Chapter 1 — Why Is Nutrition So Confusing? How Bad Science, Bad Media, and Bad Incentives Created a World of Dietary Chaos

The Hook: The man with the folder

Theo Vasquez brought a folder to his first appointment.

Not a phone — a folder. Green, cardboard, the kind you get in a pack of ten. Inside were printouts. Actual printouts, on actual paper, some of them with sections highlighted in two colors. He set it on my desk with the specific carefulness of a person who has decided to be taken seriously this time.

He is thirty-four. Software support, which means he spends nine hours a day in a chair fixing other people's problems, and he came to me because his physician had used the word prediabetes and given him a photocopied handout about the plate method, and he had gone home and — being the kind of person who fixes problems — done research.

"I've been doing this since 2016," he said. "I want you to know I've actually tried."

He had. He walked me through the folder.

2016: low fat. He cut fat, ate a lot of things labeled light, lost eleven pounds, felt hungry constantly, and put the weight back on over about fourteen months.

2018: paleo. No grains, no legumes, no dairy. He liked this one. His wife did not. It survived until a family vacation.

2020: keto. This was the serious one. Eighteen pounds in four months. He showed me a photo. Then his triglycerides came back higher than before, which confused him, and he was so tired at 3 p.m. that he was falling asleep on support calls, and when he came off it he regained twenty-four.

2022: a ten-day juice cleanse. He had the least to say about this one.

2023: a seventy-five-day training challenge. He made it to day thirty-one, which he described as failing. Thirty-one consecutive days of two workouts, a gallon of water, and a diet he'd chosen himself, and he used the word failing.

Then he opened the folder to the last section, which was the reason he'd brought it, and it was about a dozen printed articles. He'd arranged them in pairs.

Eggs raise cholesterol and increase heart disease risk. Facing: Eggs are a nutrient-dense protein source with no established link to cardiovascular events in healthy adults.

Saturated fat is the primary dietary driver of heart disease. Facing: Meta-analysis finds no significant association between saturated fat intake and cardiovascular mortality.

Red meat is a probable carcinogen. Facing: Panel recommends adults continue current red meat consumption.

Seed oils are inflammatory and should be eliminated. Facing: Replacing saturated fat with polyunsaturated fat reduces cardiovascular events.

Every one of them was real. Every one of them had come from a source a reasonable person would trust — a university press office, a major newspaper, a medical journal's own summary, a physician with credentials. He hadn't been reading garbage. That was the problem. He'd been reading the good stuff, and the good stuff didn't agree.

He looked at the pairs, then at me.

"So which one of you is lying?"

I want to be careful about how I answer that, because the honest answer has two halves and most books only give you one.

The first half is that Theo is not confused because he's stupid or lazy or lacks discipline. He is confused because the information environment around food is genuinely, structurally, almost uniquely hostile to a careful person. He did what a careful person does — he went and read — and the environment punished him for it. He has more nutrition information in his head than most people alive in 1975, and it has made his eating worse, because information that contradicts itself produces paralysis and then, eventually, cynicism.

The second half — and this is where a lot of skeptical books go wrong — is that it does not follow that nobody knows anything. Theo's pairs of headlines look like a field in total disarray. They aren't. If you sorted every claim in nutrition by how well supported it is, you'd get a very small pile of things we're genuinely certain about, a bigger pile we're fairly confident about, and an enormous pile we're guessing at. Theo's folder was drawn entirely from the third pile, because that's where the headlines live. The first pile doesn't generate headlines. Nobody writes Fiber Continues To Be Good For You, Twenty-Fourth Consecutive Year.

This chapter is about how the third pile got so loud.

Five reasons, and they compound: nutrition is structurally hard to study, we cannot measure what people eat, there's a pipeline that turns modest findings into confident content, the incentives of nearly everyone who talks to you about food are misaligned with your health, and — the one we like least — we want it to be simpler than it is.

By the end of this chapter you should be able to explain to someone why the headlines contradict each other, without shrugging and without conspiracy. And by the end of the next chapter, you'll be able to sort the piles yourself.

🏃 Fast Track: If you already accept that nutrition media is unreliable and you want the tools rather than the diagnosis, read §1.4 (the pipeline) and §1.7 (what "experts keep changing their minds" actually means), then go straight to Chapter 2. Do the Project Checkpoint first, though — it's the Belief Inventory, and it only works if you write it before you know what's coming.

🔬 Deep Dive: §1.2 and §1.3 are the methodological core — why diet resists study and why we can't measure intake. If you're taking a course, those two sections are where the exam questions live, and §1.5 (incentives) is where the essay questions live.

1.1 The eggs problem

Let's start with the single most-flip-flopped food in the Western diet, because tracking it teaches more than any general argument.

