Three scores, not one
Every analysed product carries three separate numbers, each on a 0–100 scale where higher is better. They answer different questions and are deliberately not combined into a single verdict.
- Nutrition Score — starts at 50. Whole foods, proteins, fibre and vitamins raise it; added sugars, syrups, refined fats and trans fats lower it.
- Hazard Score — starts at 100 and falls as additives with recognised safety concerns are found. A product of only whole ingredients stays at or near 100.
- Mirella Index — a food-quality read rather than a safety one: how close the product is to something you could cook at home. Reported as Gourmet (80+), Everyday (50–79) or Basic (0–49).
How a score is produced
- The ingredient list is read from the label — from a photograph or from the source record.
- Each token is matched against our additive catalogue, including the multilingual synonym layer, so «E330», «лимонена киселина» and «citric acid» all resolve to the same entry.
- Each matched ingredient carries three stored impact values, one per score.
- The scores are the sum of those impacts against each starting baseline, clamped to 0–100.
- A person reviews the result. Scores are published only after that review.
If a label cannot be read, or no ingredient has been matched yet, we show no score at all rather than a default. A dash means we do not know — it never means "average".
Where hazard levels come from
Hazard classifications reference published assessments by the European Food Safety Authority and the US Food and Drug Administration. Each additive page lists the bodies behind its classification. Allergens follow the 14 categories that Annex II of EU Regulation 1169/2011 requires to be declared.
Allergen status is tracked separately from hazard level, because they are different questions. An almond is nutritious and carries no additive hazard, and is also one of the most serious allergens on the list.
Limitations — please read
We would rather state these plainly than have you over-read a number:
- No quantity weighting. Ingredient lists give order, not amounts. A trace preservative and the main ingredient currently count the same. This is the single biggest limitation of the model.
- Presence, not dose. Regulators assess additives at defined intake levels. We flag that something is present; we cannot tell you whether the amount in this pack matters for you.
- Labels change. Manufacturers reformulate. The physical label in your hand is authoritative, not our record of it.
- Matching is imperfect. An unusual spelling can go unmatched, which shifts a score. Every product page has a report link for exactly this.
- Not medical advice. These scores are informational. They are not a clinical assessment and must not be used as one.