How Yasno decides

What happens between taking a photo and seeing a verdict, where the information comes from, and the rules Yasno follows so that it never claims more than the evidence supports.

Yasno is not medical advice. It describes what regulators and published research say about ingredients. It does not diagnose, treat or prescribe. For questions about your own health, allergies or medication, talk to a doctor or pharmacist.

1. Reading the label

When you take a photo, the app makes it smaller on your phone and sends it over an encrypted connection to our server. Our server asks an AI vision model to copy out the ingredient list and, if it is in the photo, the nutrition panel. The model is instructed to copy only what it can read and never to guess. If part of the label is unreadable, it says so, and so does the app. The photo is not stored.

You can also type an ingredient list instead of taking a photo.

2. Recognising each ingredient

Each ingredient is matched to Yasno's own database, which knows E-numbers, cosmetic (INCI) names and common English and Bulgarian names. An ingredient that can't be matched with confidence is shown as not recognised. It is never forced to the closest-looking name.

Some label words, such as "natural flavourings" or "parfum", don't say what is actually inside. Yasno says that plainly instead of guessing.

3. Where the ingredient information comes from

Every ingredient record has two parts.

What regulators say. This comes from official public sources: the European Commission's CosIng cosmetic ingredient database, the EU food additives database, and the US Food and Drug Administration's list of substances added to food. Whether something is permitted, restricted or banned is taken from these sources, not from an AI.
What the health evidence says. For each ingredient, an AI research step searches for published safety assessments. It is limited to authoritative sources such as the European Food Safety Authority (EFSA), the EU Scientific Committee on Consumer Safety (SCCS), the European Chemicals Agency (ECHA), the International Agency for Research on Cancer (IARC), the US FDA, the World Health Organization, and peer-reviewed systematic reviews. Blogs, campaign groups and other apps are excluded.

The findings are turned into three short sentences: what the ingredient is, what it does, and why it matters to you. Every explanation is shown with its sources.

Checks before anything is published

Automated checks run on every record. A record that fails is not published. The checks confirm:

every cited source really exists;
a caution or avoid verdict names a specific reason and rests on a regulator or strong research;
the explanation doesn't sound more certain than its source. "Could not be ruled out" must not become "causes";
the text doesn't contradict the official regulatory data.

After the automated checks, each record is reviewed against its sources. The reviewer is either our editor or an AI reviewer working to our written review rules. The record notes which one reviewed it. Only reviewed records reach the app. If you scan an ingredient we haven't assessed yet, the app tells you, and we research it in the background. It reaches the app only after it has passed the same checks and review.

When official regulatory data changes, the affected ingredient goes back for review before the change reaches the app.

4. The three verdicts

✅ Fine No meaningful concern at normal amounts. This is the default.
⚠️ Caution There is some evidence of a concern, it depends on the amount, or it is genuinely worth knowing.
🚫 Avoid There is a real, well-supported concern, such as a legal restriction, a classification by a major safety body, or strong agreement across research.

No reason, no warning. A caution or avoid verdict must state a specific reason and cite at least one regulator or strong research source. This is enforced in the database itself. A record that breaks the rule can't be saved. A verdict is about an ingredient, never about a brand or a company.

5. Personal warnings

If you choose to fill in your profile (allergies, pregnancy or breastfeeding, some health conditions, skin type, or your own list of ingredients to avoid), Yasno adds warnings that apply to you. These rules only ever add a warning or make one stronger. They never make an ingredient look safer than its general verdict. Every personal warning says why it is shown, for example "you've told us you're allergic to milk".

Allergen alerts appear above everything else on the result.

Always check the physical label. Recipes change and databases lag. Yasno points out allergens it finds in the text it read. It never says a product is free of an allergen or safe for you. Allergen words are recognised best on labels in English and Bulgarian. On labels in other languages, Yasno asks you to check the ingredients printed in bold yourself.

6. The score

Each product gets a score from 0 to 100, worked out with the simple arithmetic below. The score is the same for everyone. Personal warnings are shown separately and don't change it.

Start at 100.

For each ingredient:
  Avoid          −18
  Caution         −7
  Unknown         −2   (vague or not yet assessed)
  Fine             0

Ingredients are listed from most to least, so:
  first third of the list   × 1.3
  middle third              × 1.0
  last third                × 0.75

Food and drinks only, when the nutrition panel was photographed:
  high sugar (only when sugar is added)   −10
  high saturated fat                       −6
  high salt                                −6
  source of fibre                          +4
  source of protein                        +4   (fibre + protein at most +8)

If any ingredient is "Avoid", the score can be at most 40.
The score never goes below 0 or above 100.

Score	Band
80–100	Good
55–79	Okay
30–54	Poor
0–29	Bad

Nutrition thresholds. Neither the EU nor Bulgaria defines an official "high" level, so Yasno uses the UK Food Standards Agency's "high" (red) levels per 100 g: more than 22.5 g sugar, 5 g saturated fat or 1.5 g salt (for drinks, per 100 ml: 11.25 g, 2.5 g and 0.75 g). "Source of fibre" follows the EU definition of at least 3 g per 100 g (1.5 g per 100 ml for drinks), and "source of protein" means protein provides at least 12% of the energy, also the EU definition. Natural sugars in fruit or plain dairy are not penalised. The sugar penalty applies only when sugar is added.

If the nutrition panel wasn't photographed, the score uses ingredients only, and the app says so. Cosmetics are scored on ingredients only. A food score and a cosmetic score are not comparable.

7. What Yasno will never do

Say a product is free of an allergen, or "safe for you".
Give a warning without a specific reason and a trustworthy source.
Invent concern to look thorough. A clean label gets a short "this is fine".
Judge a brand or a company. Verdicts are about ingredients.
Use a product database. The label in your hand is the only source.
Tell you how to treat or prevent any illness.

8. Mistakes and corrections

Every result and every ingredient in the app has a "Report a problem" option. Reports go to a review queue with a copy of the result, with your personal warnings removed. We aim to look at every report within 5 business days. When a verdict changes, the change and its reason are recorded. Manufacturers can contact us directly; see Support.

9. Sources and credits

Regulatory data: European Commission, CosIng cosmetic ingredients database; European Commission, EU food additives database (Regulation (EC) No 1333/2008); EU Cosmetic Products Regulation (EC) No 1223/2009 and its annexes; US Food and Drug Administration, Substances Added to Food inventory. EU material is reused under the Commission's reuse policy (Decision 2011/833/EU). Explanations summarise, in our own words, assessments by EFSA, SCCS, ECHA, IARC, the FDA, the WHO and peer-reviewed research, each cited where it is used. None of these organisations endorses Yasno.