If you are over fifty, you have lived through roughly this sequence of official-sounding advice about eggs:

• Eggs are an excellent, cheap, complete protein. (Most of the twentieth century.)
• Eggs are high in cholesterol; cholesterol in food raises cholesterol in blood; limit them. (Roughly the 1970s through the 1990s. A specific numeric limit on dietary cholesterol became standard guidance.)
• Actually, dietary cholesterol has a much weaker effect on blood cholesterol than we thought for most people, because the liver adjusts its own production. The specific numeric limit was dropped from US dietary guidance in the 2015–2020 edition.
• But that doesn't mean unlimited eggs, because dietary patterns high in cholesterol also tend to be high in other things, and some observational analyses still find associations.
• And also, individual response varies enormously — there are people whose blood cholesterol is genuinely sensitive to dietary cholesterol and people in whom it barely moves.

Now: which of those steps was a lie?

None of them. That's the uncomfortable part. Each step was a defensible reading of the evidence available at the time. The reason it feels like whiplash is not that the science reversed five times. It's that a slow, hedged, population-level scientific conversation was being narrated to the public in headlines, and headlines cannot carry hedges. "Eggs are fine for most people, in the context of an overall dietary pattern, with meaningful individual variation, and we're less confident about the dietary-cholesterol mechanism than we were in 1985" is not a headline. It is barely a sentence.

So what you received was: EGGS BAD. Then, years later: EGGS FINE. Then: EGGS BAD AGAIN.

The science moved a few feet. The reporting moved a few miles, in both directions, repeatedly.

🔍 Why this works. Headlines have a structural requirement that scientific findings don't: they have to be new. A finding that slightly refines an existing estimate is not new. A finding that reverses an existing estimate is extremely new. So of the hundreds of nutrition studies published each month, the ones that get written up are systematically biased toward the surprising, the reversing, and the counterintuitive — which is precisely the subset most likely to be a statistical fluke that won't replicate. The selection process that determines what reaches you is anti-correlated with reliability. This is not anyone's fault in particular, and it is not fixable by better journalists. It is a property of the system.

Keep the eggs sequence in mind. We're going to come back to it, because it's the cleanest example of a thing that will recur for the rest of this book: most apparent reversals in nutrition are reversals in narration, not in evidence.

🔄 Check your understanding. A friend says: "They used to say eggs were bad, now they say they're fine — clearly nutrition scientists have no idea what they're doing." What's the most accurate one-sentence response?

Answer

Something like: "The recommendation shifted because the evidence on how much dietary cholesterol affects blood cholesterol got better, and the shift was small — what changed a lot was how confidently the media reported each version." The key move is separating what the evidence did (refined gradually) from what the headlines did (reversed dramatically). If you answered "because science is self-correcting," that's true but incomplete — it concedes that the science reversed, when mostly it didn't.

1.2 Reason one: diet is structurally hard to study

Here is a thought experiment I use with students, and I'd like you to actually attempt it rather than reading past it.

🧩 Productive struggle. You have unlimited funding and full ethical approval. Your task: prove definitively whether eating red meat causes heart disease.

Design the study. Take three to five minutes and write down the actual design — who's in it, what they do, for how long, what you measure, what your control group does.

Then read on and see how many walls you hit.

Here's what happens when you try.

Wall one: you can't randomize people to a diet for thirty years. Heart disease takes decades to develop. So your trial needs to run for decades. Nobody has ever successfully kept a large group of free-living humans on an assigned diet for decades. People move, get sick, get bored, get married to someone who cooks differently, and drop out — and the ones who drop out are systematically different from the ones who stay, which corrupts your results.

Wall two: you can't blind it. In a drug trial, neither the patient nor the doctor knows who got the pill and who got the placebo. That's what makes drug trials so powerful. But nobody has ever eaten a steak without noticing. Everyone in your study knows which group they're in, which means everyone's expectations, behavior, and reporting are affected.

Wall three: everyone is already eating something. There is no diet-naive human. In a drug trial, the control group takes nothing. In a diet trial, the "control group" is eating a diet too — some other diet, with its own effects. Which means you are never testing red meat versus nothing. You are testing red meat versus whatever replaced it, and the answer changes completely depending on what that is. Replace red meat with beans and you get one answer. Replace it with white bread and you get a different one. Replace it with chicken and you get a third.

This is the single most important structural fact in all of nutrition science, and we'll formalize it in the next chapter as the substitution question. Almost every argument you have ever had about food was unanswerable because nobody specified the comparison.

Wall four: the effect sizes are small and the noise is enormous. A drug might halve your risk of an outcome. A dietary change might shift it by ten or fifteen percent, over decades, against a background of genetics, smoking, exercise, sleep, income, air quality, healthcare access, and several hundred other dietary components that all changed at the same time. Finding a small signal in that much noise requires either enormous samples or very long follow-up, and usually both.

