Chapter 22 — ★ Ultra-Processed Food: What the Ward Trial Showed, and What Part IV Was Actually About

🚪 Threshold concept: processing is not a spectrum of badness. ⚠️ The question is not how far a food has been altered from nature. It is what industrial formulation does to how much of it you eat.

The Hook: One ingredients list, five arguments

A packaged chocolate-hazelnut breakfast biscuit. Four in a sleeve, 230 kcal each.

Here is its ingredients list, and here is every argument Part IV has had.

Five factions. Five theories. One biscuit.

⚠️ And here is the thing that should have been obvious four chapters ago: every one of those eliminations removes this biscuit.

Cut sugar — the biscuit goes. Cut seed oils — the biscuit goes. Buy only organic — the biscuit goes. Avoid additives — the biscuit goes. Fast until noon — the biscuit goes.

Five different theories, five different mechanisms, five different villains — and one shared result.

⚠️ Nobody in Part IV was wrong about what to do. Almost everybody was wrong about why.

This chapter is about the thing they were all pointing at.

🏃 Fast Track: §22.2 (the threshold), §22.3 (the ward trial), §22.9 (⚠️ the answer), §22.13 (what to do). Thirty-five minutes.

🔬 Deep Dive: §22.5 (mechanisms), §22.7 (⚠️ NOVA's real weaknesses, argued properly), §22.10 (additives — where the open questions genuinely are), §22.12 (⚠️ what not to conclude).

22.1 What NOVA actually classifies

The classification comes from Carlos Monteiro's group at the University of São Paulo, and its central move is unusual: ⚠️ it does not classify foods by their nutrients.

What marks out Group 4, specifically:

• Ingredients you would never have in a kitchen — hydrolysed protein isolates, modified starches, invert sugar, hydrogenated or interesterified oils, protein isolates
• ⚠️ Cosmetic additives — colours, flavours, flavour enhancers, emulsifiers, non-sugar sweeteners, thickeners, bulking and firming agents
• Industrial processes with no domestic equivalent — extrusion, moulding, pre-frying
• And a commercial purpose: ⚠️ formulations designed to be convenient, durable, highly palatable, and profitable

💡 ⚠️ Notice what the classification is actually tracking: not nutrients, not "naturalness," but whether a food is an INDUSTRIAL FORMULATION rather than a food.

This is why NOVA irritates nutrition scientists. It cuts across every category they use. A supermarket wholemeal loaf with emulsifiers is Group 4; home-baked white bread is not. Sugar itself is Group 2, not Group 4.

⚠️ Hold that irritation — §22.7 takes it seriously, because it is the strongest objection to the whole framework and it deserves better than the dismissal it usually gets.

22.2 🚪 The threshold: processing is not a spectrum of badness

Almost everyone reads NOVA as a purity ladder — Group 1 good, Group 4 bad, and the badness increasing with distance from nature.

⚠️ That reading is wrong, and it will make the rest of this chapter unintelligible.

Three facts that break the purity reading:

1. ⚠️ Group 3 contains foods with excellent outcome evidence. Tinned fish. Tinned beans. Frozen vegetables. Wholegrain bread made with four ingredients. Processing here is preservation — it makes good food cheap, safe, shelf-stable and available in February.

2. Processing per se has been overwhelmingly good for human health. ⚠️ Pasteurization. Canning. Refrigeration. Fortification — iodized salt (Chapter 14), folic acid in flour (Chapter 13), vitamin D in milk. Freezing. These are among the most successful public health interventions ever implemented, and every one of them is processing.

3. ⚠️ And Group 2 — oil, butter, sugar, salt — is not Group 4. A cake you baked yourself from flour, butter, sugar and eggs is not ultra-processed, however much sugar is in it. NOVA is explicitly not making a claim about its nutritional quality.

So what is the threshold?

⚠️ **The question is not how far a food has been altered. It is what industrial formulation does

to how much of it you eat.**

NOVA is not a purity scale. It is a proxy for a specific commercial engineering problem: how do you make a food that is cheap, durable, transportable, and consumed in large quantities?

