Chapter 10 — The Diet Wars: Low-Carb vs. Low-Fat vs. Mediterranean vs. Vegan — What the Long-Term Evidence Actually Shows

The Hook: Theo's five

Here is Theo Vasquez's diet history, which he wrote out for me on the back of one of his own printouts when I asked him to.

Five serious attempts. Eight years. Net weight change across the whole period: up about 14 pounds.

Now look at the last column, because that's the one nobody looks at.

Not one of those diets ended because it stopped working. Every single one ended because it stopped being possible — hunger, a spouse, a holiday, exhaustion, a life.

Theo believed, when he came to me, that he had failed at five diets. What actually happened is that five diets had each successfully produced weight loss for as long as he could execute them, and each had a design feature that made executing them indefinitely impossible for a man with a job, a marriage, and a social life.

The keto row is the one worth sitting with. It worked best. Eighteen pounds in four months is a genuinely good result. It also produced the largest regain, the worst side effects, and the worst triglyceride response — and it lasted the second-shortest time of anything he tried.

If you rank his diets by weight lost, keto wins. If you rank them by weight lost twelve months later, they all tie at zero.

This chapter is about why.

I have been building toward this chapter since Chapter 4, and it contains the threshold concept that Part II exists to deliver. Fair warning: it is going to disappoint people in both camps, and the disappointment is the finding.

🏃 Fast Track: §10.3 (the threshold), §10.5 (what the winners have in common), and §10.12 (how to actually choose). Twenty-five minutes, and they're the three that change what you do.

🔬 Deep Dive: §10.2 (the head-to-head trials), §10.4 (why they converge), and §10.7 (personalization, and why it failed its best test) are where students and clinicians should spend real time.

10.1 The contenders

Six positions, stated as their advocates would state them. This matters — Chapter 2's rule is that you steelman before you strike.

Each of these has a mechanism that is at least partly real, and each has produced people who genuinely, dramatically improved. That's the whole difficulty — if any of them simply didn't work, this chapter would be short.

10.2 What happens when you actually compare them

The good news is that this has been tested properly, repeatedly, with randomization. Here's what the best trials found.

DIETFITS (2018) — the best single trial in this chapter

Christopher Gardner's group at Stanford randomized 609 adults to a healthy low-fat or a healthy low-carbohydrate diet for twelve months. Both groups received extensive dietary counselling — this was not a pamphlet study — and both were instructed to maximize vegetables, minimize added sugar and refined flour, and cook at home.

Result: no significant difference in weight loss between groups. Both lost roughly 5–6 kg on average.

And here is what makes DIETFITS the most important trial in the chapter: it pre-specified two tests of personalization. Participants were genotyped for patterns hypothesized to predict low-fat versus low-carb responsiveness, and their insulin secretion was measured.

Neither predicted who did better. The genotype patterns didn't work. Insulin secretion didn't work. §10.7 returns to this.

What did vary enormously was the within-group spread. Some people lost 30 kg; some gained. The variation within each diet was far larger than the difference between diets — which is the single most important number in this chapter.

A TO Z (2007)

An earlier Gardner trial randomizing women to Atkins, Zone, LEARN, or Ornish — spanning very low-carbohydrate to very low-fat. At twelve months, Atkins showed a modest advantage, and the differences between groups were small relative to the differences between individuals.

POUNDS LOST (2009)

811 adults, four diets varying fat, protein and carbohydrate systematically, two years of follow-up. No meaningful difference in weight loss between the four.

And the pattern across the literature

Network meta-analyses comparing named diets have generally found the same thing: differences between named diets are small at six months and smaller at twelve, and largely disappear by two years.

🔄 Check your understanding. DIETFITS found no significant difference between low-fat and low-carb, and enormous variation within each group. A low-carb advocate says: "That's because the low-carb arm wasn't strict enough." Assess.

Answer

It's a legitimate objection and it's mostly a criticism of a different trial.

What's fair: DIETFITS did not run a ketogenic arm. Participants were instructed to reduce carbohydrate substantially but were not held to nutritional ketosis, and adherence drifted upward over the year — as it does in every free-living trial.

What's not fair: the drift is the finding, not a flaw. Chapter 2's question is "compared to what, in whom, for how long?" — and a trial in free-living adults with intensive counselling is measuring what happens when real people attempt a diet, which is the question that matters for a real person. A metabolic-ward trial holding people in strict ketosis would answer a different and less useful question. (And when those have been run, at matched calories and protein, they haven't shown the advantage either — Chapter 4 §4.6 and Chapter 6 §6.8.)

