Chapter 35 ★ Personalized Nutrition

⚠️ A note carried forward from Chapter 34, because it applies to everything in this chapter.

⚠️ This chapter is about products that generate individual rules about food and encourage continuous self-monitoring. ⚠️ For some readers that is not a neutral thing to be sold, and §35.12 is about it.

⚠️ If Chapter 34 §34.7 or §34.9 described something you recognized, read §35.12 before the rest.

The Hook: Two people, one banana

⚠️ Two people eat an identical banana. Both wear a continuous glucose monitor.

⚠️ One shows a modest rise. The other shows a large one — larger than they get from a slice of white bread.

⚠️ This is real. It is reproducible. It is not a measurement artefact, and it is genuinely interesting.

⚠️ It is also the single finding on which an entire commercial sector has been built, and the distance between what that finding shows and what is sold on the back of it is the subject of this chapter.

⚠️ Here is Theo Vasquez's version, and it is why he brought it to Yolanda Pierce.

⚠️ He had been sent an advertisement. It offered a test, a subscription and an app, and it made three claims:

⚠️ 1. Your body responds to food differently from everyone else's. ⚠️ 2. We can measure how. ⚠️ 3. And eating according to that measurement will make you healthier.

⚠️ The first claim is true. ⚠️ The second is partly true, and depends enormously on which test. ⚠️ The third has essentially not been demonstrated, and it is where all the money is.

⚠️ That structure — a real finding, a partly-real measurement, and an unevidenced outcome claim stacked on top — recurs in every product in this chapter, and learning to see the join is the skill worth having.

35.1 ⚠️ The genuinely impressive finding

⚠️ Say the good part clearly and first, because a chapter that opens sceptical is a chapter nobody believes when it concedes something.

⚠️ In 2015 a group at the Weizmann Institute in Israel, led by Eran Segal and Eran Elinav, published work in Cell that did something nobody had done at that scale.

⚠️ They continuously monitored the glucose responses of a large cohort of people across many thousands of real meals, and they found that individual postprandial glucose responses to the SAME food varied substantially between people.

⚠️ Not slightly. Substantially — to the point that a food producing a modest response in one person produced a large one in another, consistently, on repeat exposure.

⚠️ And the second half, which is the impressive part: ⚠️ they built a model incorporating microbiome composition, blood markers, anthropometrics, meal composition and activity, and it predicted individual responses better than carbohydrate content alone did. ⚠️ A short randomized comparison then showed that diets designed by the algorithm produced lower glucose responses than diets designed by expert dietitians using conventional criteria.

⚠️ That is a real result, it has been replicated in its core observation by independent groups, and it genuinely complicates the glycaemic index (Chapter 7).

⚠️ The GI was always a population average. This work showed how much individual variation that average conceals.

⚠️ What it establishes, stated precisely:

⚠️ That second row is the whole chapter, and §35.2 is about it.

35.2 ⚠️ The gap between a curve and an outcome

⚠️ Chapter 2's evidence ladder had eight rungs, and the distance between them is where this sector lives.

⚠️ The chain of reasoning the products rely on:

⚠️ Step 4 is doing all the work and is the least examined.

⚠️ In people with diabetes, glycaemic control is an established treatment target with hard outcome evidence behind it. That is not in question and Chapter 26 covered it.

⚠️ In people without diabetes, whether the size of a postprandial glucose excursion independently predicts disease — after accounting for what the person is actually eating, their weight, their activity, their fasting glucose and their HbA1c — is genuinely unclear. ⚠️ Glucose variability correlates with things that predict disease. Whether it is a cause, a marker, or largely an artefact of what and how much people eat is not settled.

⚠️ This is the healthy-user structure from Chapter 2 in a new costume: a measurable intermediate is observed to travel with an outcome, and an intervention on the intermediate is sold before anyone has tested whether moving it moves the outcome.

⚠️ It is beta-carotene's shape (Chapter 2), and it is the shape of very nearly everything in this chapter.