And the answer that industry converged on — ⚠️ energy-dense, soft, fast to eat, engineered for palatability, cheap per calorie, portioned generously — turns out to describe food that people eat more of.

That's the claim. It is a claim about intake, not about toxicity — and §22.3 is where it was tested.

How the food supply got here

Brief, because it explains why Group 4 dominates and why "just cook" underestimates the problem.

None of this was designed to make people fat. ⚠️ It was designed to solve real problems, and it solved them.

💡 ⚠️ Notice that each step was a solution, and that the properties which made each solution work are the same properties §22.5 identifies as driving intake.

Shelf-stable means low water and high energy density. Convenient means fast to prepare and fast to eat. Consistent and appealing means engineered texture and palatability. Cheap means refined starch, sugar and oil.

⚠️ There is no villain in this history, which is the same finding as Chapter 16's biotin case and Chapter 1's pipeline. A system optimized hard for cost, safety, shelf life and convenience produced food that people eat more of — not because anyone set out to, but because nothing in the optimization was pushing the other way.

Why this is a threshold rather than a fact

Because holding it changes what question you ask about a food.

⚠️ Not "what's in it?" — which is the question Chapters 18, 19 and 20 were all arguing about, and which produced five different villains in one biscuit.

⚠️ But "what does this food do to how much I eat, and why was it built that way?"

And it dissolves a set of arguments that cannot be resolved on their own terms. Is the sugar the problem or the fat? Is it the emulsifier or the refined flour? ⚠️ These are questions about components, asked of something engineered as a system.

🪞 Learning Check-In: predict the trial

Stop here. You have everything you need to reason about this.

Here is the design, and not the result.

Twenty adults live on a metabolic research ward for four weeks. They cannot leave. All food is provided and precisely measured.

Two weeks on an ultra-processed diet. Two weeks on an unprocessed diet. Randomized order, crossover — each person does both, so each person is their own control.

⚠️ The two diets are MATCHED for presented calories, energy density, macronutrients, sugar, sodium and fiber.

Participants are told to eat as much or as little as they want. Nothing is restricted.

Intake is measured to the gram.

Write down your predictions before reading on:

1. ⚠️ Will they eat a different number of calories on the two diets? If so, which way, and how much? Commit to a number.

2. What happens to body weight?

3. ⚠️ Name three mechanisms that could produce a difference, given that the diets are matched for everything in question 1's usual answers.

4. What are the three biggest weaknesses of this design? ⚠️ Be specific — this matters for §22.4.

5. And the honest one: ⚠️ before you read the result, what would you have said the answer was likely to be? If your answer is "no difference — calories are calories," write that down too. It was a defensible prediction.

Keep your answers. §22.16 asks you to compare.

22.3 The ward trial

Hall KD et al., Cell Metabolism, 2019. ⚠️ This is, in my view, the strongest single piece of evidence in this book, and I want to be precise about why.

The design, as above: 20 adults, inpatient at the NIH's metabolic clinical research unit, randomized crossover, two weeks per arm, all food provided, ad libitum intake measured to the gram, diets matched for presented calories, energy density, macronutrients, sugar, sodium and fiber.

The result:

⚠️ On the ultra-processed diet, participants ate approximately 500 kcal per day more.

They gained roughly 0.9 kg over two weeks on the ultra-processed diet.

They lost roughly 0.9 kg over two weeks on the unprocessed diet.

Nobody was asked to eat more or less. Nobody was restricted. The nutrients were matched.

And a mechanistic observation from the same study: ⚠️ participants ate FASTER on the ultra-processed diet — more calories per minute.

What a day on each arm actually looked like

Worth seeing, because "ultra-processed" is abstract and the trial's food was not. ⚠️ These are representative of the diets' construction rather than a transcript of any particular day.

⚠️ Both arms were matched for presented calories, energy density, macronutrients, sugar, sodium and fiber. Neither is a caricature. The ultra-processed arm is not obviously "junk" and the unprocessed arm is not conspicuously virtuous — which is precisely what makes the 500-calorie difference interesting.

💡 ⚠️ Look at the two lunches and ask yourself which one you'd finish faster.