The deeper point: "it wasn't strict enough" is unfalsifiable as usually deployed. Any null result can be attributed to insufficient adherence, indefinitely. Ask the advocate in advance what level of adherence would constitute a fair test, and whether such a trial exists.

10.3 🚪 The threshold: adherence beats composition

Here it is.

🚪 Threshold concept.

Across head-to-head trials, the differences between named diets are small and diminish with time. The variation between individuals on the same diet is far larger than the variation between diets. The best diet is the one you will actually keep doing.

This sounds like a platitude. It is a research finding, and it has teeth.

The before-and-after

Why this isn't a cop-out

The lazy reading of "adherence beats composition" is everything works, so do whatever. That's not what the evidence says, and it matters that it isn't.

Composition still constrains what's achievable. A diet that leaves you at 0.7 g/kg protein will cost you lean mass (Chapter 8). A diet supplying 10 g of fiber will underperform on satiety, lipids and glycemia (Chapter 11). A diet built on refined carbohydrate will make overeating easy (Chapter 7).

What the evidence says is narrower and more useful: among diets that are nutritionally reasonable, the differences in outcome are small, and the difference between doing one and not doing one is enormous. Composition sets the floor; adherence determines whether you ever stand on it.

💡 Aha moment. Notice what this does to how you evaluate a diet book. The interesting question stops being "is the mechanism true?" and becomes "what does this diet do to make itself followable, and what does it cost?"

Keto works by eliminating an entire macronutrient category, which removes decisions, removes most ultra-processed food, and raises protein — and costs you social eating, most restaurants, and Theo's 3 p.m. energy. Veganism works by eliminating a different category with the same decision-removal effect — and costs you B12 unless you plan, and a certain amount of family friction. The Mediterranean pattern works by addition rather than elimination — and costs the least, which may be why it has the best long-term trial evidence.

They are all elimination-or-addition strategies for producing an energy deficit and a food-quality upgrade. The macronutrient argument is a story each one tells about why its particular elimination is the correct one.

10.4 Why the trials converge

Four reasons, and understanding them is what stops "the trials show no difference" from sounding like a dodge.

1. The assigned diet is not the eaten diet. In every free-living trial, adherence drifts. A group assigned to 20% carbohydrate typically ends the year somewhere well above that. By twelve months, the groups' actual intakes are much closer together than their assignments, which mechanically shrinks any difference.

2. Dropout is large and non-random. Diet trials routinely lose 20–40% of participants, and the people who drop out are systematically those doing worst. This inflates the apparent success of every arm and compresses differences between them.

3. Every reasonable diet removes the same foods. Whatever the assigned macronutrient ratio, someone attempting it seriously stops drinking soda, stops eating pastries, stops most snacking, and starts cooking. The overlap in what actually changes is far larger than the difference in what was prescribed.

4. Energy balance is downstream of all of it. Chapter 4's threshold: every one of these diets works, when it works, by producing an energy deficit. They differ in how they get you there — appetite suppression, food elimination, satiety, decision reduction — not in whether they need to.

🧩 Productive struggle. Five minutes before reading on.

Two diets are compared in a twelve-month randomized trial. At the end, weight loss is nearly identical. Design a study that would detect a real difference between them, if one existed.

What would you have to control, measure, or change? And what would your design cost you?

What I'd say

To detect a genuine compositional difference you would need to eliminate the four confounders above:

1. Verify adherence objectively rather than by self-report — which means biomarkers, or providing all food, or metabolic ward confinement. Chapter 4 §4.7 explains why self-report can't carry this.
2. Prevent dropout, or handle it with intention-to-treat plus sensitivity analysis, and be honest that people who leave are not like people who stay.
3. Match everything else — total energy, protein, fiber, food processing level — so that the only variable is the one you're testing. Which, note, makes it no longer a test of "keto versus low-fat" as anyone practises them.
4. Run it long enough for hard outcomes, which for cardiovascular disease means decades.

The cost of your design is the point. Everything that makes the trial capable of detecting a compositional difference makes it less representative of how people eat. Metabolic ward studies are rigorous and last weeks; free-living trials are realistic and can't isolate the variable.

This is Chapter 1 §1.2's five walls, arriving in a specific form. And it's why the honest conclusion isn't "the diets are identical" — it's "under real-world conditions, the compositional difference is smaller than the adherence difference, and the study that could separate them cleanly would no longer be about real-world eating."

10.5 What the winners have in common

Here's the constructive half, and it's the actual answer to "what should I eat."

Take every dietary pattern with decent outcome evidence — Mediterranean, DASH, healthy plant-based, the diets in the National Weight Control Registry, the eating patterns of long-lived populations — and ask what they share.

