CN-11 — Humans: the numbers, without the flattery

What you are reading. The human body as a set of measured quantities, with the flattery stripped out. Nearly every famous number about the human body — how many cells, how many neurons, how many bacteria — turned out to be an unsourced ballpark that survived by repetition until someone counted. This chapter carries the corrections with the receipt of the correction, because the correction is the lesson. Nothing here raises any UNI rung — a citation to biology is never a UNI gate.

---

The spine: three famous numbers, three unsourced ballparks

Before any physiology, look at what happened three separate times.

The folklore	What it traced back to	What measurement found
"10¹²–10¹⁶ cells in the body"	no proper reference (Bianconi et al. 2013 say so in their own abstract)	3.72 × 10¹³
"10 bacteria per human cell"	one back-of-the-envelope estimate from 1972	~1.3 : 1
"100 billion neurons, 10× more glia"	Herculano-Houzel could find no supporting references — in her words, "none other than ballpark estimates"	86 billion neurons, ~1:1 glia

Three independent cases, one failure mode: a number with no denominator, no method, and no source, repeated until repetition felt like evidence. Each fell the first time anyone counted with stated assumptions. This is what happens to any claim circulating without a falsifier attached — observed in the wild, three times, in the best-studied organism there is. If the headline numbers about us were folklore, treat the interpretations built on them as suspect until each is separately sourced.

The count, and the denominator nobody stated

Bianconi et al. (2013), Ann Hum Biol 40(6):463–471 summed cell counts organ by organ from the literature and reached 3.72 × 10¹³ cells for a standard adult — explicitly a starting point, not a closed result. Sender, Fuchs & Milo (2016), PLoS Biol 14(8):e1002533 recomputed: 3.0 × 10¹³ human cells for a 70 kg reference male (2% uncertainty, 14% CV across individuals) against 3.8 × 10¹³ bacteria (25% uncertainty, 52% SD), so B/H ≈ 1.3 with 53% variation. Bacterial mass: ~0.2 kg wet weight, ~0.3% of body weight. They traced every downstream citation of the old bacterial count to a single 1972 back-of-the-envelope: 10¹¹ bacteria per gram of gut content × 1 litre of alimentary capacity. That estimate was never wrong so much as it was never a measurement.

Now the part that matters. Of those 3.0 × 10¹³ human cells, ~84% (2.5 × 10¹³) are red blood cells — anucleate, no genome, essentially haemoglobin containers. Subtract them and you have ~5 × 10¹² nucleated cells. Against 3.8 × 10¹³ bacteria, that is ~8:1.

The old 10:1 folklore was approximately right — for a denominator it never stated. The revision from 10:1 to 1.3:1 is not a wrong number replaced by a right one; it is an unstated denominator. Both ratios are defensible; neither means anything until you say which cells you are counting. That is why "the 10:1 myth was busted" is itself a slightly dishonest telling.

And: each defecation event reduces colonic content by a quarter to a third. A ratio that shifts ~30% after a bowel movement is not a constant of your being — it is a standing inventory.

The energy budget — and the "100 W human"

Clarke & Sokoloff (Basic Neurochemistry, 6th ed., 1999) give the measured basal figure: a 70 kg man consumes ~250 mL O₂/min at rest. Convert it, showing the work, using the caloric equivalent of oxygen for a mixed diet (4.82 kcal/L O₂, RQ ≈ 0.82):

0.250 L/min × 4.82 kcal/L          = 1.205 kcal/min
× 1440 min/day                     ≈ 1,735 kcal/day
× 4184 J/kcal ÷ 86,400 s           ≈ 84 W          ← BMR

The other common route: ~2,000 kcal/day × 4184 J ÷ 86,400 s ≈ 97 W. But look at that input. 2,000 kcal/day is a dietary-labelling convention, not a measured expenditure — it carries no sex, age, body mass, or activity scope, and nothing in this chapter sources it. The 97 W is that convention converted; it is not a physiological measurement, and it is fenced accordingly in the table. Both land near 100 W, and the popular "you are a 100 W lightbulb" quietly conflates BMR with total daily expenditure — they differ by ~15%. Small point, whole method: state which quantity you meant, and where its number came from.

Specific power: 84 W / 70 kg ≈ 1.2 W/kg. Set that beside NA-08's cell budget — a mammalian fibroblast runs at ~100 W/kg. Your cells are individually ~100× more power-dense than you are, because most of your mass is fat, bone, matrix, and plasma, which metabolise little or nothing. Whole-organism specific power is diluted by structural mass — it is not a measure of how hard your cells work.