Canned ravioli, soft white bread and a packaged snack take a fraction of the chewing that a salad with chicken, apple slices and seeds does. You could eat that lunch in six minutes. The other one is physically hard to eat in under fifteen.

§22.5's mechanisms are visible in the menu, and they are not chemical.

22.4 Why the trial is unusually strong — and what it doesn't establish

Everything Chapter 2 said was impossible in nutrition research, this study did.

⚠️ This is close to the maximum achievable rigour for a dietary question, and it produced a large, unambiguous effect.

Now the limitations, and I'd rather state them than have you find them:

1. ⚠️ n = 20. Small. (Though the crossover design and the effect size make it much more informative than 20 people in a parallel-group study.)

2. ⚠️ Two weeks per arm. Short. It tells you what happens to intake immediately; it does not tell you what happens over years — and Chapter 21's compensation curve is a reason to be cautious about extrapolating short-run intake findings.

3. Not blinded, and couldn't be. ⚠️ Participants could see what they were eating. Expectation effects are possible — though it's worth asking what a participant's expectation would have been, since the popular assumption is usually that "healthy food" is what you eat less of.

4. ⚠️ Fiber was matched by adding fiber supplements to beverages on the ultra-processed arm — which is a legitimate design choice for matching, and which means the arms weren't identical in how fiber was delivered (Chapter 11 §11.6's point about supplements versus food).

5. Inpatient. The food environment was controlled in a way no real life is.

⚠️ What the trial establishes: ultra-processed diets cause greater ad libitum energy intake than matched minimally-processed diets, over two weeks, in this population.

What it does NOT establish: that UPF causes disease independent of energy intake · that all Group 4 foods behave the same way · that the effect persists over years · or which specific feature of the food produced it.

⚠️ That last one is §22.5, and it is genuinely open.

And a word about the author, because it matters for how you should read this. ⚠️ Kevin Hall has been unusually careful in public about not over-claiming from his own study, has published work that complicated his earlier findings, and has been explicit about what the trial doesn't show. Chapter 1 §1.5's test — does this source ever say "I don't know" or "I was wrong"? — is passed here more clearly than by almost anyone else in Part IV.

22.5 Why do people eat more of it?

The mechanisms, ranked by how well supported they are. ⚠️ This matters because it's the difference between a category and an explanation.

1. 🟢 Energy density. ⚠️ UPF is more energy dense — more calories per gram — and energy density is one of the better-established drivers of intake. People tend to eat a fairly consistent weight of food. More calories per gram means more calories. (The Hall trial matched presented energy density, which makes this less able to explain that particular result — but it matters enormously in the real world.)

2. 🟢 Eating rate. ⚠️ UPF is softer, requires less chewing, and is eaten faster — the Hall trial measured this directly. Faster eating means more consumed before satiety signalling catches up (Chapter 3 §3.6), which takes fifteen to twenty minutes regardless of how fast you go.

3. 🟢 Texture and oral processing. Hardness and chewiness slow you down. ⚠️ A great deal of industrial formulation is explicitly about texture, and softness is commercially desirable for reasons that have nothing to do with intake.

4. 🟡 Hyper-palatability. ⚠️ Specific combinations — fat and sugar, fat and salt, fat and refined carbohydrate — occur rarely in whole foods and constantly in formulated ones. This is a real and active area, and the definitions are still being worked out. (Chapter 18 §18.11's finding lands here: nobody binges on sugar cubes.)

5. 🟡 Weakened satiety signalling. The food matrix affects gastric emptying, hormone release and the sensation of fullness (Chapters 3, 7, 11). ⚠️ Chapter 18's liquid-calorie finding is the extreme case of this, and softness sits on the same continuum.

6. 🟡 Portion and packaging. ⚠️ Serving sizes are a design decision. Chapter 33 handles this.

7. ⚗️ Additives acting on appetite or the microbiome. The most speculative and the most discussed — §22.10.

⚠️ And the honest complication, from the same researcher

Hall's group has continued the work, and the follow-up has made the picture less tidy rather than more.

