your skin has twins.
the engine already knows them.

every goodbadgray profile is a 14-metric weight vector — real skin-type weights × concern tilts × life-stage overlays, straight from the live engine's config. profiles cluster. clusters transfer knowledge: what the engine learns for your twins, it already knows for you. the cohort below is 50,000 simulated profiles (labeled illustrative) — but the geometry they land on is the engine's own math.

8,595

ingredients scored in the live engine

770

engine-exact weight combos map this space

real product-for-profile scores harvested

87.8%

of config variance in these two axes

profile-space

illustrative cohort · 50,000 simulated profiles

axes are a pca projection of the engine's real combined weight vectors (5 skin types × 22 concerns × 7 life-stage states = 770 combinations). dots are simulated; positions come from real config math. rings are the 8 real, engine-scored personas.

← irritation & barrier weighted

comedogenicity weighted →

↑ sensitization-weighted

↓ hydration & barrier weighted

simulated profile illustrative engine-scored persona real you

loved by profiles like yours

real engine scores

one-size-fits-all your twins' weighting

cold start ranks every product with the config-average weight vector — the best the system can do knowing nothing about you. warm start uses your twin persona's real engine scores. watch the ranks move.

cold start · config-averagewarm start · your twins

the flywheel

illustrative

every scan makes the next verdict smarter.

a scan becomes a profile. profiles that weigh alike become twin clusters. clusters transfer evidence — what worked, what irritated, what got gated — into sharper priors. sharper priors mean better verdicts on day one, and better verdicts bring more scans.