One photograph tells you very little. Lighting, camera and time of day move the pixels more than your body does. So we built the opposite of a one-shot reading: an instrument that learns what your usual range looks like, then reports only what moved beyond it.
De-identification is not a policy we promise. It is a step that happens first, in the pipeline, before any model sees the image.
The face is masked and the tongue is segmented out of the frame. Metadata is stripped. What continues down the pipeline is a surface, not a person.
Blur, exposure and framing are scored before analysis. A capture that can't support a reading is rejected rather than guessed at.
Colour, coating, texture, moisture and surface marks — described in a fixed vocabulary, with a confidence value and a record of exactly which image produced it.
Each result carries the hash of the image that produced it, the engine version, and a run identifier. If a reading looks wrong, it can be reproduced and examined rather than argued about.
To say something changed, you need something to compare against. There are two honest ways to get one, and you don't need to buy anything for either.
| Reference | What it gives you | What it needs |
|---|---|---|
| Shade guide in frame | A manufactured colour standard with known values lets us solve for the lighting instead of estimating it — so readings hold up across different rooms and different cameras. | A shade tab your dental office already owns. |
| Your own baseline | Three to seven captures teach the instrument how much your readings naturally wobble. Anything that exceeds that wobble is a real move, not noise. | Nothing. Just the first week. |
The second is not a lesser version of the first. It fully answers the question most people actually have — is this changing? The first answers an additional question about comparing across people, which most people don't need.
Your baseline keeps moving with you. Seasons change, phones change, lighting changes — so it's a rolling window, and a poor capture is never allowed to enter it.
An instrument that always produces a number is indistinguishable from one producing noise, and you'd have no way to tell which you were holding. So when two captures aren't genuinely comparable, we don't estimate. We say so, and we tell you how to fix it.
Refusing to show a number is what makes the numbers we do show worth anything.
Every limit below is real and none of them are scheduled to disappear quietly. If that changes, this page changes with it.
This is an instrument under active development. We are not selling lifetime passes or counting down spots. If you want to use it early and tell us where it fails, we would like to hear from you.
Practices interested in the shade-guide workflow: that path is being built with the people who would use it. Same address.