They share almost everything, and it isn't the macronutrient ratio.

Read the last two rows together. The one thing every successful pattern has in common is a low proportion of ultra-processed food. The one thing they don't agree on is the macronutrient ratio.

That is the answer, and it is Chapter 22's chapter and Chapter 37's conclusion arriving early.

10.6 Diet by diet, honestly

Verdicts, with what each gets right.

Low-carbohydrate and ketogenic

Gets right: removes most ultra-processed food and sugar-sweetened beverages · raises protein · genuinely suppresses appetite in many people · real evidence for glycemic control in type 2 diabetes (Chapter 26) · produces fast early results that sustain motivation.

Gets wrong: the metabolic advantage claim (Chapter 4 §4.6, Chapter 6 §6.8) · early loss is substantially glycogen and water (Chapter 4 §4.9) · high dropout in longer trials · restricts a category the evidence doesn't require you to restrict.

Best for: people with type 2 diabetes or significant insulin resistance, under supervision; people who genuinely find it suppresses appetite; people who prefer rules to moderation.

Worst for: endurance athletes (Chapter 6 §6.2); people with a rich social food life; anyone who finds restriction psychologically risky.

Low-fat

Gets right: fat is energy-dense, so reducing it reduces energy density · plant-based low-fat patterns have decent cardiovascular evidence.

Gets wrong: the 1980s implementation replaced fat with refined carbohydrate (Chapter 9 §9.5) · low-fat per se is not the active ingredient · very low fat diets can be hard to make satiating.

Best for: people who like high-volume, high-fiber eating; some cardiovascular contexts.

Worst for: people who find low-fat food unsatisfying — which was Theo in 2016, hungry for fourteen months.

Mediterranean

Gets right: the best long-term randomized evidence of any named pattern (PREDIMED, Chapter 9 §9.9) · additive rather than eliminative, so it's easy to adhere to · nutritionally complete · culturally flexible · not low-fat.

Gets wrong: relatively little. The main honest caveat is that "Mediterranean diet" is loosely defined and the version in trials is more specific than the version in magazines.

Best for: almost everyone, as a default. Worst for: people who need faster feedback to stay motivated, since it produces slower weight change.

Vegan and whole-food plant-based

Gets right: high fiber · high phytochemical intake · strong environmental case (Chapter 36) · cohort evidence from vegetarian populations is favourable · ethical coherence for those who hold it · some trial evidence for cardiovascular risk factor improvement.

Gets wrong (or requires attention): B12 is non-negotiable and must be supplemented · iron, zinc, iodine, calcium, omega-3 all need planning (Chapters 13, 14) · protein needs the 10–20% upward adjustment (Chapter 8 §8.8) · "vegan" includes both lentils and fried potato crisps.

Note the healthy-user problem: the cohort evidence for vegetarians comes substantially from populations who also smoke less, drink less, and exercise more. Chapter 2 applies.

Paleo

Gets right: eliminates ultra-processed food, added sugar, and most refined grain · emphasizes vegetables, meat, nuts · genuinely improves diet quality for most people who adopt it.

Gets wrong — and this is the interesting part: the premise.

🔬 Claim → Evidence → Verdict

The claim: "The paleo diet reflects what our ancestors evolved to eat. Grains, legumes and dairy are evolutionarily novel and we're not adapted to them."

Where it comes from: A real and serious idea — evolutionary mismatch — plus the genuine observation that agriculture is recent on an evolutionary timescale, and that many modern foods have no ancestral analogue.

What the evidence actually shows: The premise doesn't survive contact with anthropology. There was no single Paleolithic diet — hunter-gatherer diets varied enormously by latitude, season, and group, ranging from very high animal intake to heavily plant- and tuber-based. Contemporary hunter-gatherer populations eat substantial quantities of tubers, and in some cases honey in remarkable amounts. Modern plants and animals are not the ancestral ones — domesticated fruit is larger and sweeter, and modern livestock is fatter than wild game. And humans have demonstrably continued evolving since agriculture: lactase persistence and amylase gene copy number variation are both well-documented adaptations to agricultural foods.

Meanwhile, legumes and whole grains are among the most consistently health-associated foods in the literature (Chapter 7 §7.9).

📉 Evidence quality: The anthropological premise is weakly supported. The diet itself performs comparably to other reasonable diets in trials — because of what it eliminates, not because of the evolutionary story.

Verdict: 🟠 Probably false as a rationale. The diet is fine and the reason given for it isn't — which is a distinction worth being able to make, because it applies to a great deal of nutrition advice.