The brain, with the folklore removed

Azevedo et al. (2009), J Comp Neurol 513:532–541 counted with the isotropic fractionator: 86.1 ± 8.1 billion neurons and 84.6 ± 9.8 billion non-neuronal cells in the adult male human brain. The glia:neuron ratio is ~1:1, not 10:1. Two further facts cut against the usual framing:

• The cerebral cortex holds only 19% of your brain's neurons — Azevedo et al.'s own abstract: "With only 19% of all neurons located in the cerebral cortex" — while the cerebellum holds roughly 4× as many. The structure everyone points to when they say "this is what makes us us" contains about one neuron in five. And that fraction is not ours: in most mammalian species examined, including humans, the cortex holds ~20–25% of all brain neurons regardless of its relative size (Herculano-Houzel 2012). Note the two are different claims with different scopes — 19% is the human measurement, 20–25% is the cross-species regularity — and the regularity is the sharper anti-flattery point: our cortex is relatively enlarged and still carries the ordinary mammalian share.
• Synapses: 1.64 × 10¹⁴ in the neocortex — Tang et al. 2001, Synapse 41(3):258–273, unbiased stereology, and the scope is the number: "The average total number of synapses in the neocortex of five young male brains was 164 × 10¹² (CV = 0.17)." n = 5, one sex, one age band — a real measurement on a small denominator, not a species constant. The uncertainty around it is real too: textbook summaries span 0.6–2.4 × 10¹⁴, a ~4× spread. Anyone quoting a single flat synapse number is not quoting a measurement.

Energy — and a live 1.5× discrepancy that should not be averaged away. The brain is ~2% of body mass and takes ~20% of resting O₂ consumption (Clarke & Sokoloff). Both parts are real. But the absolute cost depends on the route:

CMRO₂ route:  20% × 250 mL O₂/min = 50 mL/min → ~347 kcal/day → ~17 W
Per-neuron route (Herculano-Houzel 2012):
              86 billion × 6 kcal/d per billion → 516 kcal/day → ~25 W

These disagree by ~1.5×. The second is a modelled extrapolation from a comparative per-neuron rate, not a direct human measurement. Carry both, name the method, do not split the difference.

What the 2%/20% number does and does not imply. It does imply neural tissue is expensive per gram: ~17–25 W over ~1.4 kg is ~12–18 W/kg, roughly 10–15× the body's average 1.2 W/kg. It does not imply human distinctiveness — per-neuron cost is conserved across primates, so the 20% share follows arithmetically from neuron count. Nor does it imply a larger metabolic share buys better cognition; no such relation is established here.

The rebuke: a scaled-up primate brain

This is the anti-flattery centrepiece, and it is Herculano-Houzel's actual finding, not a rhetorical flourish.

The human brain has about as many neurons as you would expect of a generic primate brain of its size. It is not an outlier on the primate scaling line — it sits on it. Azevedo et al. put it in their title: the human brain is an isometrically scaled-up primate brain.

The apparent specialness is real but relocated: primates scale differently from rodents, packing neurons more economically as brain size grows. Humans inherited that rule and are the largest-brained primate. The distinctive step happened at the clade, not at us. Nor do we top the raw charts — humans do not rank first in relative brain size, and at ~1.5 kg the human brain is two- to threefold smaller than an elephant's (Herculano-Houzel 2012). A cortex-neuron count is a number; it is not a warrant.

Scaling checks: where we fit, and where we plainly do not

Cross-ref NA-05 for the allometric machinery. Here is the check that matters, and it does not go our way.

Levine (1997), J Am Coll Cardiol 30:1104–1106 computed lifetime heartbeats across mammals: 7.3 ± 5.6 × 10⁸ beats per lifetime, over 15 species, excluding humans, described as roughly constant within an order of magnitude.

Look at that SD before looking at us. 5.6 on a mean of 7.3 is a coefficient of variation of ~77%. An "invariant" defended within an order of magnitude and varying by 77% is a loose invariant — the first honest reading, available before humans enter at all.

Now us. At 70 bpm for 80 years:

70 × 60 × 24 × 365.25 × 80 ≈ 2.9 × 10⁹ beats

That is ~4× the mammalian mean, ~4 SD above it. We are outside the band.

But do not stop at the flattering reading. The comparison is not clean — and that caveat is Levine's own, not this chapter's contribution: he writes that "modern humans have stretched the boundaries of biology to achieve a life expectancy of 80 years," and credits "advances in science, medicine and sociology." The mammalian figures are for animals dying on their own schedules; ours are not.

So recompute on an evolutionary lifespan — and source the input, because this one number carries the whole reversal. Two quantities get confused here, and only one of them is usable:

• Life expectancy at birth (e₀) for traditional hunter-gatherers is 21–37 years (Gurven & Kaplan 2007, Popul Dev Rev 33(2):321–365). e₀ cannot be multiplied by a heart rate. It is an average depressed largely by infant mortality — nobody lives "an e₀," and a population whose survivors reach seventy still posts a low one.
• Modal adult age at death is the quantity that means how long a human body runs. Gurven & Kaplan: "The modal age of adult death is about seven decades... human bodies are designed to function well for about seven decades in the environment in which our species evolved."

Recompute at seven decades:

70 × 60 × 24 × 365.25 × 70 ≈ 2.6 × 10⁹ beats   → ~3.5× the mammalian mean, ~3.3 SD out

Still outside the band. The "it's just modern medicine" escape moves us from ~4 SD to ~3.3 SD and leaves us outside.