Subsequent controlled feeding studies have manipulated candidate features separately — energy density, and the degree to which foods are "hyper-palatable" — and found that these account for a substantial part of the intake difference on their own.

⚠️ Which raises the objection §22.7 makes formally: if energy density and palatability explain most of it, is "ultra-processed" adding anything beyond a label for foods that are energy-dense and palatable?

⚠️ I want to flag the intellectual behaviour here, separately from the science.

The researcher whose trial produced the headline finding has published work that complicates his own result, and has said so publicly. Chapter 1 §1.5 asked whether a source ever says "I don't know" or "the picture is more complicated than my last paper suggested."

⚠️ This is what that looks like, and it is rare enough in this field to be worth pointing at directly — including as a contrast with the certainty available in every other chapter of Part IV.

💡 ⚠️ Notice that the top three are boring, physical, and have nothing to do with chemicals.

Energy density, softness, speed. You could make an ultra-processed food that was hard, dry and slow to eat, and the effect would probably shrink. Nobody does, because those are commercially bad properties.

This is why the chapter's threshold is about formulation rather than about ingredients — and why Part IV's five factions, all of whom were arguing about ingredients, all missed it.

22.6 What the population evidence says

Large cohorts across many countries have found ⚠️ dose-response associations between ultra-processed food intake and adverse outcomes — all-cause mortality, cardiovascular disease, type 2 diabetes, obesity, and in some analyses depression and colorectal cancer.

The associations are reasonably consistent, generally graded, and survive adjustment for the obvious confounders.

⚠️ And they are observational, with everything Chapter 2 said about that. UPF intake tracks income, education, food environment, time poverty, cooking facilities, smoking, physical activity and healthcare access. Chapter 20 §20.7's confounding, running in the opposite direction.

🔬 Verdict: 🟢 Probably true that higher UPF intake is associated with adverse outcomes.

⚠️ The reason this rises above the usual observational fog is convergence (Chapter 2 §2.7): a randomized inpatient trial demonstrating a large intake effect, plus plausible physical mechanisms, plus consistent graded cohort associations. Three methods with non-overlapping weaknesses, pointing the same way.

That's the same structure that got Chapter 18's sugar-sweetened beverages to ✅ — and it is why this is the strongest thing in Part IV.

22.7 ⚠️ NOVA's real weaknesses

Argued properly, because they're serious and the enthusiastic version of this chapter would skip them.

1. ⚠️ The classification is not always reliable between coders. Studies asking multiple trained raters to classify the same foods have found meaningful disagreement. A framework whose central variable can't be measured consistently has a real problem, and this is the objection I find hardest to answer.

2. The wholemeal bread problem. ⚠️ A supermarket wholemeal loaf containing an emulsifier is Group 4. A home-baked white loaf is not. Nobody thinks the first is worse for you than the second, and the classification says something that sounds like it does.

3. It mixes criteria. ⚠️ Group 4 membership can be triggered by an industrial process, by the presence of a cosmetic additive, or by the food type — three different things bundled under one label. Chapter 20's category problem, arriving in a framework this chapter is defending.

4. It's not obviously actionable in nutrition terms. ⚠️ "Eat less UPF" doesn't tell a dietitian what to change, and guidelines are built on nutrients because nutrients are what interventions can target.

5. Some Group 4 foods are unambiguously good. ⚠️ Infant formula is ultra-processed and is life-sustaining. Fortified foods that eliminated deficiency diseases are ultra-processed. Plant-based meat alternatives are Group 4 and their health profile is genuinely unresolved rather than obviously bad. Some medical nutrition products are Group 4 and keep people alive.

6. And there's a real risk it's "junk food" with a research grant. ⚠️ If Group 4 turns out to be a proxy for energy density, sugar, salt and refined starch, then it's a new name for an old finding — and the classification adds vocabulary rather than knowledge.

Mike Gibney and others have published serious critiques along these lines, and ⚠️ they should be read rather than characterized.

22.8 Why the category earns its keep anyway

Three replies, and I hold them with different strengths.