Carnivore

Gets right: it is an extremely effective elimination diet, and elimination diets genuinely help some people with gastrointestinal or autoimmune symptoms identify triggers · it removes all ultra-processed food, all added sugar, and all alcohol in one move · it is very high in protein · the testimonials are real experiences.

Gets wrong: essentially everything else.

🔬 Claim → Evidence → Verdict

The claim: "Humans thrive on an all-meat diet. Plants contain defensive compounds — lectins, oxalates, phytates — that damage us. Eliminating them resolves autoimmune and digestive disease."

Where it comes from: Genuine facts, maximally extended. Plants do contain defensive compounds (Chapter 7 §7.9). Elimination diets genuinely do identify triggers. And the personal accounts are often dramatic and sincere — people with debilitating symptoms who improved markedly.

What the evidence actually shows: There are no long-term controlled trials of a carnivore diet. What exists is testimonials (Chapter 2, rung 3), a survey of self-selected enthusiasts, and mechanistic argument. The short-term improvements reported are entirely consistent with what any aggressive elimination diet does — removing ultra-processed food, alcohol, added sugar, and any individual trigger simultaneously (Chapter 3's bundled-intervention problem). The diet contains zero fiber, which conflicts with one of the better-supported bodies of evidence in nutrition (Chapter 11). And the long-term consequences are simply unknown, because nobody has studied them.

📉 Evidence quality: Rung 3 only. No controlled trials of any duration.

Verdict: ⚗️ Untested — deliberately, rather than ❌. There is no evidence it works long-term and no evidence it doesn't, because the studies don't exist. What can be said with confidence is that people reporting improvement are reporting something real, and attributing it to the wrong variable — and that anyone claiming this diet is proven is claiming something that could not currently be true.

⚠️ Anyone considering an extreme elimination diet for suspected food-related symptoms should do it with a dietitian, systematically, and with testing first — Chapter 28's rule: test before you eliminate.

10.7 Personalization: the test it failed

The obvious rescue for the diet wars is: maybe different diets suit different people, and we just need to match them.

This is an excellent hypothesis and DIETFITS tested it directly.

The trial pre-specified two candidate predictors — genotype patterns hypothesized to indicate low-fat versus low-carbohydrate responsiveness, and insulin secretion, on the theory that high-insulin-secretors would do better on low carbohydrate.

Neither predicted who succeeded. Not the genotype patterns, not the insulin response.

This is a genuinely important negative result and it deserves care in both directions:

What it doesn't show: that personalization is impossible. It tested two specific hypothesized predictors, not the general idea.

What it does show: that the two most plausible and most commercially promoted predictors, tested prospectively in a well-designed trial, did not work — and that anyone currently selling you a DNA-based diet match is selling something that has been tested and has not delivered.

🔬 Claim → Evidence → Verdict

The claim: "There's no one-size-fits-all diet. A DNA test can tell you whether you're a low-carb or low-fat responder."

Where it comes from: The first sentence is true and important — the within-group variation in DIETFITS was enormous. The second sentence is the commercial claim built on it.

What the evidence actually shows: The variation is real; the predictability is not established. DIETFITS pre-specified and tested genotype patterns and found no effect on outcome. Direct-to-consumer nutrigenomic tests generally rest on small numbers of variants with modest individual effects, and their diet recommendations have not been prospectively validated.

📉 Evidence quality: Rung 6, pre-specified, negative.

Verdict: ❌ Not supported. The premise is right and the product doesn't follow from it. Chapter 35 takes the whole personalization market apart.

So what does predict who does well? Mostly boring things: whether they attended sessions, whether they cooked at home, whether their household was supportive, whether they had time. Adherence predictors, not biological ones — which is §10.3 arriving from a different direction.

10.8 The two-year problem

Everything above is about twelve months. Extend the window and the picture changes in a way that is rarely reported honestly.

Typical trajectory across essentially every diet studied: rapid loss for three to six months, a plateau around six to twelve months, and gradual regain from twelve to twenty-four months and beyond. Average weight at two years is commonly a fraction of peak loss.

This is not a failure of any particular diet. It's the pattern, and it's Chapter 4 §4.6 — adaptation, NEAT suppression, and hormonal changes that persist — arriving on schedule.

Two things are worth saying about it.

First, "a fraction of peak loss" is not zero, and modest sustained loss carries real benefit — improvements in blood pressure, glycemia, lipids, and function that persist even when weight partially returns. Look AHEAD, a large intensive lifestyle trial in type 2 diabetes, did not reduce its primary cardiovascular endpoint — but produced improvements across a range of secondary outcomes.