⚠️ And a practical consequence worth stating, because it is the most useful thing in the section:

⚠️ When these algorithms recommend foods, the recommendations converge substantially on what conventional advice already says — more fibre, more whole foods, fewer refined carbohydrates eaten alone, protein and fat alongside carbohydrate, and moderate portions.

⚠️ The personalization operates at the margin. It reorders a handful of foods within a dietary pattern that is not itself in dispute.

⚠️ Which is the finding that Chapter 37 is going to build on, so notice it now.

35.3 The genetic tests

⚠️ Direct-to-consumer nutrigenomic testing: send saliva, receive a report telling you which diet suits your genes.

⚠️ What the science supports:

⚠️ Monogenic conditions with dietary management — phenylketonuria, hereditary haemochromatosis, familial hypercholesterolaemia, coeliac-associated HLA types — are real, medically important, and are diagnosed through clinical pathways rather than through consumer kits.

⚠️ Common polygenic variation influencing response to ordinary diet is a different matter entirely.

⚠️ The pattern in that table: where the genetics is solid, you either already know, or it is a medical diagnosis rather than a lifestyle report.

⚠️ Where the product is selling you something you did not know, the genetics is weak.

⚠️ And the reproducibility problem, which is the single most damning practical finding:

⚠️ Investigations have repeatedly sent the same person's sample to multiple direct-to-consumer companies — and in some cases the same sample twice to the same company — and received materially different dietary recommendations. ⚠️ Different companies use different variant panels, different effect-size assumptions and different proprietary algorithms, and there is no agreed standard.

⚠️ A test that gives different answers to the same DNA is not measuring what it claims to measure.

35.3b ⚠️ The trial that tested the genetic premise properly

⚠️ The claim in §35.3's first row deserves a full treatment, because the test of it was unusually well designed and because it is the clearest example in this chapter of a hypothesis being killed honestly.

⚠️ The premise: ⚠️ that some people are constitutionally suited to low-carbohydrate diets and others to low-fat diets, and that a genetic test can tell you which. ⚠️ This is the single most commercially valuable claim in nutrigenomics — it is what most diet-gene products are ultimately selling.

⚠️ The trial — the DIETFITS study, run at Stanford by Christopher Gardner's group and published in JAMA in 2018 — did something most nutrition trials do not:

⚠️ The results, in order of how surprising they were:

⚠️ 1. Average weight change was not meaningfully different between the two diets. ⚠️ Chapter 10's finding, again, in a well-run trial.

⚠️ 2. Individual variation within each arm was enormous. ⚠️ Some people in each arm lost a great deal; some gained. The spread within one diet dwarfed the difference between the diets.

⚠️ 3. And the pre-specified genotype pattern did not predict who did better on which diet.

⚠️ 4. Nor did the insulin-secretion measure.

⚠️ Read results 2 and 3 together, because that pairing is the important part and it is routinely misreported.

⚠️ Individual variation is real and it is LARGE. The genetic test simply did not explain it.

⚠️ Why this is a good study to learn from rather than merely a useful result:

⚠️ It is a rare case of a commercially attractive hypothesis being specified in advance and then falsified by the people best placed to have found it if it were true. ⚠️ Had the pattern been sought after the data came in, something would almost certainly have been found — there were enough variants and enough outcomes to guarantee it (Chapter 3's multiple-comparisons problem).

⚠️ What the trial does NOT show, stated so it is not over-claimed:

⚠️ It does not show that genetics is irrelevant to nutrition — §35.7's monogenic conditions are real. ⚠️ It does not rule out that some future panel might predict something. ⚠️ And it tested two specific hypotheses, not every possible one.

⚠️ What it does show is that the specific claim being sold — this common genotype means you should eat this way — was tested carefully and did not hold.

⚠️ Products making that claim are being sold after the test, not before it.