Wall five: the ethics. You cannot randomize people to something you suspect is harmful. You cannot randomize children to a nutrient-deficient diet to see what happens. Some of what we most want to know is permanently off-limits to the strongest study design.

So what do we actually do? We compromise, in two main directions:

1. Short, tightly controlled trials on intermediate markers. Feed people a specific diet for twelve weeks and measure LDL cholesterol, blood pressure, insulin sensitivity, or inflammatory markers. These are rigorous — sometimes participants live in a metabolic ward and every gram is weighed — but they measure a surrogate, not the outcome you care about. Lowering a marker for twelve weeks is not the same as not having a heart attack in 2049.

2. Long observational cohorts. Recruit tens of thousands of people, ask what they eat, follow them for twenty or thirty years, and see who gets sick. These measure the real outcome over the real timescale. But they can't establish causation, and they're haunted by a problem so central that it gets its own section in the next chapter: the people who do the healthy thing also do all the other healthy things.

Nutrition's entire evidence base is built out of those two compromises, plus animal work, plus mechanism. Neither compromise is bad science. Both are incomplete science, in complementary ways — and the arguments you see online are almost always one side citing the trials and the other side citing the cohorts, without either acknowledging what their preferred design cannot do.

📊 Diagram (described). Picture a target with a bullseye labeled "Does eating X change whether I die of Y?" Now picture two archers. The first archer — the controlled feeding trial — stands very close, has excellent aim, and is shooting at a different, smaller target placed beside the real one, labeled "Does eating X change this blood marker over twelve weeks?" She hits her target almost every time, with tight grouping. But it is not the bullseye. The second archer — the long observational cohort — is aiming at the actual bullseye, and is standing three hundred yards away in fog and crosswind. Occasionally she hits it. Her grouping is wide, and you can never be sure whether a hit was skill or luck. Nutrition science is the practice of triangulating the real bullseye from one archer with great aim at the wrong target and another with poor aim at the right one. When both archers point the same direction — as they do for fiber, for trans fats, for alcohol — you can be reasonably confident. When they disagree, you get a decade of arguing.

1.3 Reason two: we cannot measure what people eat

If §1.2 didn't convince you, this will.

Suppose you solve every problem above. Long study, huge sample, real outcomes, adequate funding. You still have to answer the foundational question: what did these people eat?

And the honest answer is that we don't know, because the measuring instrument is human memory about food, and human memory about food is spectacularly, systematically unreliable.

The main tool in large studies is the food frequency questionnaire (FFQ) — a form asking how often, over the past year, you consumed each of a hundred-odd food categories, and in what portion size. Sit with that for a second. Over the past year. How many times did you eat rice? What portion? How many servings of leafy greens per week, on average, across all four seasons?

Nobody knows this. I have a doctorate in this field and I could not tell you within a factor of two how much rice I ate last year.

Other tools have their own problems. 24-hour recall — what did you eat yesterday — is more accurate for yesterday but yesterday might not be typical, so you need many recalls per person, which is expensive. Food diaries are better still, and change behavior the moment you start keeping one (people eat differently when they're writing it down, which is either a bug or, as we'll see in Chapter 4, a feature). Biomarkers — measuring nutrients in blood or urine — are objective but exist for only a handful of nutrients and don't tell you what food they came from.

And the errors aren't random. That's the crucial part. If the errors were random noise, large samples would average them out. But they're systematic:

Now stack that on top of §1.2. You have an observational study, which can't establish causation, in which the exposure — the thing you're studying — is measured by an instrument that is wrong in predictable directions correlated with the outcome you care about.

This is why serious nutritional epidemiologists spend so much of their time explaining why their own findings should be held loosely. My colleague Samir Ostrowski — you'll meet him properly in the next chapter — describes it as "doing astronomy with a telescope that fogs up more when you point it at the interesting parts of the sky."

🍽️ On your plate. This has one immediate practical consequence, and it's the first actionable thing in this book. When you see a headline about a study of what people eat, ask how the diet was measured. If it's a food frequency questionnaire in an observational cohort — which it very often is — the finding is a hypothesis, not a result. That doesn't make it worthless. It makes it a starting point that needs corroboration from a different kind of evidence before you rearrange your life around it.

🔄 Check your understanding. A study reports that people who eat more ultra-processed food have higher rates of depression. Name two measurement problems that could produce this association even if ultra-processed food had no effect on mood whatsoever.

Answer

Several good answers. Two of the strongest:

1. Differential reporting. People experiencing depression may report their diet differently — more accurately, less accurately, or with a different self-critical lens — than people who aren't. The exposure measurement is affected by the outcome.
2. Reverse causation. Depression reduces energy, motivation, and cooking. The depression could be causing the ultra-processed food intake rather than the other way around. Observational data generally cannot distinguish these directions.