And record what that correction cost this chapter. An earlier pass recomputed at "a pre-industrial-plausible 40 years," got ~1.5 × 10⁹, and called it "comfortably inside the band." That 40 was sourceless, and against the one sourced series available here it is neither quantity — above the hunter-gatherer e₀ range (21–37) and well below the modal adult lifespan (~7 decades). It delivered a reversal the sourced numbers do not support. This is the chapter's own spine biting the chapter: the flattering escape was the unsourced number. Printed, not quietly deleted.

The honest verdict is three-layered: the invariant is loose; humans sit outside it at both a modern and an evolutionary lifespan; and whether that reflects biology or the comparison's own construction is not settled by this arithmetic — least of all because Levine's mammalian coordinates are themselves averaged values from a 1956 handbook and a 1971 encyclopaedia (see the table). An outlier is data, not a trophy.

Gait: the transition is computable

The Froude number for legged locomotion is Fr = v²/(gL), with L the leg length. Bipeds switch from walking to running at Fr ≈ 0.5 — different absolute speeds, same dimensionless number. Compute it for a typical adult leg length L = 0.9 m:

v = √(Fr · g · L) = √(0.5 × 9.81 × 0.9) = √4.4145 ≈ 2.1 m/s  (≈ 7.6 km/h)

Across a normal adult range (L = 0.8–1.0 m) that is 2.0–2.2 m/s. The Fr ≈ 0.5 transition is measured, not merely modelled — Kram, Domingo & Ferris (1997), J Exp Biol 200(4):821–826: "humans and other bipeds with different leg lengths all choose to switch from a walk to a run at different absolute speeds but at approximately the same Froude number (0.5)." The dimensionless number is the sourced claim. A measured absolute preferred transition speed in m/s, with its n, is NOT-SOURCED in this pass (that paper's full text was paywalled), so "this is where people actually switch" is carried at the Froude level only — the 2.0–2.2 m/s above is computed, not observed.

The informative part: the mechanical ceiling for an inverted pendulum is Fr = 1, where the centripetal requirement exceeds gravity and the foot can no longer stay down. At L = 0.9 m that is √(9.81 × 0.9) ≈ 3.0 m/s. We switch at 2.1 m/s, well below the 3.0 m/s we could still walk at. The transition is an energetic choice, not a hard limit — a walk is an inverted pendulum trading potential against kinetic energy, a run is a spring-mass bouncer storing elastic strain, and the switch happens where the second gets cheaper, not where the first becomes impossible.

The endurance-running hypothesis (Bramble & Lieberman 2004, Nature 432:345–352) argues endurance running is a derived capability of Homo, ~2 Ma, shaping our anatomy. Its contested status must travel with the claim. Pickering & Bunn (2007, J Hum Evol) objected that persistence hunting is rare even among skilled modern trackers in ideal terrain, and that early Homo may have lacked the tracking ability; Lieberman et al. replied that over-reliance on the recent ethnographic record ("the tyranny of ethnography") is a poor test of past behaviour; Morin & Winterhalder (2024, Nat Hum Behav, DOI 10.1038/s41562-024-01876-x) assembled ~400 endurance-pursuit cases from 272 locations and argued the return rates are competitive with other pre-modern methods. Live dispute, both sides in print. Not settled.

Thermoregulation: the actual superpower, and why it works

Here the human case is genuinely strong — and the mechanism is a decoupling, not a magic. Most running mammals dump heat by panting, and in a galloping quadruped breathing is phase-locked to the locomotor cycle: the gait sets the breath, so you cannot independently crank the pant rate while running. Sweating is not coupled to respiration (Carrier 1984), so a sweating animal can raise its cooling rate without touching its gait or its breathing. That is the whole advantage, and it is structural.

The capacity, computed with CN-02's λ ≈ 2430 J/g at skin temperature:

1 L/h   → 1000 g/h × 2430 J/g ÷ 3600 s ≈  675 W   ≈ 8× BMR
2 L/h   →                                ≈ 1350 W  ≈ 16× BMR
3.7 L/h →                                ≈ 2500 W  ≈ 30× BMR

1 L/h is routine in hot environments; 3.7 L/h is the highest sweat rate reported in that literature — one runner, one race (Alberto Salazar, 1984 Olympic Marathon; primary: Armstrong et al. 1986, Physician Sportsmed 14(3):73–81, reached here via NBK236240). A record maximum at n = 1 is an observation, not a replicated constant, and is classed that way below. A system whose baseline power draw is 84 W can shed ~675 W routinely and ~2.5 kW at that single reported extreme.

And the loop closes on CN-02. This works only because water's enthalpy of vaporisation is anomalously high — 2257 kJ/kg at 100 °C, ~2430 J/g at skin temperature. Sweating is not a clever trick; it is a direct draw on a thermodynamic property of the working fluid.

