A census with a movable border
The famous 10:1 claim combined a 1972 estimate of 100 trillion intestinal bacteria with a separate rough count of human cells. The 2016 model instead estimates 38.64 trillion bacteria for a specified 20 to 30 year old, 70 kg, 170 cm reference man, using stool density as a proxy for heterogeneous colon contents; its 25% standard error describes uncertainty in the mean estimate, its 53% population variation describes spread among people, and unquantified systematic bias lies outside both. Count only nucleated human cells and the same model returns about 10:1; count all cells, or compare an independent 2023 census, and the ratio moves again. This is a model estimate, not a measurement that applies to every person.
Put your hand on the estimate ↓Layer 1 · the operable base
The opening state repeats the historical numerator. Drag colon content toward 420 g, or load the 2016 reference values. Then remove cells without nuclei.
Bacterial cells per human cell
100.00 trillion bacteria ÷ 30.00 trillion human cells = 3.333
Human denominator: Sender et al. 2016 sensitivity model, all cells included.
Historical bacterial numerator loaded. Human cells still use the 2016 census.
The bacterial side is a product. At the 2016 reference settings, the live equation is:
The 2016 all-cell sensitivity denominator allocates the paper's rounded 30T total as 25.20T red cells (its displayed 84% share), 1.47T platelets, and a balancing 3.33T nucleated remainder. These extra digits expose the arithmetic, not measurement precision. Blood volume scales both blood-cell terms; hematocrit scales only red cells. This is not a new physiological census.
At the paper's grouped resolution, the 3.33T nucleated remainder is made visible as 0.330T nucleated blood cells, 0.600T endothelial cells, 0.085T glial cells, 0.026T dermal fibroblasts, and 2.289T across the remaining non-blood types. The last term is a remainder, not a separately measured category.
Sender et al. report colon content at 17% standard error and 25% coefficient of variation, stool bacterial density at 19% standard error and 46% coefficient of variation, the bacterial numerator at 25% standard error and 52% population variation, and the human total at 2% calculated uncertainty and 14% coefficient of variation.
Their final ratio is reported with 25% uncertainty and 53% population variation. Simple independent quadrature of the displayed rounded inputs gives a value being computed. The small mismatch is disclosed because rounded inputs and any dependence structure cannot reproduce an authors' calculation exactly.
Layer 2 · the model becomes the instrument
That is the right objection. It does not make the correction trivial. It tells us to ask which terms move the answer most, which study owns each term, and what the word cell includes.
Same 2016 evidence, different definition
The reconstructed numerator stays 38.64 trillion. The rounded 2016 nucleated denominator is 3.33 trillion in this ledger.
calculatingReconstructing the displayed-input ratio. The paper describes this as about 10:1.
Hatton et al. independently counted about 36 trillion cells in a 70 kg reference male in 2023. Do not silently splice its uncertainty into the 2016 model.
calculatingKeeping the rounded-total and displayed-input routes separate.
Each row perturbs one 2016 reference input by 20% while holding the others fixed. The bars show the resulting percent change in the all-cell ratio. Colon content and bacterial density enter the numerator directly. Blood volume and hematocrit matter because non-nucleated blood cells dominate the denominator.
Calculating the effect of the cell definition.
The old numerator
1,000 g × 1.00 × 10¹¹ cells/g = 100 trillion
Sender, Fuchs, and Milo traced the often-cited bacterial numerator to Luckey's sentence-long 1972 estimate. The memorable ratio also needed a separate rough human-cell denominator.
The revised compartment
420 g × 0.92 × 10¹¹ cells/g = 38.64 trillion
The colon dominates the bacterial count. Stool measurements proxy its spatially heterogeneous density, but the paper says the product may be an upper limit and that systematic bias remains unquantified.
The check
The live page multiplies mass by density. The independent route below does not call that product: it adds one gram's bacterial count 420 times. Published values anchor the result, and a deliberately wrong input must be rejected.
Computing the ratio-scale amount. It is not presented as a confidence interval.
Computing the ratio-scale amount across the paper's standard 70 kg male population model.
The paper says using fecal density as a proxy for heterogeneous colon contents may make the product an upper limit, with unquantified systematic error outside the stated uncertainty.
Free choices named: a 70 kg reference male; whether red cells and platelets count; whether the 2016 or 2023 human census supplies the denominator; a simple local blood-scaling rule; and control bounds that stay above zero. No setting here measures your body. No result says the microbiome is biologically unimportant.
Primary record
The original article record. The 2016 paper traces the 100-trillion intestinal-bacteria numerator to its one-sentence estimate.
The compartment arithmetic, 70 kg reference man, 1.3 all-cell ratio, about 10:1 nucleated-cell ratio, uncertainty, population variation, and proxy warning.
An independent aggregation of about 36 trillion cells for a 70 kg reference male, including about 29 trillion non-nucleated and 7 trillion nucleated cells.
The browser and offline verifier carry the same source inputs but use separately written arithmetic routes. The offline script also string-matches the visible figures, uncertainty labels, metadata, and poison result. Full method and re-run instructions live in research/the-ten-to-one-person.