1. ⚠️ The Hall trial matched the nutrients and got a 500 kcal difference. That is the direct answer to objection 6. If Group 4 were only a proxy for sugar, fat, salt and energy density, matching those should have abolished the effect. It didn't. Something about the formulation, independent of the nutrient profile, changed how much people ate.

2. The nutrient framework had its chance. ⚠️ Fifty years of nutrient-based guidance produced the low-fat era (Chapter 9), the fat-free reformulation disaster, and Chapter 13's supplement graveyard. "Reduce saturated fat" was followed, and the food industry reformulated around it. A framework that describes what a food IS rather than what it contains is harder to game — ⚠️ which is a genuine advantage and also, precisely, why industry dislikes it.

3. And a classification does not have to be perfect to be useful. ⚠️ The edge cases — wholemeal bread, infant formula, tinned soup — are where the argument happens, and they are not where most people's intake is. In countries where UPF is 50%+ of energy intake, the bulk of that is not contested infant formula. It is soft drinks, snacks, confectionery, packaged baked goods and ready meals, which every framework agrees about.

⚠️ My honest position: the classification is imperfect, the reliability problem is real, the edge cases are genuinely awkward — and the ward trial means something is there that the nutrient framework was not capturing.

A better classification will probably replace NOVA. The finding it surfaced will survive it.

22.9 ⚠️ What Part IV was actually about

Chapter 20's Exercise M5 asked you to write down what these public arguments are actually about. Chapter 21's Exercise S6 asked again. Go and get what you wrote.

Here is my answer, in two parts.

Part one: they were all pointing at the same food

Look back at the biscuit.

Every faction in Part IV identified a real component of ultra-processed food and built a theory around it:

⚠️ Each elimination works. Each explanation is wrong. And they are wrong in the same way.

Chapter 19 §19.11 named it: you cannot remove seed oils without removing eight other things. Chapter 17's eighth shape named the general form: count the changes.

💡 ⚠️ The reason five incompatible theories all produced results is that they were all interventions on the same variable — and none of them knew it.

This is why the debates never resolved. Sugar people and seed oil people were arguing about which ingredient in the biscuit was responsible, when the answer was the biscuit.

Part two: and there's a harder reason they took the shape they did

⚠️ This part is less comfortable and I think it's the more important one.

A villain you can eliminate by purchase is actionable. A food environment is not.

Every Part IV faction converted a systemic problem into a shopping decision, ⚠️ and that conversion is what made each of them popular. Chapter 20 §20.13's binary — a category boundary lets you finish — applied at the scale of a whole argument.

And it explains the commercial pattern this book keeps finding. ⚠️ Every chapter in Part IV had a premium product attached — cane-sugar soda, beef tallow, organic produce, a fasting kit — and every one of them was a way of buying your way out of a problem that isn't purchasable.

⚠️ Notice that this cuts against this chapter too. "Eat less UPF" can be sold as a shopping rule, and it is being sold as one — there are UPF-free snack subscriptions now. §22.12 is about why that's a mistake.

22.10 Additives, emulsifiers, and where the real open questions are

The part that gets the most attention and has the least evidence — which does not mean none.

Emulsifiers — compounds like carboxymethylcellulose and polysorbate-80 — are used to keep formulations stable. Rodent work by Chassaing, Gewirtz and colleagues found that these compounds altered gut microbiota composition and promoted intestinal inflammation and metabolic changes in mice.

⚠️ And unusually for this literature, there is a small human study. A randomized controlled feeding trial of carboxymethylcellulose (Chassaing et al., Gastroenterology, 2022) found changes in gut microbiota composition and metabolite profiles, in a small number of participants over a short period.

🔬 Verdict: ⚗️ Untested to 🟡, depending on the specific claim.

⚠️ "Emulsifiers alter the gut microbiome in humans" — there is now preliminary human evidence. "Emulsifiers cause disease in humans at dietary exposures" — untested.

This is the honest state of it, and Chapter 27 goes into the microbiome properly.

⚠️ And the reason I'm not more excited: §22.5's top three mechanisms — energy density, softness, eating rate — are far better supported and account for the ward trial's effect more parsimoniously. The additive story is more interesting and less established, which is a combination that reliably produces over-claiming (Chapter 21 §21.6).