Carry CN-02's fence, load-bearing here: 2430 J/g is a ceiling, not a delivery. Havenith et al. (2013) measured the effective latent heat as up to 45% below the physical value under vapour resistance, because evaporation displaced from the skin cools the clothing rather than you. Sweat that drips cools nothing. In humid air the deliverable fraction collapses; every watt above is an upper bound.

Sensing: real ranges, and one frozen accident

Photons. Hecht, Shlaer & Pirenne (1942), J Gen Physiol 25(6):819–840, measured threshold energies of 2.1–5.7 × 10⁻¹⁰ ergs at the cornea = 54–148 quanta of blue-green light, and inferred that only 5–14 quanta are actually absorbed by rods — each by a different rod, implying a single rod responds to a single photon. Tinsley et al. (2016), Nat Commun 7:12172, used a heralded single-photon source and found humans report a single photon incident on the cornea at a probability significantly above chance, with a priming effect from an earlier photon on a timescale of seconds.

Fence this properly: it is a statistical result over many trials at the very edge of detectability, not a percept you can summon. Nor is it uncontested — a 2024 Nat Commun paper titled "Primate retina trades single-photon detection for high-fidelity contrast encoding" (DOI 10.1038/s41467-024-48750-y) asserts a tension on its face. Read at title level only in this pass; recorded as a pointer, not a summarised finding.

Dynamic ranges are in the table rather than here, because they are standard-reference values not primary-sourced in this pass — and are recorded NOT-MEASURED there rather than given a class they have not earned. Hearing: the ratio 0→120 dB SPL (20 µPa → 20 Pa) = 10⁶ in pressure, 10¹² in intensity follows from the dB definition, but the endpoints do not. 0 dB SPL is a reference level anchored near 1 kHz — not "the threshold of hearing": the human threshold is strongly frequency-dependent and dips below 0 dB SPL around 3–4 kHz, and the 120 dB upper endpoint is a convention with no source carried here. Vision: ~10 orders of magnitude of luminance, scotopic→photopic — no primary fetched in this pass; carried as a ballpark, not asserted.

And now the retina, which is the anti-flattery centrepiece.

The vertebrate retina is installed backwards. Photoreceptors face away from the incoming light, which must first pass through the ganglion-cell and bipolar-cell layers before reaching anything photosensitive. The axons then exit through the retina, punching a blind spot in the visual field. Cephalopods, who evolved a camera eye independently, have it the other way: photoreceptors face the light, axons exit posteriorly, no blind spot.

The vertebrate arrangement follows from developmental history — the eye derives from an invagination of the neural tube from dorsal ectoderm, and the orientation falls out of that. This is precisely NA-01's Gould & Lewontin (1979) spandrel: a frozen accident locked in by developmental constraint, not an optimum. The human eye is the standing counterexample to "nature does it this way, so it must be right."

Be fair, not sneering — then follow the argument through. Evolution patched it: Franze et al. (2007), PNAS 104(20):8287–8292 showed Müller glia act as living optical fibres — funnel-shaped, higher refractive index than surrounding tissue, aligned with the light path, forming a fibreoptic-plate-like array that carries the image through the obstructing layers with low distortion and low loss. Real result, real mechanism.

But read what it means. A patch that recovers most of the loss from a defect is evidence the defect was real and worth patching. Were the inverted retina an optimum, you would not need a dedicated glial waveguide array to undo it. And the patch is incomplete: the blind spot is still there, filled in by the visual system rather than fixed by the optics, and retinal detachment remains a failure mode this architecture invites. Franze et al. rebuts "the inversion is a disaster." It does not rebut "the inversion is an accident."

Development: the obstetrical dilemma, still open

The classic account (Washburn's "obstetrical dilemma"): a large neonatal brain and a narrow bipedally-adapted birth canal are in antagonistic selection; the compromise is to truncate gestation, producing a helpless newborn.

Dunsworth et al. (2012), PNAS 109(38):15212–15216 challenged it with EGG (Energetics of Gestation and Growth): birth is triggered when fetal energy demand approaches the mother's maximum sustained metabolic scope, put at ~2.0–2.1 × BMR of the non-pregnant, non-lactating female — so maternal metabolism, not the pelvis, sets gestation length. Warrener et al. (2015), PLoS ONE 10(3):e0118903 then tested the mechanical premise directly and found a wider pelvis does not increase locomotor cost, removing the dilemma's assumed trade-off at its root.

And EGG is itself now contested — carry all three layers. Later work reports maternal metabolic rate exceeded the 2.1 × BMR ceiling in about a third of studies, and that simulations suggest the fetus cannot realistically push the mother across that threshold at term (Evol Med Public Health 11(1):415, 2023). Three positions in print: pelvic constraint, metabolic constraint, neither-as-stated. Printing it as settled in any direction would be the defect.

The blanket: where the cardinal rule lands hardest

Per NA-04 and M12, the human is one system with two typed Markov blankets — WORLD ⊥ BODY ⊥ MIND. And the human is the only system in this wing where we possess first-person testimony about internal states. You can ask it how it feels. Nothing else in the corpus talks back.

That testimony is an HONEST signal. It is never a TRUE one.