📊 Diagram (described). Picture a graph with months across the bottom (0 to 24) and weight change up the side, going down from zero.

Now draw four lines, one per diet — low-fat, low-carb, Mediterranean, plant-based — and here is the shape every one of them makes.

Months 0–3: all four drop steeply. The low-carb line drops fastest, and it is genuinely ahead — though a meaningful part of its early lead is glycogen and its bound water (Chapter 4 §4.9).

Months 3–6: all four begin to flatten. The gaps between them are at their widest here, which is also exactly when most diet books stop reporting.

Months 6–12: the lines converge. Not because the slow ones speed up, but because the fast ones slow down and start to turn. The low-carb line's early advantage erodes.

Months 12–24: all four curve back upward, roughly in parallel. By twenty-four months they are bunched together well above their trough and — this is the finding — close enough that no reasonable trial could distinguish them.

Two features are worth staring at. First: the widest separation happens early, and that's the window every marketing claim is drawn from. "Lose more weight on X" is almost always a statement about month three. Second: the shape is the same for all four lines. The regain is not a property of any particular diet — it's Chapter 4 §4.6's coupled system arriving on schedule, and it will arrive for whichever line you're standing on.

Now add a fifth element: a wide grey band around every line, representing the within-group spread. That band is taller than the distance between the lines. Which is the whole chapter in one image — the person matters more than the diet.

Second, the people who maintain do exist, and Chapter 24 examines what they do. The short version: they didn't find a better diet. They found a version of eating they could keep doing, and they kept doing it after they stopped calling it a diet.

🔄 Check your understanding. A friend loses 20 lb on keto in four months, then regains 25 over the following year. They conclude: "My metabolism is wrecked from keto." Using this chapter and Chapters 4–6, give the four-part answer.

Answer

1. A meaningful share of the 20 lb was never fat. Carbohydrate restriction depletes glycogen, and each gram of glycogen carries roughly three grams of water (Ch 4 §4.9, Ch 6 §6.2). Some of the "loss" and a good part of the "regain" is that water returning — it looks alarming on a scale and isn't fat.

2. The regain curve is universal, not keto-specific. §10.8's diagram: every diet studied shows rapid loss → plateau → gradual regain. Attributing the shape to the particular diet mistakes a property of the system for a property of the intervention.

3. Adaptation is real and modest (Ch 5 §5.6) — NEAT falls, RMR drops somewhat beyond what the smaller body predicts, ghrelin rises and leptin falls and stay shifted. That's hundreds of calories, not a wrecked metabolism, and it happens on any deficit.

4. The diet ended, and nothing replaced it. This is the actual answer. They went from a structured regime that removed an entire food category to no structure at all, on a single day, with elevated appetite. §10.9: it didn't stop working, it stopped being executable — and nobody had a plan for day one after.

What I'd say to them: "Nothing is wrecked. What happened is what happens, and the useful question isn't which diet to try next — it's which two things from the keto period you actually liked and could keep."

10.9 Why diets "fail" — and why that's the wrong word

Return to Theo's table.

Not one of his diets stopped producing results. Every one stopped being executable.

"The diet failed" and "I failed the diet" are both wrong framings, and the second one does real damage. The accurate framing is: the diet was badly matched to the life it had to survive in.

Consider what each of his five actually demanded:

That last row deserves attention. A regime whose rules specify that a single lapse restarts the count is a regime designed to end. It converts one bad day into total failure, which is the what-the-hell effect (Chapter 4 §4.10) written into the terms and conditions.

Theo called stopping on day thirty-one "failing." He had completed thirty-one consecutive days of two daily workouts and a self-selected diet. A regime that classifies that as failure has a defective scoring system, and he'd internalized it as information about himself.

10.10 Why the argument is so bitter

One more thing needs explaining, because it's genuinely puzzling: why is this argument so unpleasant?

Nobody gets abusive about whether to take vitamin D. People get abusive about keto. The diet wars are conducted with a hostility that is wildly out of proportion to the effect sizes involved — we are, per §10.2, arguing about differences that trials can barely detect.

Three things are happening, and recognizing them defuses a lot.

1. A diet is an identity, not a behaviour. People do not say "I am currently reducing my carbohydrate intake." They say "I'm keto." "I'm vegan." "I'm carnivore." The diet becomes a category of person, complete with an in-group, a vocabulary, a set of shared enemies, and a conversion narrative.

Once that's happened, evidence against the diet is evidence against you — and people defend identities far more vigorously than they defend hypotheses. This is Chapter 1 §1.6's myth shapes operating at full strength: restriction as virtue, purity and contamination, the suppressed truth.