Also credit for: confounding by income, time poverty, or living alone; and for noting that "ultra-processed" is itself a hard category to capture on a questionnaire. If you got two of these, you're already doing Chapter 2's job.

1.4 Reason three: the pipeline

Now we come to the part that Theo's folder was made of.

I want to trace, step by step, how a real, modest, honest finding becomes a confident claim in your feed. I'm going to use a composite example — the specifics are illustrative, but every step in this pipeline is something I've watched happen, repeatedly, with real studies.

Step 0 — The finding

A research group runs a study in mice. They're investigating how a particular fatty acid affects markers of inflammation in liver tissue. They feed one group of mice a diet where a large fraction of calories comes from a specific oil, and observe elevated inflammatory markers in the liver compared to controls.

This is fine work. It generates a hypothesis. In the paper, the authors are appropriately careful: they note the dose is far higher than typical human consumption, that mice metabolize fats differently from humans, that they measured markers rather than disease, and that human studies are needed.

What the study established: at a high dose, in mice, this oil raised a marker. That's it. That is rung 2 on the evidence ladder you'll learn next chapter, and rung 2 tells you what might happen, not what does.

Step 1 — The press release

The university's communications office writes a press release. Their job — their actual, funded, performance-reviewed job — is to generate coverage for the institution. So the release is titled something like "Common Cooking Oil Linked to Liver Inflammation, Study Finds."

Notice what happened. "Mice" left the headline. "High dose" left the headline. "Marker" became "inflammation." Nothing in the release is technically false — the caveats are all there in paragraph six — but the emphasis has moved. The press officer is not a villain. They are doing the job they were hired to do, and they probably ran the release past the senior author, who probably approved it while distracted, because the alternative is a press release nobody reads.

Step 2 — The news article

A health writer covering four stories that day reads the press release. Maybe they read the abstract. Very few read the full paper; there isn't time, and often there isn't access.

The article becomes: "Is Your Cooking Oil Making You Sick? New Research Raises Concerns." Paragraph four mentions the study was in mice. Paragraph nine has an outside expert saying the findings are preliminary. Roughly nine in ten readers never reach paragraph four.

Step 3 — The aggregation

Six other outlets rewrite the first article without going back to the source. Each rewrite is a degraded copy of a degraded copy. By the fourth iteration, "linked to markers of inflammation in mice" has become "causes inflammation." The mice are gone entirely. So is the dose.

Step 4 — The reel

Now Coach Brandt gets it.

I want to be precise about Brandt, because he's going to recur throughout this book and it matters that you understand him correctly. He is a composite of a very common type: a fitness influencer with about two million followers, genuinely in excellent shape, hard-working, personally disciplined, and — this is the part people get wrong — sincere. He is not knowingly lying. He reads a lot of health content. He believes he is helping people, and by his own lights he sometimes is; some of his training advice is good.

He makes a forty-second video. Quick cuts, a bottle of oil held up to the camera, an incredulous expression. "They're putting this in EVERYTHING and it's inflaming your liver. There's a study. Look it up."

There is a study. He is telling the truth about that.

The video gets eleven million views, because the algorithm rewards emotional response, and outrage and fear are the cheapest emotional responses to produce. A calm forty-second video saying "a mouse study found a marker changed at a high dose, we should look into it in humans" would get four thousand views, and Brandt knows this, and the knowing is what slowly bends him.

Step 5 — The belief

Six weeks later, Theo Vasquez throws out a bottle of oil.

Nobody in this chain lied. The researchers were careful. The press officer did their job. The journalist met their deadline. The aggregators aggregated. Brandt shared something he believed. And the output is a man changing his behavior based on a mouse study he's never seen, at a dose he'll never eat.

💡 Aha moment. The pipeline doesn't require anyone to be dishonest. It only requires that each participant be slightly more confident than the previous one — because at every single stage, the incentives reward confidence and punish hedging. Five slight amplifications compound into a total distortion. This is why "who's lying to me?" is usually the wrong question. Nobody is. The system distorts without needing liars, which makes it much harder to fix and much easier to fall for.

🔄 Check your understanding. At which step in the pipeline was the first factually false statement made?

Answer

Arguably step 3 — the aggregation — where "linked to markers of inflammation in mice at high dose" became "causes inflammation," dropping the species and the dose. Steps 0–2 were each technically accurate but progressively less complete.

The important insight is that the most damaging step was not the first false one. Steps 1 and 2 did most of the work by moving the emphasis while staying technically true, and by the time an outright falsehood appeared, the ground had already been prepared. Misleading-but-true is more dangerous than false, because it survives fact-checking.

