Skip to content

Trust

How ProofLoop works, and what it can't tell you

Plain answers about the measurement, your data and the matching logic.

What is measured

Your framed photo is sent from the ProofLoop server to the YouCam AI Skin Analysis API (POST /s2s/v2.0/task/skin-analysis), which returns a score from 1 to 100 for eight concerns: hydration, oil balance, redness, texture, pores, blemishes, radiance and under-eye darkness. YouCam documents these as scores where higher is better, adjusted to be encouraging. ProofLoop shows them as returned; if a concern isn't returned, nothing is shown for it.

Normal variation. On 11 October 2026 we ran ten live analyses of one face (eight different framings plus two exact repeats). The same photo twice gave identical scores. Reframing the face (65–80% of the photo width), shifting it, or making the photo 18% darker moved any score by at most 2 points. The same photo cropped in the browser and re-encoded moved pores by 5 and texture and hydration by 3 — those fine-detail measures are sensitive to image sharpness. ProofLoop therefore only calls a change beyond ±3 points for radiance and under-eye, ±5 for hydration and ±7 for pores and texture. Blemishes, oil balance and redness scored 99–100 on the test face, so their variation wasn't measured. Rather than guess, ProofLoop gives them the widest band it measured for any concern (±7) and labels it “assumed” wherever it's used. This is a small, single-face experiment, not a validation study; real re-captures on different days vary more, which is why every comparison also checks lighting and capture method and says “we can't call this one” when they differ too much.

Where your data goes

  • Your photo is decoded, stripped of all metadata (including location) and re-encoded on our server, sent to YouCam, and then discarded from memory. ProofLoop never writes it to disk or a database.
  • In your browser, the framed photo is kept in this tab's session storage so YouCam's detection masks can be overlaid and a preview generated. It disappears when you close the tab or delete your data.
  • On our server we keep scores, the time of each scan, the photo's size and average brightness (for comparability), a one-way fingerprint (SHA-256) of the cleaned photo so a later check-in or instrument check can't silently reuse the baseline photo (the photo can't be rebuilt from it), and your trials — linked to an anonymous session token held in your browser. No name, email or account.
  • YouCam keeps different data for different periods, according to its documentation. The photo ProofLoop uploads, and its file ID, are stored for 30 days, and uploaded and generated images are removed automatically after 30 days. Skin Analysis results are retained for 24 hours after the analysis completes. YouCam's own terms govern its processing.
  • Deletion: “Delete all my data” on My trials removes the session and everything linked to it immediately. A usage ledger keeps only unit counts, with no link to you.

How products are matched

Matching is a fixed set of rules, not an AI model. A product qualifies only if the catalogue lists your focus as one of its targets. It's then placed in a tier — strong, good or possible — depending on whether your focus is its main job, whether any caution applies to your measurements (for example, an exfoliating acid when redness scored low), and whether it fits your texture preference. Budget and fragrance-free are hard filters, and anything set aside is listed with the reason.

There is no match percentage, because there's no data that would make one meaningful. The catalogue is a fictional sample built to demonstrate the logic.

Skin-type reading and your answers

Where skin profiles are switched on, the same YouCam request that measures your eight concerns also returns a skin-type reading of the photo for the whole face, the T-zone and the cheeks (for example, “T-zone: oily”). It costs 14 units instead of 12; adding it left the eight scores unchanged when we tested the same photo with and without it. The reading is stored with the scan, like the scores, and describes how your skin looks in that one photo. It isn't a diagnosis, it can change with season, washing and light, and ProofLoop never uses it to measure your trial or to choose products. A redness reading describes appearance only; it never means your skin is sensitive.

Your own answers are optional. They stay in this browser tab unless you tick “Save my answers with this trial”, in which case they are stored with that trial, together with the reading, as they were when the plan was fixed. Changing them later is recorded beside the original and never changes the plan or the result. “Remove profile” on the trial page deletes the answers and the readings; “Delete all my data” deletes everything. The answers and readings are never logged and are never sent to Gemini.

About goal previews

Previews are generated by the YouCam AI Skin Simulation API (POST /s2s/v2.0/task/skin-simulation) from your own photo, only when you ask, and only for concerns that API supports. They show what a goal looks like. They are not a prediction of what any product will do.

The fair-test plan

When you start a trial, ProofLoop fixes its plan before any check-in exists. The plan holds one focus score, the band that score must move beyond, the exact scores that would count as up or down, a planned check-in date (day 28), and the baseline photo's lighting and capture method. Later changes to ProofLoop's bands don't move a running trial's lines.

  • One score decides. A check-in shows all eight scores, but only the focus decides the verdict. With eight scores compared at once, one can cross its band by chance, so the others are shown as context only.
  • Some focuses can't show a gain. The scale stops at 100. A score of 99 with a ±7 band would need 107 to count as an improvement, so ProofLoop says so before you start, and lists those focuses last.
  • Free check before the scan. A check-in photo is first compared with the baseline's lighting and capture method, without calling YouCam. If they differ enough that the verdict could only be “we can't call this one”, you're told before any units are spent.

What ProofLoop cannot conclude

  • Why a score changed. A product, the season, sleep, stress or the rest of your routine can all play a part.
  • Anything medical. ProofLoop doesn't diagnose or treat; see a professional for skin concerns.
  • How you compare with other people. Scores are most useful compared with your own earlier scans.
  • Anything from sample data. Sample and synthetic data are labelled everywhere and never mixed into real trials.