2. Everyone involved has a genuine success story. This is what makes it intractable rather than merely heated. The keto advocate really did lose forty pounds. The vegan really did reverse their lipid panel. The carnivore really did have their joint pain resolve. Nobody is lying, and each of them has rung-3 evidence about themselves that feels far more compelling than any trial.

Chapter 2's whole argument is that a personal result doesn't generalize. Chapter 2's whole argument is also almost impossible to accept about your own transformation.

3. There's a conversion cost. Adopting a restrictive diet is expensive — socially, financially, and in effort. Someone who has spent two years explaining their food choices at every dinner has paid a great deal for a belief, and the more you've paid for something the harder it is to revise. That's not stupidity; it's an entirely normal human response to sunk costs, and it operates on all of us.

💡 Aha moment. Notice that this chapter's conclusion is the one that offers the least identity. "Adherence beats composition, pick something reasonable you'll keep doing" gives you no in-group, no enemy, no vocabulary, and nothing to post.

That's a real disadvantage in an attention economy, and it's a large part of why the boring answer loses the public argument while winning the trials. A conclusion that can't be worn as a badge travels badly — which is Chapter 1's media analysis arriving one last time, in the place it matters most.

And the practical consequence: if you're trying to change someone's mind about a diet, you are not having an argument about evidence. You are asking them to give up a membership. That's a much larger request, it usually fails, and the only thing I've seen work is asking about their last column — what ended it, and what they'd do if it ended again.

10.11 The ones we didn't cover, briefly

Six positions isn't the whole field. Quick verdicts on the rest, so you're not left wondering.

DASH (Dietary Approaches to Stop Hypertension). Not really a diet-war combatant, and it has excellent trial evidence — the DASH and DASH-Sodium trials produced substantial blood pressure reductions, and it's one of the best-evidenced dietary patterns in existence. High in vegetables, fruit, whole grains, legumes, low-fat dairy; low in sodium, sweets, red meat. It looks a great deal like the Mediterranean pattern with more dairy and an explicit sodium target. Chapter 26 covers it properly. If you have high blood pressure, this is your pattern.

MIND (a Mediterranean-DASH hybrid aimed at cognitive decline). Promising observational data, and a randomized trial has since produced more modest results than the cohort work suggested — a familiar pattern by now. Reasonable to follow; don't expect it to prevent dementia on its own.

Flexitarian / "mostly plants." Not a protocol, which is its strength. Most of the benefit of a plant-based pattern with none of the B12 planning problem or the social friction. For a lot of people this is the honest recommendation and it's unmarketable because it has no rules.

Whole30 and similar 30-day eliminations. Effective as a short-term reset and as a bundled intervention (Chapter 3, Case Study 1) — and by design they end, which means the question is always what happens on day 31? Useful as a diagnostic elimination if done deliberately; not a diet.

"Clean eating." Not a dietary pattern but a moral vocabulary applied to food, and Chapter 20 and Chapter 34 both take it seriously — the first for what the labels mean, the second for what the framing costs. The word "clean" implies its opposite, and that's the problem.

Intermittent fasting. Deliberately deferred to Chapter 21, because it's a timing strategy rather than a compositional one and it deserves its own treatment. Short preview: it works for some people, largely by reducing intake, and the mechanistic claims outrun the human evidence.

🧾 Cost check. The diet wars have a price dimension that almost never appears in the argument. Rough weekly cost for one adult, mid-range shopping:

Pattern	Weekly	Annual	Driver
Beans-and-grains plant-based	~$35–$50	~$1,800–$2,600	Legumes and grains are the cheapest calories and protein available
Mediterranean	~$55–$75	~$2,900–$3,900	Olive oil, fish, nuts, produce
Standard mixed diet	~$60–$80	~$3,100–$4,200
Low-carbohydrate / keto	~$85–$120	~$4,400–$6,200	Meat, cheese, nuts, and specialty replacement products
Carnivore	~$100–$160	~$5,200–$8,300	Meat is the most expensive food category by energy

Two things fall out. The diets with the best long-term evidence are among the cheapest, and the two most restrictive are the most expensive — by a factor of two to four over a plant-forward pattern.

And this connects to the Ortiz-Lindqvist ledger (Chapter 7, Case Study 1). At $6.43 per person per day, a keto or carnivore diet for a family of four is not a lifestyle choice they are declining — it's arithmetic they cannot do. A large fraction of the diet wars is being conducted between people for whom food cost is not a binding constraint, which is worth noticing when the advice reaches someone for whom it is.