35.4 Microbiome testing

⚠️ Chapter 27 gave the verdicts. This section extends rather than repeats them.

⚠️ What Chapter 27 established: the microbiome is genuinely important, and almost nothing sold to fix it works yet — the threshold, and both halves of it.

⚠️ What is specific to the TESTING products:

⚠️ That last row is the important one and applies to almost everything in this chapter.

⚠️ The advice is good advice (Chapter 27 §27.4). ⚠️ It is also the advice you would have been given without the test, and the test is what you paid for.

⚠️ Where microbiome analysis IS clinically useful, and it is a short list: ⚠️ specific diagnostic testing in clinical contexts, faecal microbiota transplant for recurrent C. difficile infection, and research. ⚠️ Not a consumer report suggesting foods.

35.5 ⚠️ Continuous glucose monitors in people without diabetes

⚠️ The fastest-growing product in this chapter, and the one that most requires a careful answer rather than a dismissive one.

⚠️ A CGM measures interstitial glucose continuously and displays it on a phone. In diabetes it is a genuine advance in care and this section is not about that.

⚠️ In people without diabetes, three honest positions, and I hold all three:

⚠️ 1. It shows you something real. ⚠️ Your responses do differ, they differ from other people's, and seeing your own data is a legitimately powerful demonstration of things this book has spent chapters asserting — that refined carbohydrate eaten alone behaves differently from the same carbohydrate eaten with protein, fat and fibre; that a walk after a meal changes the curve; that sleep and stress matter.

⚠️ 2. There is no evidence that acting on it improves outcomes in non-diabetic people. ⚠️ Step 4 of §35.2, unresolved.

⚠️ 3. And the interpretation problem is severe.

⚠️ That last row is where the harm is concrete. ⚠️ A person optimizing for a flat curve will reduce exactly the foods Parts II, III and V spent the book recommending — because fibre-containing carbohydrate raises glucose, and a slice of cheese does not.

⚠️ Optimizing a single intermediate marker at the expense of the whole diet is Chapter 2's error with a subscription attached.

⚠️ The honest summary: ⚠️ 🟡 Unclear, with genuine educational value for some people over a short period, no demonstrated outcome benefit, and a real risk of the interpretation problems above.

⚠️ If you are going to use one, three rules make it safer:

⚠️ Use it for a defined period as a learning exercise, not indefinitely · compare MEALS rather than FOODS, because context is most of the variance · and decide in advance that you will not eliminate a food category on the basis of a curve.

35.6 ⚠️ The rest of the market, briefly

⚠️ Grouped, because they share a structure.

⚠️ The blood-type diet deserves its one sentence of elaboration because it is an unusually clean example.

⚠️ A study assigned people to the different blood-type diets and then checked their actual blood types. Some diets produced better markers than others. ⚠️ The benefit was unrelated to whether the person had the blood type the diet was supposedly for — the diets that worked better were the ones containing more plants and less processed food.

⚠️ Which is the cleanest available demonstration of this chapter's recurring finding: the personalization did nothing, and the underlying dietary quality did everything.

35.7 ⚠️ What personalization actually has evidence for

⚠️ This is the constructive section, and it is deliberately unglamorous.

⚠️ Nutrition genuinely should be individualized. The question is on what basis — and the bases with evidence are the ones nobody sells a subscription for.

⚠️ That last row is the honest centre of the chapter.

⚠️ Chapter 10's threshold was that adherence beats composition. ⚠️ Which means the single most consequential piece of personalization available is matching the diet to what you will actually eat — to your culture, your budget, your kitchen, your schedule, your household, and what you like.

⚠️ That IS personalized nutrition. It requires no test, produces larger effects than anything in §35.3 to §35.6, and cannot be sold as a subscription — which is a substantial part of why it is not what the word has come to mean.

35.8 ⚠️ Why the responder question is genuinely hard

⚠️ A methodological section, because "some people respond differently" is true, is used to justify everything in this chapter, and is much harder to establish than it sounds.

