A little context changes everything.

Every ingredient has a job. The useful question is whether those jobs add up to the cosmetic result you want.

Each category has its own goal.

Each goal has a function profile. We compare it with the identified functions across your products, accounting for their role in your routine. Alternative functions can fill the same gap. Hair and skin routines are considered separately.

Your product goal-match estimate.

We compare the ingredient function groups relevant to this product’s role with your selected cosmetic goals. Goals and optional profile answers set the weights. One matching function can satisfy a group; repeated ingredients add no extra credit. Conflicting functions reduce the result. An ingredient you explicitly avoid sets the match to 0%. We need a selected goal, a known product role and at least 60% of ingredient names mapped to estimate a percentage. Missing data is never scored as 0%.

Calculation: 100 × matched group weight ÷ (all relevant positive group weights + present conflict weights), rounded. An explicit avoid-list ingredient sets an otherwise estimable match to 0%. Preferred ingredients add context, not score points. A 100% match describes the recorded function groups for this product role; it does not mean perfect suitability.

Label-based estimate, not a safety rating or a prediction of results. Label coverage describes mapped names, not a probability of results. Current mappings are provisional and need expert validation. Ranking compares candidates on the current page, not every product worldwide.

A label has limits.

An ingredient list does not tell us concentration, the complete behaviour of a formula, or how it will feel on you. A function being present is a clue, not proof of a result. UV filter names cannot establish SPF or broad-spectrum performance; use a labelled sunscreen and follow its directions.

Evidence should be visible.

This build includes a provisional starter catalogue. Its goal mappings have not yet been reviewed by a cosmetic formulation expert. Ingredients without a specific mapping stay neutral. Unknown ingredients remain visible and are excluded from the calculation.

• Regulatory: an exact fact from an official record.
• Strong: multiple relevant human studies or a systematic review.
• Moderate: some human data or substantial mechanistic support.
• Weak: limited or indirect evidence.
• Marketing claim: a claim with no independent evidence attached.

“Not yet assessed” means that an evidence review has not been done. We keep that separate from “marketing claim”; a missing citation is not a completed literature search.

One change, if one is useful.

The calculation identifies a gap or possible mismatch and suggests one modest change. When the key functions are present, it suggests keeping your routine. Your optional preference can refine the explanation, but a language model cannot change the calculated result.

Your labels stay yours.

With an account, your routine, extracted label text and corrections are private to you through database access policies. Label photos are sent to OpenAI for extraction and are not retained by this app. The demo stores its routine on this device only. Sharing downloads a finding card containing only your goal and a generic function finding, without product names or personal notes.

Where reference data belongs.

Ingredient identities and declared functions can be checked in the European Commission’s CosIng database. Regulatory records require the applicable current annex and product context. A database listing alone is not an approval or a finished-product assessment.