1.5 Reason four: everybody's incentives are wrong

This is the section where I have to be even-handed in a way that will annoy people on both sides, which is generally a sign you're doing it right.

There are two industries talking to you about food, and both of them are misaligned with your health. Most books pick one to be angry about. That's a mistake, and it's the mistake that makes people vulnerable — because a person who correctly distrusts Big Food is very often the person who then buys $2,244 of supplements a year from someone else.

The food industry

Food companies operate on thin margins in a market where total human stomach capacity is fixed. They grow by capturing a larger share of a fixed pie. That means designing products that are extraordinarily easy and pleasant to over-consume, and marketing them effectively.

They also fund research. This is where it gets genuinely complicated, and where I want to be careful.

The best-documented historical case is the Sugar Research Foundation episode. Internal industry documents that came to light decades later showed that a sugar industry body funded a review of dietary factors in coronary heart disease in the 1960s, and that the funding relationship was not disclosed in the published review, and that the review's framing was favorable to sugar and unfavorable to fat. The historians who surfaced these documents made a strong case that industry funding shaped the emphasis of an influential piece of scientific literature at a formative moment for the field.

That is a real, documented, serious episode. It happened.

And here is where I have to hold the line, because this episode has become the load-bearing citation for an enormous number of claims it does not support. It is routinely invoked to argue that the entire dietary guidance apparatus is an industry fabrication, that saturated fat was "exonerated" and framed, that everything you've been told is inverted. That's a much larger claim than the evidence carries. What the documents show is that industry funding influenced the framing of some mid-century research. What they don't show is that every subsequent finding across sixty years and dozens of countries and thousands of independent investigators is therefore void.

🔬 Claim → Evidence → Verdict

The claim: "Nutrition science is bought and paid for by industry. Dietary guidelines are written by people on food-company payrolls. You can't trust any of it."

Where it comes from: Real and documented instances of industry influence — the Sugar Research Foundation episode being the best known — plus genuine, ongoing problems: food companies fund a substantial share of nutrition research; some guideline committee members have industry ties; agricultural commodity groups fund studies about their own commodities; and there is a well-documented general pattern in which industry-funded studies produce results favorable to the funder more often than independently funded ones do.

What the evidence actually shows: The funding-effect problem is real and should change how you read a single study. But the inference from "some research is compromised" to "all conclusions are void" doesn't hold, for three reasons. First, the strongest nutrition conclusions are supported by many independent lines of evidence across countries with completely different agricultural industries and funding structures — the fiber evidence looks the same in Finland and Japan. Second, industry influence pushes in competing directions: the dairy industry, the beef industry, the sugar industry, the grain industry, and the supplement industry all fund research, and they want incompatible things. Third, the claim is usually deployed selectively — to discredit findings the speaker dislikes, while the speaker's own preferred findings, often funded by the supplement industry, are exempted.

📉 Evidence quality: Mixed. Strong historical documentation for specific episodes; strong meta-research on funding effects; no support for the total-invalidation conclusion.

Verdict: 🟡 Unclear / it depends. Industry influence on nutrition research is real, documented, and a legitimate reason to check who funded a given study. It is not a reason to discard conclusions that rest on many independent lines of evidence — and "follow the money" is only honest if you follow it in every direction, including toward the person telling you to follow it.

The wellness industry

Which brings us to the other side.

The global dietary supplement market runs to tens of billions of dollars annually in the US alone, and it is built on a structural advantage that most consumers don't know about: under the Dietary Supplement Health and Education Act of 1994 (DSHEA), supplements in the United States are regulated much more like food than like drugs. Manufacturers generally do not have to demonstrate efficacy before selling a product, and the burden falls largely on regulators to demonstrate harm after it's on the market. We'll spend a whole chapter on what that means in practice (Chapter 16).

Add to that: influencers with affiliate codes. Wellness brands funding "studies" designed as marketing. Practitioners selling the tests that diagnose the conditions they then treat. Subscription supplement boxes. $200 microbiome analyses. Detox protocols with a product attached.

Here is the asymmetry that should bother you: the food industry wants you to eat more of its product, and the wellness industry wants you to be afraid of the food industry's product so you'll buy its product. Fear of the first is the business model of the second. That's why wellness content is so much angrier than food advertising — anger is the acquisition channel.

🧾 Cost check. Walt Prosser, whom you'll meet properly in Chapter 16, is sixty-eight, retired from the postal service, and living on a fixed income. He spends $187 a month — $2,244 a year — on nine supplements. Reviewed against the evidence, roughly two of them are defensible for him, two are harmless-but-pointless, and two are actively risky, including one that nearly caused a catastrophic misreading of an emergency-room blood test. At his budget, $2,244 a year is a great deal of fresh produce, or a chunk of a dental bill, or six months of not worrying about the electricity. The cost of bad nutrition information is not only measured in health.

