← All posts

Food scores explained: how an honest 0–100 actually works

2026-04-25 · 4 min read

If a food-scoring app cannot show you exactly how it computes a number, it is selling you a vibe. WellSignal is allergic to vibes.

Our score is fully deterministic. The same nutrition data plus the same user profile always produces the same score — there is no model in the loop for the core score itself. We use a model only to extract nutrition from a photo or a label, not to invent the verdict.

There are eight dimensions: nutrient density, processing level, added sugar, sodium, saturated fat, fiber, satiety, and additives. Each maps to a 0–100 sub-score, then weights are applied. Personalization adjusts the weights — never the inputs. Hypertension raises sodium's weight; PCOS or prediabetes raises added-sugar weight. GLP-1 dose-up loosens calorie penalties for low-volume meals.

Allergens are not a weight. They are a hard filter. Anything containing a flagged allergen is blocked from one-tap logging entirely. The only way to override is an explicit "I know, log it anyway" confirmation.

We refuse three things: opaque scoring systems, paid product placement, and brand sponsorships from food companies. The product is the score. The score has to be trustworthy. End.