Event can hold several viewpoints and still not have earned the mark. The app shows which state it is in.
You see the viewpoints that exist. We never imply an angle that nobody captured.
Corroboration is about the occurrence. It says these cameras saw the same thing. It does not adjudicate what the thing was.
Let’s start with one event
A different perspective.
A practical first step.
Tell us about your venue or show. We will explore what a supported pilot could look like.