The repo's cardinal rule lands hardest here, because here the temptation is strongest. Testimony is lived experience: respected exactly as lived, and never calibrated — calibrating a person's report of their own experience against an "objective" measure is a category error that destroys the thing it claims to check. It is also not falsifiable, not reproducible, not recalibratable, so it is never admitted as TRUE, and no chain of reasoning may pass from a report of experience to a measured claim.

The interoceptive hardware signals (M12) — heart rate, core temperature, CMRO₂, sweat rate — are TRUE-class. The report "I am too hot" is HONEST-class. They may correlate perfectly and still never merge. A design that reads a thermistor and a self-report into one store has committed the cardinal sin — and that both concern the same body is exactly what makes the error easy to make.

The numbers

Symbol	Value	Units	Scope	Class	Source	Falsifier
N_cells,Bianconi	3.72 × 10¹³	cells	standard adult; organ-by-organ bibliographic sum	MODELED	Bianconi et al. 2013, Ann Hum Biol 40(6):463–471	Independent systematic count outside 2–6 × 10¹³
N_cells,Sender	3.0 × 10¹³ (2% est. unc.; 14% CV)	cells	70 kg reference male	MODELED	Sender, Fuchs & Milo 2016, PLoS Biol 14(8):e1002533	Recount outside stated uncertainty
f_RBC	~84 (2.5 × 10¹³)	% of human cells	70 kg male	MODELED	Sender et al. 2016	Recount; RBC fraction outside 75–90%
N_nucleated	~5 × 10¹²	cells	3.0 × 10¹³ − 2.5 × 10¹³	MODELED	Computed here from Sender et al. 2016	Arithmetic error
N_bacteria	3.8 × 10¹³ (25% est. unc.; 52% SD)	cells	70 kg male, mostly colon	MODELED	Sender et al. 2016	Independent estimate outside stated band
R_B/H	1.3 (25% unc.; 53% variation)	dimensionless	all human cells as denominator	MODELED	Sender et al. 2016	Recount outside band
R_B/H,nucleated	~8:1	dimensionless	nucleated human cells as denominator	MODELED	Computed here from Sender et al. 2016	Arithmetic error
R_B/H,folklore	10:1 — superseded	dimensionless	traced to one 1972 back-of-envelope (10¹¹/g × 1 L)	INADMISSIBLE as stated (no method, no denominator)	Sender et al. 2016 (the tracing)	Produce the original measurement it was said to rest on
m_microbiota	~0.2 (0.3% of body mass)	kg wet weight	70 kg male	MODELED	Sender et al. 2016	Direct measurement >2× off
Δ_defecation	25–33	% of colonic content per event	normal adult	OBSERVED-REPLICATED	Sender et al. 2016	Measurement outside range
V̇O₂,basal	250	mL O₂/min	70 kg man, basal state	OBSERVED-REPLICATED	Clarke & Sokoloff, Basic Neurochem, 6th ed. 1999	Indirect calorimetry >20% off
k_O₂	4.82 (≈20.1 kJ/L)	kcal/L O₂	mixed diet, RQ ≈ 0.82 (fat 4.68, CHO 5.03)	OBSERVED-REPLICATED	standard indirect calorimetry	Bomb-calorimetry value outside 4.6–5.1
P_BMR	~84 (≈1,735 kcal/day)	W	70 kg man; from V̇O₂ × k_O₂	MODELED	Computed here from Clarke & Sokoloff	Arithmetic error, or V̇O₂ refuted
P_TDEE	~97 (2,000 kcal/day)	W	the 2,000 kcal/day dietary-labelling convention, converted — not a measured TDEE, and not BMR. No sex/age/mass/activity scope. The input is NOT-MEASURED	MODELED on an unsourced input	Computed here; the 2,000 kcal/day input is NOT-SOURCED in this pass	Conversion: arithmetic error. Input falsifier (the one that matters): a doubly-labelled-water TDEE for a stated cohort outside 1,800–2,200 kcal/day
p_body	~1.2	W/kg	84 W / 70 kg	MODELED	Computed here	Arithmetic error
p_cell	~100 (mammalian) / ~1,000 (E. coli)	W/kg	cross-ref NA-08	OBSERVED-REPLICATED	BNID 111474/111475; 109687 (see NA-08)	see NA-08
N_neurons	86.1 ± 8.1 × 10⁹	neurons	adult male human brain, isotropic fractionator	OBSERVED-REPLICATED	Azevedo et al. 2009, J Comp Neurol 513:532–541	Independent count outside ±2 SD
N_nonneuronal	84.6 ± 9.8 × 10⁹	cells	same	OBSERVED-REPLICATED	Azevedo et al. 2009	As above
R_glia:neuron	~1:1	dimensionless	whole human brain	OBSERVED-REPLICATED	Azevedo et al. 2009	A count returning ≥3:1 by an unbiased method
R_glia,folklore	10:1 — superseded	dimensionless	no supporting reference locatable	INADMISSIBLE as stated	Herculano-Houzel 2012, PNAS 109(S1):10661–8 (the tracing)	Produce the primary source it rested on
N_neurons,folklore	100 × 10⁹ — superseded	neurons	no supporting reference locatable	INADMISSIBLE as stated	Herculano-Houzel 2012 (the tracing)	As above
f_cortex	19	% of all brain neurons	adult male human, isotropic fractionator — the human-specific unbiased count	OBSERVED-REPLICATED	Azevedo et al. 2009 (abstract: "With only 19% of all neurons located in the cerebral cortex")	Unbiased count outside 17–21%
f_cortex,mammals	20–25	% of all brain neurons	most mammalian species examined, including humans — regardless of relative cortex size. A cross-species regularity, not the human measurement (see f_cortex)	OBSERVED-REPLICATED	Herculano-Houzel 2012	A mammalian clade whose cortex holds a materially different share
R_cerebellum:cortex	~4:1	neurons	human; consistent with f_cortex = 19 (cortex 19% + cerebellum ~80% + rest ~1% = 100). Note it does not close at 25% — 25 × 4 = 100 leaves nothing for the rest of the brain	OBSERVED-REPLICATED	Herculano-Houzel 2012; Azevedo et al. 2009	Unbiased count outside 3–5:1
