The four proportions are measured exactly. What they add up to is a tier on a 0 to 8 scale, and that mapping is still being calibrated against real photos, so treat the tier as provisional. There is no percentile claim here and there won't be one: this analyzes a single photograph, not a person.

The measurements are exact — they're geometry computed on detected landmarks. What's still being calibrated is the mapping from those measurements to the 0–8 PSL scale, because that mapping needs a large sample of real photos to be meaningful. Treat the tier as provisional and the individual proportions as the real output.

No. The number comes from a fixed formula over measured distances and areas. No model is asked for an opinion, which is exactly why the result is reproducible.

It's read by your own browser, and the scoring runs there. If you check the optional box on the results page, only the resulting numbers — the four scores and the photo-quality measurements — are shared to help calibrate the scale. The image itself is not part of that, and neither are the landmark coordinates.

Because at least one required measurement couldn't be taken: the head is turned too far, the image is too soft, too dark or too bright, the face is cropped at the frame edge, or more than one face is present. A rejection is the honest answer — a score built on an unmeasurable photo would just be a guess wearing a number.

No. This measures proportions in one photograph. It is not a health assessment, not a cosmetic recommendation, and not advice of any kind. Camera lens, distance, and angle all shift the numbers, so treat it as a measurement of one image rather than a verdict on a face.