10.12 How to actually choose

A decision framework, since the evidence won't hand you a winner.

Step 1 — Establish the floor. Whatever you choose must deliver: adequate protein (Chapter 8), adequate fiber (Chapter 11), adequate micronutrients (Chapters 13–14), and low ultra-processed food (Chapter 22). Any diet failing these is disqualified regardless of its macronutrient story.

Step 2 — Ask the adherence questions. These predict outcome better than any mechanism:

• Would I still be doing this in a year? Be honest. Not "could I," but "would I."
• Does it survive a bad week? A regime that only works when everything is going well fails exactly when you need it.
• Can I eat with other people? The most common failure point in Theo's table.
• Can I afford it, in money and time? (Chapter 7's ledger; Chapter 32.)
• Does it require me to become a different person?
• Does it have a reset-on-failure rule? If so, reject it.

Step 3 — Match the mechanism to the person, not to the biology. Some people do better with rules (clear categories, no decisions) and some with flexibility. That's a temperament question and it's a better predictor than any genotype.

Step 4 — Prefer additive to eliminative. Chapter 7's principle. "Add beans, add vegetables, add protein at breakfast" outlasts "remove carbohydrate," in general and in my clinic.

Step 4b — Run the reversibility test. Before starting anything, answer one question in writing: "What is my plan for the day this ends?"

Every diet ends. Some end at a target, some at a holiday, some at a bad week. Theo's five all ended, and in every case the answer to that question was nothing — which is why the regain was total. A regime with no exit plan is not a plan; it's a suspension of normal eating with an undefined resumption date, and the resumption is where the outcome is actually determined.

A good answer looks like: "When this ends, I keep the higher protein and the vegetables, drop the strictness, and expect my weight to settle two or three kilograms above the trough." A bad answer is "I won't stop" — because you will.

Step 5 — Default to Mediterranean if you have no strong preference. Best long-term randomized evidence, additive rather than restrictive, nutritionally complete, culturally adaptable, and it doesn't require you to give up a food category.

🍽️ On your plate — what Theo did. Not a diet. He didn't name it, and I think that mattered.

He made four changes (Chapter 4): beer from five nights to two, two of four coffees black, measure the cooking oil, sit away from the office bowl. Then three more over the following months: protein to 154 g, fiber from 14 g toward 38 g, and sardines twice a week.

That's it. No macronutrient ratio. No category eliminated. Nothing his wife had to cook differently.

Nine months: 96 kg → 89 kg, triglycerides 186 → 118, HbA1c 5.9% → 5.6%.

Slower than keto. He lost 18 lb in four months on keto and 15 lb in nine months on this. And at the eighteen-month mark he was still doing three of the seven changes and had not regained — which is the only comparison that has ever mattered.

⚠️ When to see a professional. Some of the diets in this chapter are genuine clinical interventions in specific circumstances and need supervision rather than enthusiasm. Very low-carbohydrate approaches with diabetes medication require dose adjustment — insulin and sulfonylureas can cause hypoglycaemia when carbohydrate drops suddenly, and SGLT2 inhibitors carry the euglycemic DKA risk from Chapter 6 §6.8. Any restrictive diet in pregnancy or breastfeeding needs a dietitian. Any elimination diet in a child needs supervision, because growth is at stake and because a child cannot consent to a nutritional experiment. And any diet undertaken by someone with a history of disordered eating should be discussed with a clinician first — restriction is both the intervention and the illness, and Chapter 34 covers why that matters more than anything else in this chapter.

What we don't know

We don't know how to predict who will adhere to what. This is the single largest gap in weight management, and it's a psychology question that nutrition research is poorly equipped to answer. We know adherence dominates; we cannot yet match a person to a pattern in advance, and we mostly find out by trying.

We also don't have good very-long-term randomized comparisons — five years, ten years — with hard outcomes, for any of these diets. PREDIMED is the closest thing and it's one trial in one population. Every confident claim about which diet is best over a lifetime is an extrapolation.

Spaced Review

Answer before reading on.

1. (Chapter 4) Why does every diet in this chapter produce weight loss when it works, regardless of composition?

Because they all produce an energy deficit — the accounting identity (Ch 4 §4.2). They differ in the mechanism by which they get you there: appetite suppression, food elimination, satiety, decision reduction. Composition is a delivery mechanism, not an alternative to energy balance.

2. (Chapter 9) The Mediterranean pattern has the best long-term randomized evidence and is not low-fat. What does that tell you about the diet wars' central framing?

