Processed Food Scanner Evidence: NOVA, Additives and Limits

Processed food scanner evidence is a classification problem, not a shopping trophy. Apps compress NOVA, additives and house rules into one glance. Experts do not always agree on the same yogurt. This is a source-reviewed guide, not a hands-on lab test.

A red “ultra-processed” badge at the fridge door feels like a lab result. It is a label read through a rulebook. Processed food scanner evidence depends on whether humans can apply that rulebook the same way twice.

This page sits beside the best meal planning apps hub. For claimed product features see the Ivy review and the Yuka review. Photo calorie tools such as Foodvisor answer a different question. We do not invent a scanner league table.

Do not treat a processing class as a diagnosis. Fortified foods, medical nutrition and affordable staples can sit in NOVA 4 and still have a role. If labels fuel fear, step back.

What NOVA is designed to do

Monteiro and colleagues (Public Health Nutrition, 2019) describe four groups. Group 1 is unprocessed or minimally processed food. Group 2 is culinary ingredients such as oil, sugar and salt. Group 3 is processed foods that combine those. Group 4 is ultra-processed industrial formulations, often with little intact group-1 food left.

NOVA does not score protein quality, fortification or price. A whole-grain extruded cereal and a confectionery bar can share group 4. That is a feature of the system. It is also the main consumer confusion.

SACN’s July 2023 statement, and its April 2025 rapid update, reviewed eight processing classifications. Only NOVA met the committee’s initial screening criteria for possible UK use. SACN still worried about practical application. Some NOVA assignments clash with nutrient-based ideas of a “healthier” food. The 2025 update said higher (ultra) processed intake still looks concerning. It also said the evidence cannot yet separate processing itself from the fact that many of these foods are high in energy, saturated fat, salt or free sugars.

What processed food scanner evidence actually shows

Reliability is the first evidence question. If trained people disagree, a phone will disagree too.

Braesco’s team asked French food and nutrition specialists to assign NOVA groups. One list had 120 marketed products with ingredients (159 evaluators). One list had 111 generic foods without ingredients (177 evaluators). Fleiss’ κ was 0.32 and 0.34. That is poor agreement. A cluster of marketed foods was mostly called NOVA 4. Other clusters were mixed. The authors said current criteria do not allow robust, functional assignment.

Bleiweiss-Sande and colleagues (Nutrients, 2019) coded the top 100 foods eaten by US children in NHANES 2013–2014. Inter-rater rank correlation was 0.97 for the UNC system, 0.78 for IFIC and 0.76 for NOVA. Sodium and added sugars helped predict “highly processed” classes. The authors said the systems may not flag nutrient-dense processed foods children actually eat.

Trained double-coding can look better. A 2023 American Journal of Clinical Nutrition analysis of preschooler recalls reported Cohen’s κ 0.75 after certified coder pairs used NDSR. That is a research workflow, not a shopper tapping a barcode.

Hall and colleagues (Cell Metabolism, 2019; NIH news release) ran an inpatient crossover in 20 adults. An ultra-processed menu and a minimally processed menu were matched for presented nutrients. People ate more energy on the ultra-processed menu and gained weight in two weeks. That trial supports a mechanism story. It does not validate a consumer scanner.

We found no peer-reviewed paper that takes Ivy’s house rating or Yuka’s 100-point mark and compares them with expert NOVA coding on a stated product set. Ivy’s terms (captured 13 September 2026) say the app uses Open Food Facts and manufacturer labels, and that the publisher does not analyse composition in a lab. Yuka’s help page is a Nutri-Score-plus-additives-plus-organic mix, not a NOVA engine.

Additives are not a personal risk colour

Many scanners paint E-numbers. Yuka assigns green-to-red risk and can cap a product at 49/100 if it marks an additive high-risk. That is a house reading of EFSA, IARC and selected papers. It is not a personal exposure assessment.

An approved additive has a permitted use. An IARC hazard classification is not the same as a dietary risk at the dose in one yogurt. Observational links from one cohort should not become a scarlet letter on every emulsifier.

Ivy’s public copy lists seed oils, pesticides, microplastics and heavy metals among score inputs. Those are different scientific questions. A barcode cannot measure microplastics in the pack you hold unless a lab result is attached. If the method page does not show a lab, treat the extra marks as inferred, not assayed.

What a scanner cannot see

Your portion. A class is per product, not per gram you eat.

Your week. One NOVA 4 bread does not describe a pattern. SACN’s caution is about habitual diets.

Home recipes. A photo of leftovers is not a barcode. Processing class from an image is still experimental. We do not treat it as settled.

Nutrient needs. Protein after illness, iodine in pregnancy, or a low-potassium clinical diet can require products a scanner paints red. Our everyday protein intake guide stays in food language for that reason.

How to use a scanner without a purity test

Use it to notice formulation. Long additive lists and sweetened drinks are fair flags. Then read the Eatwell plate for the week. Swap one daily sugary drink. Add a bag of frozen vegetables beside a ready meal. That is a pattern change. It is not a ban.

Compare two similar products. Two baked beans. Two yogurts. If both are NOVA 4, the lower sugar and salt pack still matters. Nutrient profile and processing are complementary, as Sarda, Julia and colleagues have argued for Nutri-Score updates.

Ignore health promises on the store listing. Ivy’s Play copy (captured 13 September 2026) claims users lose weight and fix anxiety. Those are marketing sentences. They are not trial results. We omit them as facts.

What remains unverified

We cannot verify 2026 Ivy or Yuka accuracy against a gold-standard NOVA panel. We cannot say a red scan causes disease. We cannot convert a week of green scans into a health outcome.

Processed food scanner evidence is strongest at the level of systems (NOVA reliability, SACN caution, nutrient-score mismatch). It is weakest at the level of one branded app. Keep those layers separate.