⚠️ The problem: in any trial, some participants improve more than others. It is tempting to call them responders.

⚠️ But the spread of individual results in a trial arm reflects at least four things, and true individual response is only one of them:

⚠️ How you separate them: a REPLICATE CROSSOVER design. ⚠️ Each person does each condition more than once. If a person's response is genuinely individual, it should recur when they repeat the same condition.

⚠️ When this has been done properly, the evidence for true individual response varies by outcome. ⚠️ It is reasonably good for postprandial glucose — which is why §35.1 is a genuine finding — and much weaker or absent for many of the outcomes marketed on the responder concept.

⚠️ The practical rule: "some people respond differently" is a hypothesis, not a finding, until somebody has shown the same person responds the same way twice.

⚠️ Most personalized nutrition products have not shown that.

35.9 The n-of-1 experiment, taken seriously

⚠️ You can run a personal experiment. It can be genuinely informative, and it is much harder to do well than it looks.

⚠️ What an n-of-1 CAN tell you:

⚠️ Whether a specific, short-latency, clearly measurable effect happens in you — does dairy give you symptoms · does caffeine after four o'clock affect your sleep · does eating before training help · does a particular food produce a large glucose response.

⚠️ What it CANNOT tell you:

⚠️ Anything about long-latency outcomes. ⚠️ You cannot n-of-1 your cardiovascular risk, your cancer risk, or your lifespan, because the outcome takes decades and you have no control condition.

⚠️ How to do one properly, and each of these is where amateur attempts fail:

⚠️ Step 7 is the one nobody does, and it is what separates an experiment from a justification.

⚠️ The honest verdict: 🟢 for short-latency, self-evident, high-signal outcomes. ⚠️ ❌ for anything requiring years, and ⚠️ 🟠 for anything where the outcome is how you feel and you know which arm you are in.

35.10 ⚠️ Why the recommendations converge

⚠️ The chapter's central observation, and it is an empirical one rather than a rhetorical flourish.

⚠️ Collect the outputs of the personalized systems in this chapter — the glucose-response algorithms, the microbiome reports, the genetic panels, the app-based plans — and compare what they actually tell people to do.

⚠️ They converge, substantially, on the following:

⚠️ Eat more plants · eat more fibre · eat more variety · eat fewer refined carbohydrates on their own · eat protein, fat and fibre alongside carbohydrate · eat less ultra-processed food · move after meals · sleep · and keep portions moderate.

⚠️ Every one of those is in this book already, and none of them required a test.

⚠️ Where genuine personalization appears in these outputs, it is at the margin: the reordering of a handful of foods within a dietary pattern that is not itself in dispute.

⚠️ Three reasons this happens, and they are worth distinguishing:

⚠️ 1. Because the underlying dietary pattern really is that robust. ⚠️ The things that are good for most people are good for most people, and the individual variation sits on top of a large shared component.

⚠️ 2. Because the algorithms are constrained. ⚠️ A system that recommended cheese and cured meat over lentils, because they produce flatter glucose curves, would be recommending a worse diet — and the sensible products build in constraints that prevent it. ⚠️ Which means part of the "personalized" output is conventional advice, hard-coded.

⚠️ 3. And because the data does not support more than marginal personalization yet.

⚠️ Which sets up Chapter 37, and it is worth stating plainly here so it does not arrive as a surprise.

⚠️ The most sophisticated personalization available, applied honestly, produces recommendations that look remarkably like the boring ones. ⚠️ That is not a failure of personalization. It is information about how much of nutrition is shared.

35.10b ⚠️ Two reports, one person, side by side

⚠️ The abstract version of §35.3's reproducibility problem is easy to nod along to. Here it is concretely.

⚠️ Theo bought two products. He did not tell either about the other. Same person, same month, same body.

⚠️ Four things to notice, and the fourth is the one that matters most.