On non-sugar sweeteners: Chapter 18 §18.13. 🟡, and the same caution applies.

22.10b Reformulation, and why the framework matters practically

⚠️ This is the argument for NOVA that has nothing to do with biology, and it may be the strongest one.

Chapter 18 Case Study 2 showed what happened when the UK taxed soft drinks by sugar content: manufacturers reformulated below the threshold. ⚠️ That worked, and it is the clearest policy success in this book.

Now consider what a nutrient-based rule does more generally.

💡 ⚠️ Every nutrient-based rule has been met by reformulation that satisfied the rule and changed little else.

This is not a conspiracy. It is what a rule does when it specifies a measurable target and a profitable industry is asked to hit it. The system optimizes for the metric.

A framework that classifies by what a food IS rather than what it CONTAINS is much harder to reformulate around — ⚠️ because the only way to leave Group 4 is to stop being an industrial formulation.

That is a genuine advantage, and it is also exactly why the framework is unpopular with manufacturers — which is a claim about incentives that you can check rather than take on trust.

⚠️ The counter-argument is real and §22.7's fourth objection makes it: a category that can't be reformulated around also can't be acted on incrementally. "Make this product less ultra-processed" has no first step, where "take 20% of the sugar out" does.

Which is the honest tension: the framework is harder to game and harder to use.

22.11 ⚠️ The problem this chapter can't solve

Ultra-processed food is cheap, shelf-stable, requires no equipment, no skill and no time, and it doesn't spoil.

Those are not incidental properties. They are the reasons it exists and the reasons it dominates the diets of people with the least money and the least time.

Marisol Ortiz's $180 a week (Chapter 20) buys substantially more calories as Group 4 than as Group 1. ⚠️ That's not a failure of her knowledge or her values. It's arithmetic.

What cooking from scratch actually requires:

⚠️ Any version of this chapter that ends at "cook from scratch" has described a solution available to people who already have one.

Chapter 32 handles the economics and Chapter 36 the food system. ⚠️ But it needs saying here, because this is the chapter where the temptation to moralize is strongest.

22.12 ⚠️ What not to conclude

Five things, and I'd hold all five firmly.

1. ⚠️ UPF is not a moral category. Chapter 20 §20.13's clean-eating framing is available here and must be refused. "Ultra-processed" describes how a food was made. It does not describe the person eating it.

2. This is not "everything in a packet is poison." ⚠️ Tinned beans, frozen vegetables, tinned fish, plain yoghurt, wholemeal pasta, bagged salad, frozen fruit — Group 1 and Group 3, cheap, excellent, and the practical backbone of §22.13.

3. ⚠️ Nobody needs to reach zero. There is no evidence for a threshold, no reason to think one exists, and the dose-response associations are gradients rather than cliffs. Going from 60% of energy to 40% is most of the available benefit for most people.

4. ⚠️ The framing carries a documented risk of disordered eating. A binary, morally-loaded food category with no defined endpoint is exactly Chapter 20 §20.13's structure. If "avoiding UPF" starts narrowing, generating guilt, or making eating with other people difficult, it has become the thing Chapter 34 is about.

5. And don't let it become another purchase. ⚠️ "UPF-free" is already appearing on packaging — including on products that are, by NOVA's own criteria, Group 4. Chapter 19's Ren bought seed-oil-free ultra-processed snacks; the same market response is arriving here.

22.13 What to actually do

A strategy that isn't "cook everything from scratch."

1. ⚠️ Start with drinks. Chapter 18 §18.7 is ✅ and soft drinks are Group 4 by definition. This is the highest-yield, lowest-effort change available and it doesn't require cooking anything.

2. ⚠️ Use Group 3 aggressively. Tinned beans, tinned tomatoes, tinned fish, frozen vegetables, frozen fruit, plain yoghurt, oats, wholemeal pasta, rice, eggs. Cheap, shelf-stable, require almost no skill, and don't spoil. This is the practical answer to §22.11 and it is not a compromise — it's most of the Chapter 10 §10.5 pattern.

