A named diet is two interventions wearing one name: an exclusion (the foods it drops) and a distinctive inclusion (the foods it makes you eat more of). Its measured or felt benefit is a sum of the two, and the two often have opposite signs. The confound is that the diet’s identity — its memorable additions — collects credit that its exclusions earned.
Keto is “eat fat”; carnivore is “eat meat”; Whole30 and paleo are “eat real food.” Each name points at what the diet adds. But each one also silently deletes ultra-processed food, sugar-sweetened beverages, and refined-carbohydrate snacks — and the fabric holds that the deletion is a real lever while the signature addition frequently is not. A person who improves on such a diet, and a testimonial that credits the beef or the butter, are both committing the attribution the fabric already forbids for single foods, run at the level of a whole pattern.
No source decomposes a real diet into these two arms and measures their relative share, so the claim here is directional, not a quantity: the fabric’s evidence supports the exclusion arm and often runs against the distinctive addition. Which arm is larger for a given person is not something the fabric can weigh.
The decomposition
Split any named diet into two exposures and price them separately:
-
The exclusion arm. What the diet removes — usually ultra-processed food, added sugar, and sugar-sweetened drinks, because almost every popular diet removes those regardless of its ideology. The fabric rates this lever as real: a controlled-feeding trial had people eat markedly more on an ultra-processed diet at matched macros -> Ultra-Processed Food and Health Outcomes, and sugar-sweetened drinks carry a clean additive signal for diabetes and weight -> Free Sugars Intake. Whatever the diet is called, this is the arm the fabric’s evidence actually supports — strongest on intake and weight surrogates, with the hard-outcome ultra-processed signal still composition-confounded.
-
The distinctive-inclusion arm. What the diet’s name actually promotes — the beef, the butter, the fat, the exclusion of plants. This is the marginal exposure, the part that separates this diet from another equally junk-free one, and it is where the fabric’s directional evidence often runs the wrong way (saturated fat and LDL; a low-carbohydrate pattern raising LDL; fibre and plant foods removed with nothing substituted) -> Low-Carbohydrate vs Balanced-Carbohydrate Diets.
Why the comparator hides it
This is The Comparator Problem applied to a bundle, and the trick is in the default comparator. Set the diet against a standard junk-heavy Western diet and the exclusion arm dominates, so the whole bundle looks good — and its distinctive additions ride along on that verdict. Set it against the honest comparator — an equally junk-free diet that also keeps fibre, plants, fish and plant oils — and the exclusion arm is matched out on both sides of the contrast. What remains is the distinctive inclusion alone, judged on its own evidence. The diet’s apparent effect shrinks to the marginal exposure, and that is usually the part the fabric is least happy about.
The move is the same one the fabric already makes for single foods: an observed healthy population is not evidence for any one of its components, and the benefit of a fruit-and-vegetable pattern resists being pinned to any single compound -> Is the Food Category Doing Any Work, Fruit and Vegetable Intake and Health. Here the “population” is a diet and the “component” is its signature food.
Not every diet is pure removal — the discriminator
The confound is a diagnostic, not a verdict that every named diet is only its exclusions. A diet earns credit for its additions when a distinctive inclusion has its own evidence that survives the honest comparator. The Mediterranean pattern is the nearest contrast case: its hard-endpoint trial signal comes from named additions (olive oil, nuts) tested against a fat-reduction control -> Named Diet Programs Compared, Dietary Patterns. That control is an imperfect comparator, not the fully junk-free one this page defines, but it is non-trivial, because the control was not a junk-heavy diet either. So at least part of the credit sits with the addition rather than a junk removal alone — even though the trial cannot separate the oil from the nuts from the whole gestalt. The test is therefore per diet: does a distinctive addition survive a comparator that is not itself junk-heavy, or does only the exclusion?
Decision relevance
- Separate the two arms before crediting a diet. When someone reports a diet “worked,” ask what it removed and what it distinctively added, then ask which arm the evidence supports. Most of the time the answer is: keep doing the removal, and the branded addition is optional at best.
- The removal is the transferable lever. Because almost any diet can drop ultra-processed food and sugar, the benefit is reachable without adopting the ideology attached to it — which is the Layer-1 point that the big rock is the junk removal, not the diet’s name -> Layer 1 - Ranking Interventions for a Stratum.
- The distinctive addition is where harm hides. The marginal exposure that survives the matched-comparator is exactly where a diet can carry a route-(c) contraindication (a high-saturated-fat pattern for an atherogenic-lipoprotein responder), so it deserves the scrutiny the removal’s halo would otherwise buy it off.
- This is an open loop. The fabric can say which arm the evidence supports; it cannot observe whether a given person, on a given named diet, was better off — only whether a well-informed advisor would credit the removal or the addition.