A sugary breakfast cereal can show a longer vitamin list than a bag of spinach. A candy-adjacent snack bar can claim more fiber than oatmeal, courtesy of chicory-root inulin added at the factory. Grade food on any single number and the industry will find the loophole — it has had decades of practice.
That is why Nutrication never grades on one axis. Every food you scan runs through six independent frameworks — processing level, overall nutrient balance, ingredient quality, micronutrient density, blood-sugar impact, and fat composition. Each comes from published nutrition science, each earns points on its own scale, and the points combine into a single 0–100 score that maps to a letter: A, B, C, or D.
This page is the complete methodology — the weights, the math, the edge cases, and the rule we consider most important: when the data isn't good enough to judge, the app says "not rated" instead of guessing.
Six measures, one score
Each measure answers a different question and carries its own weight:
| Measure | Max points | The question it answers |
|---|---|---|
| NOVA processing | 35 | How industrially processed is this food? |
| Nutri-Score | 30 | How does its overall nutrient profile balance out? |
| Ingredient flags | 20 | Does the label contain additives worth knowing about? |
| Nutrient density | 20 | What beneficial micronutrients does it actively deliver? |
| Glycemic load | 15 | How hard will a serving hit your blood sugar? |
| Fat quality | 15 | Is its fat mostly unsaturated or mostly saturated? |
Those maximums total 135, but the engine never divides by 135. It sums the points a food earned across the measures that could actually be computed, divides by the maximum possible for those measures, and scales the result to 100. That distinction matters — more on it below.
What each measure looks at
NOVA processing (35 points). The NOVA classification, developed at the University of São Paulo and used in research by the World Health Organization, sorts foods into four groups by how they are made. Unprocessed and minimally processed foods earn the full 35 points; culinary ingredients like oil and butter earn 28; processed foods earn 18; ultra-processed foods earn just 4. That last cliff is deliberate — ultra-processing is the strongest single signal in the grade.
Nutri-Score (30 points). The front-of-pack system officially adopted in France, Germany, Spain, and other European countries. It penalizes calories, saturated fat, sugar, and sodium per 100g, and credits fiber, protein, and fruit or vegetable content. Its A-to-E letter converts to 30, 24, 16, 8, or 2 points. We apply two corrections: naturally occurring sugar in a whole food — an apple, plain yogurt — takes no sugar penalty, while industrially added fibers like inulin and chicory root fiber earn only half the fiber credit. Both loopholes close in the honest direction. Beverages are held to much stricter energy and sugar thresholds; among drinks, only water can earn an A.
Ingredient flags (20 points). Every food starts at 20 and loses points for flagged ingredients: partially hydrogenated oils cost 10, sodium nitrite costs 6, high-fructose corn syrup 5, generic artificial colors 2. Credible certifications add points back — organic and whole grain each add 3. A minor cosmetic additive sitting near the bottom of a long ingredient list counts at half weight, since ingredient lists are ordered by amount. Serious flags never get that discount.
Nutrient density (20 points). The only measure that rewards what a food actively gives you. It tracks six micronutrients — potassium, calcium, iron, magnesium, zinc, and vitamin C — against FDA Daily Values, capping any single nutrient at 50% of its Daily Value so one mega-dose can't max the score. Ultra-processed foods get half credit here, because their micronutrients are usually synthetic fortification. A vitamin-fortified sugary cereal should not read as nutrient-dense as spinach.
Glycemic load (15 points). Estimates how much a realistic serving raises blood sugar, using glycemic load rather than glycemic index. Databases don't publish glycemic index, so the app estimates it from the food's dominant carbohydrate category, then computes the load from a realistic serving size. A load of 10 or under earns the full 15 points; 11–19 earns 8; 20 or more earns 2. Foods with a genuinely fiber-rich carb profile pick up a 3-point bonus.