🍽️ On your plate. A simple, portable heuristic: whenever someone tells you what to eat, ask what happens to their income if you agree. Apply it to food companies. Apply it to influencers, supplement brands, functional-medicine clinics, and diet-book authors. Apply it to me — I'm a dietitian; my profession has an interest in you believing that dietitians are useful. This isn't cynicism and it doesn't mean everyone with an incentive is wrong. It means the incentive is a reason to check, and the people most worth listening to are usually the ones who lose nothing if you ignore them.

1.6 Reason five: we want it to be simple

The four reasons so far are all external — the science is hard, measurement is bad, media distorts, incentives corrupt. Comfortable reasons. Nothing is our fault.

Here's the fifth, and it's the one that makes the other four work.

We want a villain.

Real nutrition is multifactorial, probabilistic, context-dependent, and slow. Nobody wants that. What people want — what I want, when I'm not being careful — is a single cause with a single fix. It's carbs. It's seed oils. It's gluten. It's lectins. It's sugar. It's the cooking method. It's the timing. One thing, identified, removed, problem solved.

This is a deep and old cognitive pattern, and it shows up in a small number of recurring shapes. Learn the shapes and you can spot the next myth before it has a name — which is the actual skill, because the specific myths turn over every eighteen months.

Notice that these shapes are about morality, not biology. Clean and dirty. Virtue through sacrifice. Corruption by industry. Fall from a natural state. Nutrition is unusually vulnerable to this because eating is one of the few biological processes that is also a daily moral performance — we describe food as sinful, guilty, indulgent, virtuous, and we describe ourselves as good or bad based on what we ate at 3 p.m.

That's not a scientific vocabulary. It's a religious one, and it got attached to a scientific field.

🔍 Why this works. Consider why "restriction as virtue" is so persistent despite being unsupported as a general principle. Eliminating a food category produces three things at once: immediate identity ("I'm someone who doesn't eat X"), a visible in-group, and a legible metric of effort. All three are psychologically rewarding right now. Meanwhile, the actual health effect — if any — is invisible, delayed by years, and probabilistic. Given a choice between an immediate certain psychological reward and a distant uncertain physical one, the brain is not close to neutral. The behaviors that feel most like health are selected for feeling, not for outcome, which is exactly why the diets that feel most virtuous are so often the ones with the least evidence and the highest dropout rates.

🔄 Check your understanding. Someone tells you that a specific ancient grain, eaten daily, "resets your metabolism." Which two shapes from the table is that claim using?

Answer

The lost golden age (ancient, therefore correct — our ancestors knew something we forgot) and the hidden simple fix (one food, one profound systemic effect). Often a third is implied — the suppressed truth — via some version of "the food industry doesn't want you to know about this."

Note that identifying the shapes doesn't prove the claim false. It tells you why the claim is spreading, which is a different question from whether it's true — and it tells you to demand better evidence than you'd otherwise require, because the claim has a tailwind that has nothing to do with its accuracy.

1.7 "But experts keep changing their minds"

Time to deal with the sentence directly, because it's the load-bearing complaint underneath Theo's folder and it deserves a serious answer rather than a defensive one.

There are three different things this complaint can mean, and they have three different answers.

Meaning 1: "The reporting changed" — usually true, and not the scientists' fault

This is the eggs case from §1.1, and it's the majority of instances. The underlying scientific position moved modestly and the public narration swung wildly. When you feel whiplash, check whether the guidelines actually reversed or whether the headlines did. They're very different documents, and the guidelines change slowly and with visible reasoning that you can go read.

Meaning 2: "The evidence genuinely changed" — sometimes true, and this is science working

Sometimes the science does move, substantially, and it should.

The clearest example in modern nutrition is infant allergen introduction. For years, guidance in several countries advised delaying the introduction of allergenic foods like peanut to infants at risk. Subsequent randomized evidence indicated that early introduction reduced the development of peanut allergy, and guidance reversed.

That is a genuine, substantial reversal. It is also exactly what you want a scientific field to do. The earlier advice was a reasonable inference from weaker evidence; better evidence arrived; the recommendation changed. A field that never reverses is a field that isn't testing itself.

The failure mode isn't reversal. It's reversing without saying why, or refusing to reverse when the evidence demands it.

Trans fats are the other clean example. Partially hydrogenated oils were once promoted as a heart-healthy alternative to butter. Evidence accumulated that they were substantially worse than what they replaced. Regulators eventually acted, and they have been largely removed from the food supply in many countries. That is not a scandal. That is a system detecting an error and correcting it — slower than it should have, at real human cost, but correcting it.