3. Attack the snack layer first, not meals. ⚠️ For most people, Group 4 intake is concentrated in snacks, drinks and breakfast — not dinner. Nico's sports drinks. Theo's station pastry. The biscuit in the car.

4. ⚠️ Slow down. §22.5's second and third mechanisms. Eating rate is modifiable and free. Harder, chewier, more intact food does it automatically (Chapter 7's intact-versus-refined continuum).

5. Cook slightly more than you do, not everything. ⚠️ Two additional home-cooked meals a week is a real change. "Cook everything" is a plan that fails in week two (Chapter 10 §10.4).

6. ⚠️ And read the ingredients list rather than the nutrition panel. Five or fewer ingredients, all of which you recognize as food, is a crude and surprisingly effective heuristic. It's not NOVA and it doesn't need to be.

22.14 Where this sits against everything else in the book

⚠️ Chapter 20 §20.12b's comparative move, applied to Part IV.

⚠️ The top four are where the evidence is, and three of the four are about ultra-processed food directly or indirectly. The arguments that got the most attention are near the bottom.

⚠️ How firmly I hold these

What we don't know

⚠️ Whether UPF causes harm beyond the energy it delivers. The single biggest open question. Which features of formulation matter most — the ward trial matched several candidates and still found an effect, which narrows it without settling it. Whether a better classification exists (§22.7). And whether any of this is modifiable at population scale without changing what's on the shelves — which is Chapter 36's problem, not this chapter's.

22.15 The two honest positions, side by side

⚠️ And note: both columns would give exactly the same practical advice — less soft drink, fewer packaged snacks, more Group 1 and Group 3 food.

Sixth Part IV chapter in a row. ⚠️ Which is §22.9's whole point, stated one more time.

Now go back to your 🪞 Learning Check-In predictions.

1. ⚠️ Did you predict a difference, and how big? Most people predict a difference and underestimate it. If you predicted none, that was defensible and the trial says otherwise.
2. Did your three mechanisms include energy density, eating rate, or texture? ⚠️ Or were they all about ingredients? That's the Part IV reflex, and this chapter is about it.
3. Did you get the limitations? ⚠️ n=20, two weeks, unblinded. If you found all three before §22.4, your ten-minute check is working.

🧾 Economics

Eleventh consecutive chapter where the best-evidenced change is the cheapest — ⚠️ and the first where the cheapest food overall is the problem. That tension is the honest ending of Part IV.

Spaced Review

1. (Chapter 2) Why does the ward trial rise above every other piece of dietary evidence in this book?

It solved Chapter 2's problems simultaneously. ⚠️ Randomized · crossover (each person their own control) · inpatient with intake measured to the gram · no self-report · nutrient-matched comparator. Healthy-user bias eliminated by randomization; confounding eliminated by crossover. (Limits: n=20, two weeks, unblinded.)

2. (Chapter 19) Chapter 19 said you can't remove seed oils without removing eight other things. ⚠️ What does §22.9 do with that observation?

Generalizes it to all of Part IV. ⚠️ Every faction eliminated a component of ultra-processed food and credited the component. Sugar, seed oils, pesticides, additives and late eating are all removed by the same intervention — which is why five incompatible theories all produced real results. Each elimination works; each explanation is wrong; they're wrong in the same way.

3. (Chapter 17) Which myth shape does §22.9's second part describe, and why is it uncomfortable for this chapter?

It's not a shape from §17.1 — it's a shape about shapes. ⚠️ Every faction converted a systemic problem into a purchase, because a purchasable villain is individually actionable and a food environment isn't. ⚠️ It's uncomfortable because "eat less UPF" is already being sold as a shopping rule, including on Group 4 products labelled "UPF-free."

Project Checkpoint: Your UPF Audit

Component twenty-two. Three days, and count before you judge.

Step 1 — Classify. ⚠️ For three days, sort everything you eat into NOVA's four groups. Use §22.1. When unsure, look at the ingredients list: five or fewer, all recognizable as food, is usually not Group 4.

⚠️ Population averages in high-income countries commonly run above 50% of energy from Group 4. Most people guess low about themselves.

Step 2 — Where is it? ⚠️ Sort your Group 4 by occasion, because this determines what to do.

