Methodology · rubric v2.2
Sixteen sub-factors. One formula. No black box.
A score you can't interrogate is a horoscope. The model only does perception - it scores sixteen named sub-factors against a fixed written rubric. Every number you see (feature scores, your overall, your potential) is then computed by us with a published weighted formula. The model never hands you a number it made up.
The five features and their sub-factors
Jawline
22% of overall- Definition - Visibility of the jaw-to-neck transition.Partly40%
- Under-chin - Fullness under the chin (body-fat and hydration responsive).Movable35%
- Contour - Underlying jaw shape as presented (not changeable by habit).Structural25%
Skin
24% of overall- Clarity - Visible breakouts and blemishes.Movable35%
- Evenness - Redness and tone uniformity.Movable25%
- Texture - Surface smoothness and visible pores.Partly20%
- Hydration - Dull and dry versus healthy and supple.Movable20%
Eyes
20% of overall- Under-eye - Dark circles and puffiness of the under-eye area.Movable45%
- Brows - Brow grooming and upkeep (not brow shape genetics).Movable30%
- Freshness - How rested and alert the eye area looks - a fatigue signal, never an emotion read.Movable25%
Balance
14% of overall- Balance - Left-right evenness as presented in a front-facing photo.Partly45%
- Grooming evenness - Evenness of grooming across the face.Movable30%
- Proportion - Presented facial proportions (not changeable by habit).Structural25%
Hair
20% of overall- Cut - Quality and freshness of the current cut.Movable40%
- Condition - Health, shine, and frizz.Movable35%
- Style fit - How well the style frames the face as presented.Movable25%
Movable, partly, or structural
Every sub-factor is tagged by how much a 90-day protocol can actually move it. Movable sub-factors (skin clarity, under-chin fullness, brow upkeep) respond to grooming and lifestyle. Structural ones (bone contour, facial proportion) do not - so we lock them and they contribute zero to your potential. Your potential is your current overall plus only the headroom on movable sub-factors, never raised above 88 on any single one. That is why our potential is a promise we can keep, not a number designed to flatter you.
Confidence, not false precision
We run the scoring several times and measure how much the passes agree. Tight agreement reads as high confidence; lighting, angle, or a hat that blurs a sub-factor lowers it - and we tell you, rather than printing a precise-looking number we don't trust. Low-confidence reads come with a nudge to retake in even light.
How a score is produced
- 01Your photo is normalized deterministically - same resize, same crop, same processing every time, so identical photos produce identical model input.
- 02The model scores sixteen sub-factors against fixed written band definitions (the rubric) - pure perception, no overall, no improvised judgment. The rubric is versioned; your score is stamped with the version that produced it.
- 03Inference runs with zero sampling randomness.
- 04The full pass runs three times and we keep the median per sub-factor. If the passes disagree, we automatically run more passes before trusting the number, rather than reporting false precision.
- 05The app - not the model - computes your feature scores, overall, potential, and confidence from those sub-scores using the published weighted formula below.
- 06The photo is deleted from memory before the response returns. Only the numbers persist.
The result: scan the same photo twice and you get the same score. Why most apps can't say that →
What we will never score
The output schema physically has no fields for these - it is not a policy toggle, the model cannot return them:
- ✕Race or ethnicity
- ✕Age
- ✕Gender expression
- ✕Emotion or mood
- ✕Eye shape or colour
- ✕"Masculinity" or "femininity"
Some competitors advertise "masculinity" scores. We think that's both scientifically hollow and corrosive - and we built the schema so we can't drift into it.
The honest caveats
This is a measurement of a photo, not of you - lighting, angle, and camera quality are part of what any vision model sees, which is why re-scans ask for consistent conditions. Scores are rubric-anchored opinions of an AI model, useful for tracking your own change over time - they are not clinical assessments, percentile claims against humanity, or anything medical. For skin or health concerns, see a professional, not an app.
Photo handling details: what happens to your photo.