Meaning 3: "Nobody agrees, so nothing is known" — false, and the most damaging of the three

This is the conclusion Theo had reached, and it's the one that hurts people.

Because there is a large body of nutrition knowledge that has been stable for decades, is agreed on across essentially every independent expert body worldwide, and generates no headlines whatsoever. A partial list:

• Severe deficiency of specific micronutrients causes specific, well-characterized diseases, and correcting the deficiency cures them. This has been settled for a century.
• Folic acid before and during early pregnancy reduces neural tube defects. Randomized evidence. Not contested.
• Trans fats increase cardiovascular risk. Settled enough to have driven regulation.
• Alcohol is a Group 1 carcinogen in the IARC classification — the same category, in terms of strength of evidence for carcinogenicity, as tobacco and asbestos. (Group 1 describes confidence that it causes cancer, not magnitude of risk — an important distinction we'll unpack in Chapter 12.)
• Higher fiber intake is associated with lower cardiovascular disease and colorectal cancer risk, consistently, across cohorts, with supporting trial evidence on intermediate outcomes and a plausible mechanism.
• Dietary patterns high in vegetables, fruit, legumes, whole grains, and nuts, and low in ultra-processed food and sugar-sweetened beverages, are associated with better health outcomes — across countries, cuisines, and study designs, for fifty years.
• Sugar-sweetened beverages are associated with weight gain and type 2 diabetes risk beyond what their calories alone would predict, with supporting trial evidence.

None of that is fringe. All of it is boring. The boring parts are where the certainty lives, and the reason is almost mechanical: a claim only becomes exciting when the evidence is weak enough to argue about. Where evidence is overwhelming, there's no argument, so there's no content, so you never hear about it.

🔬 Claim → Evidence → Verdict

The claim: "Nutrition experts constantly reverse themselves, so nutrition science can't be trusted and you might as well eat whatever you want."

Where it comes from: A completely real experience. Anyone who has followed nutrition coverage for twenty years has watched eggs, fat, salt, coffee, and red wine each be condemned and rehabilitated, sometimes more than once. The frustration is legitimate and the people expressing it are not being unreasonable.

What the evidence actually shows: Three separate things are being conflated. (1) Media narration reverses constantly — true, and mostly not a reflection of the underlying science. (2) Formal guidance changes slowly and incrementally, with published reasoning — the actual guideline documents show far less volatility than the coverage. (3) Genuine reversals do occur (infant allergen introduction being the clearest), and those represent the field working correctly. Meanwhile, the stable core — deficiency diseases, folate in pregnancy, trans fats, fiber, dietary patterns, alcohol — has not reversed and shows no sign of doing so.

📉 Evidence quality: The premise is accurate about media. The conclusion doesn't follow from it.

Verdict: 🟠 Probably false. The observation is right; the inference is wrong. "Experts keep changing their minds" mostly describes journalism, and the practical conclusion — therefore eat whatever — throws away a large body of stable knowledge to escape a much smaller body of genuinely unsettled argument.

🔬 Claim → Evidence → Verdict

The claim: "There's a study for everything, so you can prove anything you want about food."

Where it comes from: Genuinely true observations: tens of thousands of nutrition papers publish annually; publication bias favors positive findings; small studies produce noisy results; and anyone can search until they find support for a prior belief. Both sides of every nutrition argument really do have citations.

What the evidence actually shows: True at the level of individual studies, false at the level of bodies of evidence. You can find a single study supporting almost anything. You cannot find a systematic review of well-conducted trials plus consistent cohort data plus a plausible mechanism plus independent replication supporting almost anything. The whole point of the evidence ladder in Chapter 2 is that individual studies are the weakest unit of analysis, and the skill being taught is to stop reading nutrition one study at a time.

📉 Evidence quality: Accurate description of the literature's surface; false description of how conclusions are actually reached.

Verdict: 🟡 Unclear / it depends — true for single studies, false for evidence bases. The distinction is the entire subject of the next chapter.

1.8 What this book is going to do about it

Let me tell you the plan, so you know what you're signing up for.

This book will not give you a diet. I've watched diets work for people and then fail the same people two years later, and the pattern has almost nothing to do with the diet's macronutrient composition. Chapter 10 goes through the head-to-head evidence in detail; the short version is that the diet wars have a winner and the winner is "whichever one you'll actually keep doing."

It will teach the science from the ground up — digestion, energy, metabolism, macronutrients, micronutrients, the microbiome, clinical applications — assuming no chemistry, no biology, and no statistics.

And it will teach you to evaluate claims yourself, which is the part that survives the book going out of date. Every chapter runs real claims through the same device you've now seen three times: 🔬 Claim → Evidence → Verdict, with a six-level scale that includes "unclear," "untested," and "well supported."