N_synapses	1.64 × 10¹⁴ (164 × 10¹², CV = 0.17; textbook summaries span 0.6–2.4 × 10¹⁴)	synapses	human neocortex, n = 5 young male autopsy brains, unbiased stereology	OBSERVED-SINGLE (single study, n = 5; no independent replication located)	Tang et al. 2001, Synapse 41(3):258–273 — primary abstract read this pass: "The average total number of synapses in the neocortex of five young male brains was 164 × 10¹²"	An independent stereological count outside 0.6–2.4 × 10¹⁴; any replication at n > 5
f_brain,mass	~2	% of body mass	~1.4–1.5 kg / 70 kg	OBSERVED-REPLICATED	standard anatomy; Herculano-Houzel 2012	Measurement outside 1.5–2.5%
f_brain,O₂	~20	% of resting O₂ consumption	70 kg man, basal	OBSERVED-REPLICATED	Clarke & Sokoloff 1999	CMRO₂ measurement outside 15–25%
P_brain,CMRO₂	~17 (≈347 kcal/day)	W	from 20% × 250 mL O₂/min	MODELED	Computed here from Clarke & Sokoloff	Arithmetic error
P_brain,perneuron	~25 (516 kcal/day; 25% of TDEE)	W	86 × 10⁹ × 6 kcal/d per 10⁹ neurons	MODELED	Herculano-Houzel 2012	Direct human measurement settling the 1.5× gap
P_brain spread	~1.5× — do not average	dimensionless	CMRO₂ route vs per-neuron route	OBSERVED-CONTESTED / MODELED	the two rows above	A method reconciling both under one measurement
p_brain	~12–18	W/kg	17–25 W over ~1.4 kg; ~10–15× body mean	MODELED	Computed here	Arithmetic error
c_neuron	6	kcal/day per 10⁹ neurons	primates; comparative regression	MODELED	Herculano-Houzel 2012	Direct per-neuron measurement >2× off
human brain vs primate line	on the line — not an outlier	—	neuron count vs brain size, primates	OBSERVED-REPLICATED	Azevedo et al. 2009 (title claim); Herculano-Houzel 2012	A primate scaling regression placing humans significantly off-line
m_brain,human vs elephant	human ~1.5; elephant 2–3× larger	kg	humans also not first in relative brain size	OBSERVED-REPLICATED	Herculano-Houzel 2012	Comparative measurement refuting
N_beats,mammal	7.3 ± 5.6 × 10⁸ (CV ≈ 77%)	beats/lifetime	15 mammal species, excluding humans. *Provenance, per Levine's own Fig. 1 caption — "Most coordinates represent average values (4–6)":* refs 4–6 are Spector, Handbook of Biological Data (1956); Encyclopaedia Britannica (Chicago: W Benton, 1971); and White et al., "The electrocardiogram of the elephant" (1938). Species lifespan/HR coordinates not independently traced, and "life expectancy" is left undefined per species	OBSERVED-CONTESTED (a loose "invariant", on averaged and untraced coordinates)	Levine 1997, J Am Coll Cardiol 30:1104–1106	Expand the species set; a tighter or looser CV; trace the coordinates to primary measurements
N_beats,human	~2.9 × 10⁹ (2.945 × 10⁹; ~4× mean, ~4 SD out)	beats/lifetime	70 bpm × 80 yr	MODELED	Levine 1997 states ~3 × 10⁹ himself at these inputs — "If humans are predetermined to have ~3 billion heart beats/lifetime... a reduction in mean heart rate from 70 to 60 beats/min throughout life would increase life span from 80 to 93.3 years" — recomputed here as 2.945 × 10⁹. Not this chapter's arithmetic	Arithmetic error; Levine's own stated figure moving
N_beats,human@70yr	~2.6 × 10⁹ (2.577 × 10⁹; ~3.5× mean, ~3.3 SD — still OUTSIDE the band)	beats/lifetime	70 bpm × modal adult age at death ≈ 7 decades (lifespan in the evolutionary environment)	MODELED	Computed here; lifespan input sourced to Gurven & Kaplan 2007, Popul Dev Rev 33(2):321–365 — "The modal age of adult death is about seven decades"	Arithmetic error; a sourced evolutionary lifespan materially ≠ ~7 decades
e₀,hunter-gatherer	21–37	years	life expectancy at birth, traditional hunter-gatherers. Not a per-individual lifespan and cannot be multiplied by a heart rate — an average depressed largely by infant mortality. Recorded to block exactly that multiplication	OBSERVED-REPLICATED (cross-cultural synthesis)	Gurven & Kaplan 2007 ("the average life expectancy at birth (e₀) varies from 21 to 37 years")	A cross-cultural series placing hunter-gatherer e₀ outside 21–37
N_beats,human@40yr	~1.5 × 10⁹ (~1.3 SD, "comfortably inside the band") — WITHDRAWN this pass	beats/lifetime	70 bpm × 40 yr, "pre-industrial-plausible"	INADMISSIBLE as stated: the 40 yr was unsourced, and matches neither sourced quantity — above hunter-gatherer e₀ (21–37), well below modal adult lifespan (~7 decades). It was the single input carrying the chapter's headline reversal	previously "Computed here"; the input was never sourced — the falsifier on the row ("Arithmetic error") could not catch it, because the arithmetic was clean	Produce a sourced pre-industrial lifespan of ~40 yr that is a lifespan and not an e₀
Fr_transition	~0.5	dimensionless	Fr = v²/(gL), L = hip height above ground; bipeds, walk→run	OBSERVED-REPLICATED	Alexander 1984, "The Gaits of Bipedal and Quadrupedal Animals", Int J Robot Res 3(2):49–59 (dynamic similarity); Kram, Domingo & Ferris 1997, J Exp Biol 200(4):821–826 — abstract: bipeds of differing leg length switch "at approximately the same Froude number (0.5)"	A biped switching at a materially different Fr