⚠️ 1. They contradict each other directly on the central question. ⚠️ One says reduce carbohydrate; the other says increase a major category of it. ⚠️ Both are confident, both are personalized, and both cite his data.

⚠️ 2. Where they agree, they agree on cheese. ⚠️ Because cheese does not raise glucose and does not appear in the carbohydrate-sensitivity panel. ⚠️ Two entirely different methods converged on recommending a food that most of this book has treated as fine in moderation and not as a foundation — which is §35.5's "flatter is better" problem arriving from two directions at once.

⚠️ 3. Both recommended supplements they sell. ⚠️ Six in total, at a combined cost comparable to several weeks of Chapter 32's food budget, and Chapter 13 and Chapter 16's verdicts apply to every one of them.

⚠️ 4. And where they agreed on something useful, they agreed on the boring things.

⚠️ Strip out the contradictions, the supplements and the colour codes, and what both reports actually told him was: eat more vegetables, more fibre, more variety, fewer refined carbohydrates on their own, and less ultra-processed food.

⚠️ He paid a substantial sum for two documents that disagreed about oats and agreed about vegetables.

⚠️ Yolanda Pierce's assessment, which is the most useful sentence in this section:

"Look at what they disagree about. That's the part you paid for."

⚠️ The disagreements are the personalization. The agreements are conventional advice, arriving with a colour code attached. ⚠️ And when the personalization from two products points in opposite directions, at least one of them is wrong — and you have no way of telling which.

⚠️ One fair qualification, because this comparison is not quite a controlled experiment:

⚠️ The two products measure different things and are not strictly answering the same question. ⚠️ A genetic panel and a glucose-response model could in principle both be right about different aspects of the same person.

⚠️ But that defence has a cost: if the outputs of two personalized systems can point in opposite directions about whether to eat oats, and both are correct within their own frame, then neither is telling the person what to do. ⚠️ And what to do is the thing being sold.

35.11 ⚠️ What would change my mind

⚠️ A section this book has not used before, and this is the right chapter for it.

⚠️ It would be easy to write this chapter as blanket scepticism, and blanket scepticism is not a position — it is a reflex. ⚠️ So here is what would move me, specified in advance.

⚠️ If those arrived, I would change the verdicts in this chapter, and I would expect to.

⚠️ The glucose-response work is a real scientific finding in a young field, and young fields improve. ⚠️ What I object to is not the science. It is the products sold years ahead of it.

35.11b ⚠️ Precision nutrition is a real research programme

⚠️ A distinction this chapter needs to draw explicitly, because failing to draw it is how scepticism turns into cynicism.

⚠️ "Personalized nutrition" as a consumer category and "precision nutrition" as a research programme are not the same thing, and they deserve different verdicts.

⚠️ Major national research initiatives are now doing this properly — recruiting large, deliberately diverse cohorts, collecting genomic, microbiome, metabolomic, continuous-glucose, dietary and environmental data together, and asking whether individualized prediction outperforms population guidance on outcomes that matter.

⚠️ That work is serious, it is well designed, and it will produce answers. ⚠️ Some of those answers will probably vindicate parts of this sector.

⚠️ What it has not yet produced is a result that justifies a consumer product, and the honest position is that the products arrived roughly a decade ahead of the evidence they cite.

⚠️ Two things worth watching for over the next decade, because they are where genuine progress is most likely:

⚠️ Metabolomic and proteomic signatures of dietary intake — objective biomarkers of what someone has actually eaten, which would fix the single largest methodological weakness in the whole of nutrition science (Chapter 3).

⚠️ And stratified rather than individualized recommendations — identifying subgroups for whom a different pattern genuinely works better, which is a far more tractable problem than a bespoke plan for each person and is where most of medicine's actual personalization lives.

⚠️ Notice that both of those would improve the population advice rather than replace it. ⚠️ The likeliest future of this field is better general guidance with a few well-evidenced subgroups inside it — not seven billion individual diets.