That scale matters more than it might look. A book in which every verdict is ❌ is not skepticism — it's contrarianism, which is credulity pointed the other way. Some things the wellness world believes are true. Creatine works. Vitamin D helps people who are actually deficient. Fiber is as good as advertised. Older adults do need more protein than the RDA specifies. Fermented foods look modestly promising. You'll see those verdicts too, in the same voice I use to tell you your liver doesn't need a cleanse.

What we don't know

Every chapter in this book contains a section like this one, because a book about evidence that never admits ignorance is doing the thing it criticizes.

So, honestly, for this chapter: I don't know how much of nutrition's replication problem is measurement error versus genuine heterogeneity between people. When a dietary intervention works beautifully for one person and does nothing for another, there are at least three explanations — they didn't actually do what they reported, there's real biological variation in response, or the effect was never there and we're looking at noise. These are very hard to distinguish, and the honest state of the field is that we often can't. Chapter 35 looks at whether personalized nutrition can help with this yet. The preview: less than it's selling.

I also don't know how much of the observed benefit of "healthy dietary patterns" is the food and how much is everything correlated with eating that way — money, time, education, healthcare access, neighborhood, stress, sleep. This is the central unresolved question in nutritional epidemiology, and anyone who tells you it's settled is selling something. It's the subject of the next chapter, and I'm telling you up front that we're going to end that chapter without a clean answer.

That's not a failure of the book. That's the actual state of the knowledge, and knowing where the uncertainty lives is more useful than a confident number.

Project Checkpoint: Your Belief Inventory

This is the first component of Your Nutrition Framework, the project you'll build across all 38 chapters. It takes about twenty minutes, and it only works if you do it now, before you read Chapter 2 — because its whole value comes from being written while you still believe what you currently believe.

The task: write down ten things you currently believe about food and nutrition, and where each one came from.

Use whatever's handy — paper, a notes app, the worksheet in Appendix F. Four columns:

Rules that make this worth doing:

1. Include beliefs you're confident about. The comfortable ones are the point. Anything you'd rate a 5 is more valuable here than something you already doubt.
2. Be specific about the source. "The internet" is not a source. Try to actually remember: was it a person, a class, an article, an ad, a video, a doctor, your family? A surprising number will turn out to have no traceable origin at all — you just know it. Those are the most interesting entries on the sheet.
3. Include at least two beliefs that changed your behavior — something you buy, avoid, time, or spend money on.
4. Don't research anything. Not now. Record what you believe today.
5. Then seal it. Fold the paper, or don't reopen the file. You will not look at this again until Chapter 17, where the checkpoint is to grade it.

Why the sealing matters: hindsight bias is powerful and fast. If you keep the list open while reading, you'll unconsciously revise it, and in Chapter 17 you'll find yourself thinking well, I never really believed that. You did. Write it down while it's true.

A note for the honest: most readers, in Chapter 17, find that between two and four of their ten were wrong or substantially overstated — and that the ones they rated most confident are overrepresented among the errors. That's not a failure. It's the single best evidence that the book worked, and it's the same thing that happens to me every time I read outside my specialty.

Next checkpoint (Chapter 2): your Claim Filter — the six-question card you'll run every future nutrition claim through.

Chapter Summary

Why nutrition is confusing — the five compounding reasons:

The three bins. Every nutrition claim belongs in one of these, and most of this book is about learning to sort them:

• What we KNOW — small, boring, stable for decades, generates no headlines
• What we're FAIRLY CONFIDENT about — dietary patterns, mostly; larger than skeptics admit
• What we're GUESSING — enormous, and where nearly everything anyone tries to sell you lives

The four questions to ask any nutrition claim (a preview of Chapter 2's Claim Filter):

1. Compared to what? — every dietary change is a substitution, never a deletion
2. In whom? — mice? twelve men in a metabolic ward? a population unlike you?
3. How was diet measured? — if it's a questionnaire in a cohort, it's a hypothesis
4. Who benefits if I believe this? — asked in every direction, including at the skeptic

This chapter's verdicts:

The one thing to remember: Theo isn't confused because he's careless. He's confused because he was careful in an environment that punishes carefulness. The fix isn't to try harder at reading headlines. It's to stop reading nutrition one study at a time.

What's Next

Everything in this chapter was diagnosis. Chapter 2 is the treatment, and it's the most important chapter in this book — the one chapter with no acceptable skip.

You'll get the eight-rung evidence ladder, and you'll learn to place any study on it in about ten seconds. You'll learn healthy-user bias, which is the single biggest reason nutrition headlines are wrong and which, once you see it, you will never stop seeing. And you'll get the question that quietly unlocks every nutrition argument you will ever have: compared to what, in whom, for how long?

After that, the rest of the book is just applying it.