← All food signals
Governance

The Ultra-Processed Line

Machine-learning classifiers already grade the food supply by degree of processing, scoring well against the NOVA research scheme. California has written the ultra-processed line into statute for school meals.

What is it?

Machine-learning models already sort food by how processed it is, scoring above 0.95 against the NOVA scheme. Governments are still drawing the legal line. The grade reached the shelf before the definition did, and California’s school-meal statute is the only line written in law so far.

Why now?

In August 2026 HHS sent a white paper defining ultra-processed food for White House review, contents undisclosed. A 2026 review counted nine AI studies on the task. Courts draw lines too: injury claims against makers fail on causation, and one maker is suing a scoring app.

Who’s building it?

Research teams behind FoodProX, scoring apps such as Yuka, California lawmakers, and federal health agencies.

FoodProXYukaHHSCalifornia legislatureNOVA researchers
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — research classifiers, scoring apps and state legislators.
Buyers — shoppers reading app scores, school caterers, and makers whose products get graded.
Gatekeepers — federal health agencies, the White House review office, and courts.
What’s at stake
Whether the legal line matches the one NOVA-trained models already draw, or leaves graded products stranded.
Whether a single app score survives legal challenge.
Whether school meals set the template for everyone.
What’s unsettled
What the federal white paper actually says about ultra-processed food.
Whether any regulator will ever accept a model’s classification.
Whether the Third Circuit keeps causation as the barrier in injury claims against makers.
02 · How the demand matureda progression, not a timeline
Novelty
Researchers show a model can grade processing from nutrients alone.
Retention
Scoring apps put a number on the shelf, product by product.
Willingness to pay
Schools, agencies and courts need one definition that holds, not scores varying by app.
Category
Processing becomes a regulated attribute, checked against law rather than a research scheme.
03 · Where the market goesthe forks, not a prediction
→The law follows NOVA: existing classifiers become compliance tools overnight, and the models that graded the shelf early turn into the regulator’s instrument.
→The law departs from NOVA, and every trained classifier needs a new benchmark before its grades count.
→The white paper stays guidance, and courts and apps keep drawing the working line between them.
→California’s school-meal statute spreads state by state into a patchwork of definitions.
The forks turn on whether the definition is written as ingredient and process rules a machine can apply, or as a judgement only people can make.
04 · The Graded Shelf deskadjacent market, moving now
The Readable Label
Scanning apps grade products at scale.
Yuka reached 85 million reported users worldwide by August 2026.
BBC News via The Star
The Clean Purge
Undefined label terms like ‘clean’ meet the courts.
A class action over Ulta’s ‘clean’ claims advanced past dismissal.
Cosmetics & Toiletries
Machine Dietetics
Models that judge food are now scored themselves.
AI models significantly overestimated fat across 15 weighed hospital meals.
Nutrients
05 · The Wire
01
HHS sent a white paper on ultra-processed food definitions.
Haynes Boone ↗
02
Several of nine AI studies report F1 scores of 0·86 to 0·98
British Journal of Nutrition ↗
03
A federal judge refused leave to amend in Martinez v. Kraft Heinz on 30 June 2026, finding the complaint deficient on causation.
FDLI ↗
04
Yuka scores foods from 0 to 100 on nutrition, additives and organic status, and declines makers’ requests for help with their scores.
FoodNavigator ↗
05
California’s AB 1264 gives the first statutory definition of ultra-processed food and phases the most concerning products out of school meals.
Office of the Governor of California ↗
06
Model puts over 73% of the US food supply as ultra-processed
Nature Communications ↗
07
Goya Foods filed a Lanham Act case against Yuca Corp. in the US District Court for New Jersey.
GovInfo (US District Court, D.N.J.) ↗
The panel on this signal
They agreed on
–Classifiers and apps already grade processing at scale, so any legal definition lands on a shelf that is pre-sorted.
–Every AI study benchmarks against NOVA, not law.
–Causation, not the missing definition, is what has stopped injury claims against makers so far.
The food-process engineer disagreed on
–Nutrient data cannot show how a food was made; a model reading labels infers processing, it never observes it.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.