35.12 ⚠️ Self-monitoring, and who this is not neutral for

⚠️ Chapter 34 §34.7 listed the ways nutrition material can act as a vector. Read that list against this chapter's products.

⚠️ That last row is the specific problem, and it is worse than a generic diet.

⚠️ A conventional diet's rules can be argued with. A rule generated by a test of your own biology cannot easily be — ⚠️ it arrives with the authority of personal data, and disagreeing with it feels like disagreeing with your own body.

⚠️ For a person for whom restriction has become rule-governed (Chapter 34 §34.9), a device generating continuous personalized justifications for narrowing the diet is close to the worst available purchase.

⚠️ Signals that a monitoring tool has stopped being informative:

⚠️ You check it before deciding whether you are allowed to eat something · a reading changes how you feel about yourself · your diet has narrowed and has not widened · you have stopped eating with other people because of it · or you feel unable to stop wearing it.

⚠️ If those are recognizable, Chapter 34 §34.14 is the section, and taking the device off is a reasonable first step.

⚠️ And the general point, which applies to readers who have none of those concerns:

⚠️ Continuous data about a normal physiological process is not automatically useful. ⚠️ Watching your glucose rise after eating is watching your body work correctly, and there is a version of this technology that teaches people to be alarmed by health.

35.13 ⚠️ Who this chapter is for — and who it isn't

⚠️ The distinction that matters: personalization prescribed for a diagnosed condition is medicine. Personalization sold to a well person on the basis of a consumer test is a product, and the evidence standards for those two things are not the same.

🧾 What it costs

⚠️ The systemic-problems-as-purchases pattern, in its clearest form.

⚠️ Compare Chapter 32 §32.13: a nutritionally excellent week of food for one adult costs a small fraction of a CGM subscription.

⚠️ The person spending $2,000 a year discovering which foods suit them could have spent it on food. ⚠️ That is not a rhetorical flourish — it is the comparison the sector most needs made, and it is the same pattern this book has flagged since Chapter 16.

35.14 What we don't know, and how firmly I hold this

⚠️ What this field is bad at:

⚠️ Short trials · surrogate outcomes · comparison against no advice rather than against good advice · and a literature substantially produced by the companies selling the products.

⚠️ The last is not an accusation of dishonesty — it is a structural fact about who funds this research, and Chapter 3's framework for handling funded science applies unchanged.

⚠️ And the thing I am most confident of: this field will produce something genuinely useful, and it has not yet, and the products are being sold as though it had.

Spaced Review

From Chapter 2: ⚠️ What is the beta-carotene template, and how does §35.2 show this sector repeating it?

From Chapter 7: ⚠️ What is the glycaemic index, and how does §35.1's finding complicate it?

From Chapter 27: ⚠️ What was the microbiome threshold, and why does §35.4 not need to re-argue it?

From Chapter 28 §28.13: ⚠️ Why are IgG food sensitivity panels ❌?

From Chapter 10: ⚠️ What is the adherence threshold, and why does §35.7 call preference the biggest personalization lever?

From Chapter 16: ⚠️ What did Walt's supplement audit cost, and how does §35.13's table rhyme with it?

From Chapter 34 §34.7: ⚠️ What are the ways nutrition material acts as a vector, and how many does a CGM subscription commit?

Project Checkpoint: Price the Personalization

⚠️ Progressive project, Phase 5.

Step 1 — ⚠️ Take one product you have used, been offered, or are curious about.

Step 2 — ⚠️ Split its claims into the hook's three.

⚠️ What it claims your body does differently: __ ⚠️ What it claims to measure: __ ⚠️ What it claims will improve if you act on it: __

⚠️ Then find the join. Which of the three is doing the selling?

Step 3 — ⚠️ Run §35.11's six questions against it.

⚠️ Hard outcome? · Compared against GOOD conventional advice? · Durable beyond weeks? · Independently replicated? · Individual response demonstrated twice? · Does the test replicate?