v_transition	~2.1 (range 2.0–2.2 for L = 0.8–1.0 m)	m/s	L = 0.9 m, g = 9.81 m/s². Computed, not observed — a measured absolute preferred transition speed with its n is NOT-SOURCED in this pass (Kram et al. 1997 full text paywalled; only their dimensionless Fr ≈ 0.5 is sourced)	MODELED	Computed here: v = √(Fr·g·L), from Fr_transition above	Arithmetic error; a measured preferred transition speed outside 2.0–2.2 m/s
v_Fr=1	~3.0	m/s	inverted-pendulum mechanical ceiling, L = 0.9 m	MODELED	Computed here	Arithmetic error
λ_sweat	~2430 (2257 at 100 °C)	J/g	water, skin temperature — cross-ref CN-02	OBSERVED-REPLICATED	see CN-02	Calorimetry outside ±1%
Ṡ_routine	1	L/h	working in hot environments	OBSERVED-REPLICATED	Nutritional Needs in Hot Environments (NCBI NBK236240)	Field measurement outside 0.5–2 L/h
Ṡ_max	3.7	L/h	n = 1 — Alberto Salazar, 1984 Olympic Marathon. A single reported record maximum: one runner, one race, uncorroborated. A record max from n = 1 is the opposite of replicated	OBSERVED-SINGLE (single report)	primary: Armstrong, Hubbard, Jones & Daniels 1986, Physician Sportsmed 14(3):73–81; secondary route: NBK236240 — "The highest sweating rate reported in the literature is 3.7 liters per hour, measured for Alberto Salazar during the 1984 Olympic Marathon (Armstrong et al., 1986)"	A higher documented rate; any replication at n > 1
P_evap@1L/h	~675 (≈8× BMR)	W	ceiling, assumes full evaporation at the skin	MODELED	Computed here from λ_sweat	Arithmetic error
P_evap@3.7L/h	~2500 (≈30× BMR)	W	ceiling, same assumption	MODELED	Computed here	Arithmetic error
λ_eff reduction	up to 45	% below physical λ	vapour resistance displaces evaporation off-skin	OBSERVED-REPLICATED	Havenith et al. 2013 (see CN-02)	Manikin study finding λ_eff ≈ λ
panting decoupling	breathing phase-locked to gait in galloping quadrupeds; sweating is not	—	mechanism, not magic	OBSERVED-REPLICATED	Carrier 1984	A galloping quadruped independently modulating pant rate
E_threshold	2.1–5.7 × 10⁻¹⁰ ergs = 54–148 quanta	at the cornea	blue-green, optimal conditions	OBSERVED-REPLICATED	Hecht, Shlaer & Pirenne 1942, J Gen Physiol 25(6):819–840	Replication outside range
n_quanta,absorbed	5–14	quanta at the rods	inferred after corneal/media/rhodopsin losses	MODELED	Hecht et al. 1942	A loss model changing the inference
single-photon detection	above chance	probability	single photon at cornea; heralded source	OBSERVED-CONTESTED	Tinsley et al. 2016, Nat Commun 7:12172; tension asserted by DOI 10.1038/s41467-024-48750-y (title-level reading only)	Read the 2024 paper; a replication failing above-chance detection
DR_hearing	10⁶ pressure / 10¹² intensity (0→120 dB SPL; 20 µPa → 20 Pa)	dimensionless	The ratio follows from the dB definition; the endpoints do not. 0 dB SPL = the 20 µPa reference level, anchored near 1 kHz — not "the threshold of hearing": the human threshold is strongly frequency-dependent and falls below 0 dB SPL near 3–4 kHz. 120 dB upper endpoint = a convention, no source carried	MODELED (the ratio) / NOT-MEASURED (both endpoints)	Computed here from the 20 µPa reference level; no primary carried for either endpoint	Ratio: arithmetic error. Endpoints (the ones that matter): an audiometric threshold series showing the 20 µPa reference is not ≈ threshold at 1 kHz; or a sourced upper endpoint ≠ 120 dB SPL
DR_vision	~10 — carried as a ballpark, not asserted	orders of magnitude of luminance	scotopic→photopic	NOT-MEASURED — no source located this pass. "Standard reference" is not a source; it is the same provenance status this chapter's opening table indicts three famous numbers for having	none — NOT-SOURCED in this pass	Fetch a primary for the scotopic→photopic luminance range; a sourced range materially ≠ ~10 orders
retina orientation	inverted (photoreceptors face away; axons exit through it → blind spot)	—	vertebrates; cephalopods verted, no blind spot	OBSERVED-REPLICATED	Franze et al. 2007, PNAS 104(20):8287–8292; cephalopod comparative literature	An anatomical demonstration that vertebrate photoreceptors face the light
Müller cells as waveguides	funnel-shaped, higher refractive index, aligned to light path	—	the patch, not a refutation of the accident	OBSERVED-REPLICATED	Franze et al. 2007	Optical measurement showing no fibre-like transfer
ceiling_EGG	2.0–2.1	× BMR (non-pregnant, non-lactating)	maternal max sustained metabolic scope	OBSERVED-CONTESTED	Dunsworth et al. 2012, PNAS 109(38):15212–15216; contested by Evol Med Public Health 11(1):415 (2023)	A study confirming the ceiling is never exceeded at term
wider pelvis → locomotor cost	no measured increase	—	men and women, dynamic hip-abductor model	OBSERVED-REPLICATED	Warrener et al. 2015, PLoS ONE 10(3):e0118903	A replication finding a metabolic penalty for pelvic width
first-person testimony	—	—	the human is the only system here that reports its states	HONEST-class, never TRUE	repo cardinal rule; NA-04, M12	None. Testimony is never calibrated — that is the point, not a gap.