⚠️ For most people it is concentrated in the first three, which is good news — those are the easiest to change and don't require cooking.

Step 3 — The rate check. ⚠️ Time one Group 4 meal and one Group 1/3 meal.

Minutes to eat: Group 4 _ · Group 1/3 _

§22.5's second mechanism, measured on yourself.

Step 4 — Pick three changes, and make them structural rather than moral:

⚠️ If more than one requires cooking, you have designed §22.11's plan and it will fail in week two.

Step 5 — And write the honest constraint:

"The reason I eat Group 4 food is __."

⚠️ If the honest answer is time, money, energy, skill or access — say so. That is not a failure of knowledge and this chapter is not going to pretend otherwise. Chapter 31 and Chapter 32 are about exactly that.

Next checkpoint (Chapter 23): your training fuel — what you actually eat around exercise, if you exercise.

Chapter Summary

🚪 The threshold: processing is not a spectrum of badness. ⚠️ The question is not how far a food has been altered — it's what industrial formulation does to how much of it you eat. Group 3 contains excellent food; processing per se — pasteurization, canning, freezing, fortification — is among the great public health successes; and your homemade cake is not ultra-processed.

⚠️ The ward trial (Hall et al., Cell Metabolism, 2019): 20 adults, inpatient, randomized crossover, two weeks per arm, diets matched for presented calories, energy density, macronutrients, sugar, sodium and fiber, eating ad libitum. Result: ~500 kcal/day more on the ultra-processed diet; ~0.9 kg gained versus ~0.9 kg lost. And they ate faster.

Limits stated plainly: n=20 · two weeks · unblinded · fiber matched via supplemented beverages · inpatient.

⚠️ §22.9 — what Part IV was actually about, in two parts:

One: five factions each identified a real component of ultra-processed food and built a theory around it. Sugar, seed oils, pesticides and additives, and meal timing are all removed by the same intervention. ⚠️ Each elimination works. Each explanation is wrong. They are wrong in the same way, and that's why the debates never resolved — everyone was arguing about which ingredient in the biscuit was responsible, and the answer was the biscuit.

Two, and harder: ⚠️ every faction converted a systemic problem into a shopping decision, because a purchasable villain is individually actionable and a food environment is not. Which explains the premium product attached to every chapter in Part IV — and applies to this chapter too.

⚠️ §22.11: cooking from scratch requires time, skill, equipment, storage, energy, access and waste tolerance, all unequally distributed. Any version of this chapter that ends there has described a solution available to people who already have one.

The one thing to remember: ⚠️ Twenty people, the same nutrients, and a 500-calorie difference. Nobody was told to eat more. The food did it.

Part IV, in one table

⚠️ Six chapters, seventy-odd verdicts, and this is what survived.

⚠️ Count the third table. Thirteen entries, on the questions people argue about most.

That is not a failure of the field and it is not a failure of this book. It is what the evidence supports, and Chapter 17 §17.14 warned you that 🟡 and ⚗️ are the correct answer far more often than anyone finds comfortable.

⚠️ And the practical column — the first table — contains almost nothing anyone argues about.

Drink less soft drink. Eat more vegetables. Eat fish. Wash your hands. Don't reuse frying oil. Take folic acid if you might get pregnant. Eat less ultra-processed food.

None of that is contested by anybody in any of these debates. It is also, more or less, the whole of what Part IV establishes — which was §22.9's point, and is the honest summary of six chapters of argument.

What's Next

Part IV closes here. Chapter 23 opens Part V — Nutrition for Life Stages and Goals, and the register changes completely.

⚠️ The arguing is over. Part V is about specific situations with specific answers: training and competition, weight management, ageing, chronic disease, the microbiome, allergies and intolerances, and clinical nutrition.

Chapter 23 starts with sports nutrition — where the evidence is unusually good, because athletes can be measured, outcomes are objective, and performance is a much cleaner endpoint than "health." Carbohydrate periodization, protein timing tested properly, hydration during competition, the supplements that actually work (there are about five), and — returning to Devi — what happens when fuelling goes wrong.