That the framing — how much of each macronutrient? — is the wrong axis. PREDIMED's Mediterranean arms were higher in total fat than the control arm advised to reduce fat, and did better. The variable that mattered was food quality and pattern, not ratio.

3. (Chapter 2) Cohort evidence favours vegetarian populations. Name the specific problem, and say what would strengthen the inference.

Healthy-user bias (§2.2). Vegetarian cohorts — particularly the Adventist populations that supply much of this data — also smoke less, drink less, exercise more, and differ in income and education. What would strengthen it: convergence with trial evidence on intermediate outcomes, dose-response, and consistency across populations with different confounding structures (§2.10).

Project Checkpoint: Your Diet-History Autopsy

Component ten. Twenty minutes, and it's the one readers report finding most uncomfortable and most useful.

Build Theo's table for yourself.

Include everything: named diets, "eating clean," a gym phase, cutting out sugar, a fasting window, a January. If you've never dieted, list any deliberate eating change you've attempted.

Then answer four questions in writing.

1. Look only at the last column. How many ended because they stopped producing results, and how many because they stopped being possible? For almost everyone the answer is: none, and all of them.

2. What was the specific mismatch each time? Hunger? Other people? Cost? Time? Social eating? Boredom? Energy? A rule that guaranteed eventual failure?

3. Is there a pattern? Most people find one — the same constraint ending three different attempts. That constraint, not the diet, is your actual problem, and it's the thing to design around.

4. What language did you use? Go back over what you wrote. Did you say "I failed," "I cheated," "I fell off"? Or did you say "it didn't fit"? Circle the moral vocabulary.

Theo wrote "failed" next to thirty-one consecutive days of two daily workouts. Notice yours.

Non-tracking alternative. Skip the table. Write three paragraphs instead: what you've tried, what made it stop, and what you'd need to be true for something to last. The insight is in the why it ended column either way.

Next checkpoint (Chapter 11): your fiber audit and a personal ramp plan — which, for most readers, is the single highest-yield change in Part II.

Chapter Summary

The trials: DIETFITS (609 people, 12 months, healthy low-fat vs. healthy low-carb) — no significant difference, and genotype and insulin secretion both failed to predict who did better. A TO Z — small differences. POUNDS LOST (811 people, 2 years, four ratios) — no meaningful difference. Network meta-analyses: differences small at 6 months, smaller at 12, gone by 24.

🚪 The threshold: adherence beats composition. The variation within each diet is far larger than the variation between diets. The best diet is the one you'll actually keep doing.

And the necessary qualifier: composition sets the floor — protein, fiber, micronutrients, ultra-processed food. Among nutritionally reasonable diets, differences are small.

Why the trials converge: assigned ≠ eaten · dropout is large and non-random · every reasonable diet removes the same foods · everything runs through energy balance.

What the winners share: whole food · vegetables · legumes · adequate protein · high fiber · no sugary drinks · little ultra-processed food · home cooking. They do not share a macronutrient ratio — no two of them agree.

This chapter's verdicts:

Why diets "fail": they don't. They stop being executable. "The diet failed" and "I failed the diet" are both wrong — the accurate framing is that the diet was badly matched to the life it had to survive in. And any regime with a reset-on-failure rule is designed to end.

How to choose: establish the floor → ask the adherence questions → match the mechanism to your temperament → prefer additive to eliminative → default to Mediterranean if you have no strong preference.

Why the argument is so bitter (§10.10): a diet is an identity, not a behaviour — so evidence against it is evidence against you. Everyone involved has a genuine personal success story that feels more compelling than any trial. And the conversion cost is high, which makes revision expensive. This chapter's conclusion offers the least identity of any position in the argument, which is a real disadvantage in an attention economy and no disadvantage at all in a clinic.

Cost matters and never gets mentioned: a beans-and-grains plant-based pattern runs roughly $1,800–$2,600 a year; carnivore runs $5,200–$8,300. The diets with the best long-term evidence are among the cheapest, and the two most restrictive are the most expensive — which is worth remembering before the advice reaches a household at $6.43 per person per day.

The one thing to remember: if you rank Theo's diets by weight lost, keto wins. If you rank them by weight lost twelve months later, they all tie at zero.

What's Next

The diet wars are over and nobody won, which leaves the question of what to actually do — and the answer runs through the two chapters that close Part II.

Chapter 11 is about the nutrient with the best evidence-to-attention ratio in the entire food supply. Almost nobody gets enough, the effects are large and consistent, and there is no industry promoting it because nobody can patent a bean. For most readers it's the single highest-yield change in this book.

Theo's fiber is 14 g against a target of 38. That gap is doing more damage than every macronutrient argument in this chapter combined.