Fat quality (15 points). Nutri-Score sees how much saturated fat a food has; this measure sees composition — what share of the total fat is unsaturated. Seventy-five percent or more unsaturated earns the full 15; a mostly saturated profile earns 2. Trans fat overrides everything: more than a trace of it, or the words "partially hydrogenated" anywhere in the ingredients, caps this measure at a single point. Foods with under 3g of fat per 100g skip the measure entirely — the fat profile of an apple is meaningless.
"Not rated" beats a wrong guess
Every measure has an availability condition. NOVA needs an ingredient list or a database processing tag. Nutri-Score refuses to run when saturated fat, sugar, and sodium are all missing, because treating them as zero would hand out fake "low sodium" credit. Nutrient density needs at least two micronutrients on file. Fat quality needs both total and saturated fat reported.
When a measure can't be computed, it is excluded from the grade and the remaining measures re-normalize to 100. Nothing is zero-filled, and no fabricated "neutral" value stands in. A measure counts only when it was computed from real data.
Sparse data still has guardrails. A grade resting on a single measure is capped at C, whatever the score. A grade resting on two measures is capped at B — and capped at C if the food is processed or ultra-processed, so a clean-looking label can't float a data-poor packaged food up to a B. A genuine whole food like plain frozen spinach faces only the gentler B cap: missing paperwork alone never drags it down further. With three or more measures, no cap applies. Every result also carries a confidence tag (high, medium, or low) based on how many of the core signals were available.
The consensus safety net
Algorithms have blind spots. Systems built around penalizing fat tend to underrate foods that are nearly all fat yet backed by strong outcome evidence. Nutrication carries a curated list of eleven overrides for exactly these cases: extra virgin olive oil, whole avocado, avocado oil, salmon, sardines, mackerel, herring and anchovies, walnuts, almonds, pistachios, and chia or flax seeds — foods that major cardiology guidelines recommend and landmark trials such as PREDIMED support. Each is pinned to an A, with scores between 86 and 92.
Three rules keep this honest. First, an override is a floor, not a ceiling: it can only raise a score, never lower one. Second, matching is strict — containing a food is not being that food. Almond milk is not almonds, a fish spread is not plain fish, and in a tin of sardines the fish must be the first ingredient. Third, there is a refusal list. Butter, coconut oil, dark chocolate, honey, and red wine all have devoted advocates, but the evidence is contested, and overriding the algorithm there would be closer to advocacy than to science. They take whatever grade the six measures produce.
What the letters mean
- A (80–100): An excellent choice — typically a whole or minimally processed food with a strong nutrient profile. Eat it freely.
- B (65–79): A good everyday choice, usually with one thing worth keeping an eye on — sodium, say, or a moderate glycemic load.
- C (45–64): Middle of the road. Fine sometimes, not an everyday staple.
- D (0–44): Best kept to an occasional treat — and the breakdown will name the main reason why.
One small detail we sweat: the score is rounded before the letter is assigned, so the number you see always matches the letter you see. A 44.7 never displays as a "45" wearing a D.
How Nutrication uses this
When you scan a product, the app pulls its data, runs all six measures, and shows the grade with a per-measure breakdown. Each measure explains in plain language why it scored what it scored, and any measure that couldn't run says "not rated" rather than pretending. If a consensus override lifted the grade, the app discloses the original algorithmic score alongside it. Everything then distills into a short summary with up to four "good" and four "watch" bullets, leading with whatever mattered most.
Meal photos get special handling. Glycemic load is computed per item and summed — the textbook mixed-meal method — so the crackers on your plate don't set the glycemic index for the chicken. And because a meal's "ingredients" are just item names, the ingredient-flags measure bows out rather than award a perfect score to an ingredient list that doesn't exist.
What to actually do
- Treat the letter as a starting point, then tap the breakdown to see which measure moved it.
- Check the confidence tag — an A built on rich data means more than one built on two measures.
- Comparing two similar products? The breakdown shows exactly where they diverge — usually processing or sugar.
- Don't stress over a single D. The grade describes a food, not your diet; patterns matter more than any one item.
- "Not rated" means unknown, not bad — try another brand, or the whole-food version.