Three steps, and an honest account of what each one can and cannot tell you.
Front on, hair back, no makeup, daylight from a window rather than overhead or direct sun. The analysis reads five things off the pixels — average luminance, luminance variance, local edge density, red-channel dominance and saturation spread — which stand in for surface hydration, evenness, texture, redness and congestion. It runs in your browser using the canvas API. The image is never uploaded, and closing the tab discards it.
A photograph cannot feel tight at eleven in the morning, and it cannot know that every new serum stings for a minute. The questions carry what the camera misses. Answer three and you get a reading; answer six and add a photo and you get a much better one.
Your six scores make a shape. We take your two lowest axes and score all fourteen formulas against them, weighted so a formula treating both of your weak points beats one treating a single strong one. The match is re-run on our server so it cannot be edited in your browser. That is the whole logic — deliberately simple, and we would rather show it to you than call it proprietary.
FACET is a cosmetic recommendation tool. It cannot see below the surface, it does not diagnose anything, and it has no opinion on moles, rashes, cysts or anything that hurts. Those belong to a doctor. Nothing here is a prescription and no product in any box is a prescription product.
Skin in February is not skin in August. On the plan you re-scan every ninety days and the next box follows the new reading — different formula if the shape changed, same price band, no charge to switch.