⚠️ Count the yeses. Most products score zero or one.

Step 4 — ⚠️ Read its actual recommendations against §35.10's list.

⚠️ How many are already in this book? _ ⚠️ How many are genuinely individual? _

Step 5 — ⚠️ Cost it against food.

⚠️ Annual cost of the product: $______ ⚠️ Weeks of Chapter 32 §32.13's food that buys: ______

Step 6 — ⚠️ And do the honest personalization instead, which is free.

⚠️ My diagnosed conditions and allergies: __ (Ch 26, 28) ⚠️ My medications and their interactions: __ (Ch 16) ⚠️ My life stage and activity: __ (Ch 23, 25) ⚠️ My budget, kitchen, schedule and household: __ (Ch 31, 32) ⚠️ What I actually like eating: __ (Ch 10)

⚠️ That list is your personalized nutrition. It required no test and it will outperform anything in this chapter.

Chapter Summary

⚠️ The structure to learn: a real finding, a partly-real measurement, and an unevidenced outcome claim stacked on top. ⚠️ Every product in this chapter has that shape, and the skill is seeing the join.

⚠️ §35.1: individual glycaemic responses to identical foods genuinely differ, substantially, are reasonably stable within a person, and are partly predictable. ⚠️ That is real work, it has been replicated in its core observation, and it complicates the glycaemic index properly.

⚠️ §35.2: what has NOT been shown is that eating to lower those responses improves any outcome in people without diabetes. ⚠️ Step 4 of the chain is doing all the work and is the least examined — which is Chapter 2's beta-carotene template with a subscription attached.

⚠️ §35.3: DTC genetic diet tests are ❌ to 🟠, ⚠️ and the damning practical finding is that the same sample can return different advice. ⚠️ Where the genetics is solid you either already know, or it is a medical diagnosis rather than a lifestyle report.

⚠️ §35.4–35.6: microbiome reports have no reference standard and give converging generic advice; IgG panels, hair analysis and blood-type diets are ❌. ⚠️ The blood-type diet was tested directly: some diets worked better, and it had nothing to do with blood type.

⚠️ §35.5: CGMs in non-diabetic people are 🟡 — real educational value briefly, no outcome evidence, and severe interpretation problems. ⚠️ A postprandial rise is your body working, and "flatter is better" pursued honestly recommends avoiding fruit, legumes and whole grains.

⚠️ §35.7: what personalization actually has evidence for is allergy, coeliac disease, intolerance, diagnosed disease, medication interactions, life stage, athletic demand, documented deficiency — and PREFERENCE, CULTURE, BUDGET AND SCHEDULE, which is the biggest lever of all (Chapter 10).

⚠️ §35.8: "some people respond differently" is a hypothesis until someone shows the same person responds the same way twice.

⚠️ §35.10: the outputs converge — more plants, more fibre, more variety, fewer refined carbohydrates alone, less ultra-processed food, move after meals, sleep, moderate portions.

⚠️ Partly because the pattern is robust, partly because sensible algorithms hard-code conventional advice to stop themselves recommending cheese over lentils, and partly because the data does not yet support more than marginal personalization.

⚠️ §35.13: an enthusiastic first year runs $1,500–4,000. The person spending it discovering which foods suit them could have spent it on food.

What's Next

⚠️ Chapter 36 closes Part VI by going in the opposite direction from this one.

⚠️ This chapter looked as far into the individual as the technology allows and found that most of the answer is shared. ⚠️ Chapter 36 looks outward, at the systems that determine what is on the shelf, what it costs, what it is made of, and what it does to the planet that grows it.

⚠️ It is where the policy questions Chapter 32 §32.11 explicitly deferred finally get discussed — subsidy, marketing, taxation, labelling, land use and waste.

⚠️ And it carries a Learning Check-In 🪞, because Part VI has been making an argument across seven chapters and it is worth stopping to see whether it has landed before Part VII tries to assemble it.