Some claims survive honesty. You can state your uncertainty, label your opinions, invite correction, and collect real data — and still produce a guaranteed false positive, because your intervention reaches your measurement without passing through the biology. When that happens the result was decided before anyone enrolled. Here is the shape, four times.
This failure is different from the others in this series, and more dangerous, because sincerity is no defence against it. Every other error we’ve catalogued — the floating number, the missing sensor, the borrowed phenomenon — gets weaker when the person is careful and honest. This one doesn’t. A scrupulous investigator who names their assumptions, admits the evidence is thin, and gathers data in good faith can walk straight into it and come out holding a confident, replicated, completely spurious finding. It isn’t a reasoning failure at all. It’s a measurement failure, and good intentions have no purchase on it.
Start with the simplest instance, because once you see it you can’t stop seeing it. Most things we measure in blood are reported as a concentration — an amount divided by a volume. Milligrams per decilitre. Millimoles per litre. Nanograms per millilitre.
That means every one of those numbers has a denominator. And dehydration — or acute plasma volume shift from hard exercise — is an intervention on the denominator. Shrink the volume and the concentration rises with nothing whatsoever added to the numerator.
The concentration artifact is one route. There are three, and they share a structure: the intervention touches the number without going through the biology you’re trying to study.
One — the testosterone spike. Heavy lifting appears to raise testosterone acutely. Part of that rise is real. Part is plasma volume contraction from exercise, plus reduced hepatic clearance during the session. Same hormone, smaller volume, bigger number. The “spike” is partly an artifact of the ruler.
Two — ketones on a dry fast. Blood BHB is mmol/L. Dehydrate someone and the reading rises for identical ketone production. So a dry-versus-water ketosis comparison will find dry “faster” whether or not it is — and every ketone measure has the problem: blood, breath acetone, and especially urine strips, which are maximally concentrated exactly when you’re driest.
Three — lean mass on a dry fast. DXA estimates lean soft tissue assuming it is a fixed ~73% water. BIA estimates it from total body water. Dehydration violates both.
Four — protein breakdown. The valid method is nitrogen balance: urinary urea nitrogen, exactly as the classical starvation literature did it. Which requires urine, from someone who isn’t drinking. The instrument itself is disabled by the intervention.
Imagine the good-faith version: someone honestly says “there’s no research on this, I looked hard, this next part is just my opinion” — and then, admirably, proposes to gather data: have people record how long it takes to reach ketosis on a water fast versus a dry fast. Every instinct there is right. The result is still guaranteed. Hundreds of sincere reports will converge on “dry fasting reaches ketosis faster,” the consensus will look overwhelming, and it will be plasma volume. A false finding built by careful people is more persuasive than one built by careless ones — because the care is real, and everybody can see it.
The fix is almost embarrassingly simple, which is what makes it worth saying loudly: measure the denominator. Take haematocrit or haemoglobin at the same moment as the ketone reading. Red cells concentrate right alongside everything else, so plasma volume contraction shows up there directly. If haematocrit is up ~10%, roughly 10% of any concentration rise is dilutional rather than metabolic. Correct for it and the comparison becomes meaningful.
Skip it, and the study cannot fail — in the worst sense of that phrase.
Worth separating the two failures, because they’re independent. Even with perfect measurement, ketosis ≠ autophagy. The counterexample is sitting in plain sight: on a ketogenic diet you can hold 1–3 mmol/L BHB while eating — mTOR active, autophagy suppressed. High ketones and low autophagy, at the same time, in the same person. So the poll has a confounded instrument and a proxy that doesn’t track the outcome. Fixing the first still leaves you unable to answer the question in the title.
Before believing a measured effect — including one you measured yourself — ask: “could my intervention have moved this number without going through the biology I care about?”
Three specific follow-ups do most of the work. Is the reading a concentration? Then find the denominator and ask what the intervention did to it. Does the method rest on an assumed constant? Then ask whether the intervention moved that constant. Does the valid measurement require something the intervention removes? Then no amount of data collection will rescue it.
And the honest closing note, which is the same one every piece in this series ends on: a confounded measurement doesn’t prove the effect is absent. It proves you can’t see it. Dry fasting might reach ketosis faster for real metabolic reasons. The point is that this study — and every poll like it — will report that it does either way, which means the report carries no information. Not a refutation. A blank.
A study confirms itself when the intervention reaches the measurement without passing through the biology — and sincerity is no defence, because it’s a measurement failure, not a reasoning failure. Most blood values are concentrations, so they have a denominator — and dehydration is an intervention on the denominator: 24 units in 3.0 L reads 8.0/L; contract the plasma 20% and the same 24 units read 10.0/L. +25% on the meter, zero new molecules. Three routes: the concentration artifact (you changed the volume), the broken assumption (DXA assumes lean tissue is a fixed 73% water), and the disabled instrument (nitrogen balance needs urine from someone who isn’t drinking). Four instances, one shape — the testosterone “spike”, ketones on a dry fast, DXA/BIA lean mass, nitrogen balance — and they don’t all flatter the claim: two push up, one pushes down, one erases the reading, which is what makes it a broken instrument rather than a bias. So a good-faith citizen-science poll on dry-vs-water ketosis will find dry “faster” whether or not it is. The fix is cheap: measure haematocrit alongside — up ~10% means ~10% of the rise is dilutional. The limit that survives the fix: ketosis was never a marker of autophagy (a keto diet gives 1–3 mmol/L BHB while eating, mTOR on, autophagy off). And the discipline: a confounded measurement doesn’t prove the effect is absent — it proves you can’t see it.
This article is for educational purposes only and is not medical advice, diagnosis, or treatment, and does not recommend any fasting protocol, diet, or self-experiment for any individual. It examines the structure of measurement and inference in publicly circulated claims, not the character or practice of any named individual. Nothing here should be read as encouragement to undertake self-experimentation: extended dry fasting is potentially lethal, survival without water is typically measured in days and varies with temperature, activity, and renal function, and dehydration causes acute kidney injury, electrolyte disturbance, arrhythmia, and death without reliably signalling itself through how you feel. Risk is substantially higher with kidney disease, kidney stones, diabetes, cardiovascular disease, pregnancy, older age, eating-disorder history, and many medications — including GLP-1 receptor agonists, which blunt thirst as well as appetite and carry a documented acute kidney injury signal driven by volume depletion. The measurement principles described are general and simplified; figures are illustrative rather than clinical, and interpretation of any laboratory value belongs with a qualified physician who knows your individual situation.
This lesson relates to these health systems — health works as a connected system, not isolated topics.
Prerequisite: The Evidence Is the Injury