Falsifier (operable)

The chapter's central structural claim — that the famous headline numbers about the human body were unsourced ballparks that moved when measured, and that human "specialness" survives none of the scaling checks applied here — is refuted by exhibiting either:

1. A primary source, predating the corrections, that actually measured any of: 100 billion neurons, 10:1 glia, 10:1 bacteria, or a whole-body cell count — with a stated method and denominator. (Sender et al., Bianconi et al., and Herculano-Houzel each independently report failing to find one. Produce one and the spine breaks.) Or:
2. A primate neuron-count-vs-brain-size regression placing humans significantly off the primate line. Azevedo et al. (2009) and Herculano-Houzel (2012) put us on it. Off-line humans would restore the exceptionalism this chapter denies.

Note carefully what does not refute it: the heartbeat outlier. Humans genuinely sit ~4 SD outside Levine's band at a modern lifespan — and ~3.3 SD outside at a sourced evolutionary one — and the chapter says so. An outlier on a loose invariant, whose own species coordinates are averaged values from a 1956 handbook and a 1971 encyclopaedia, is data about the invariant and the comparison, not a warrant for specialness.

Secondary falsifiers are row-local: any number in the table found outside its stated scope under its stated conditions moves that row and only that row.

Recorded INADMISSIBLE / NEGATIVE (first-class, inline)

• "10 bacteria for every human cell." — INADMISSIBLE as stated: no denominator, no method. Receipt: Sender et al. (2016) traced every downstream citation to one 1972 back-of-the-envelope (10¹¹/g × 1 L). Recorded, not mocked — and the popular retelling omits the honest twist: against nucleated cells the ratio is ~8:1. The defect was the missing denominator, not the arithmetic.
• "100 billion neurons and 10× as many glia." — INADMISSIBLE as stated. Receipt: Herculano-Houzel (2012) could locate no supporting reference; measured values are 86.1 ± 8.1 × 10⁹ and ~1:1.
• "The human brain is anomalously large / uniquely encephalised for our size." — NEGATIVE against the primate line. Receipt: we carry the neuron count expected of a generic primate brain of our size; we are not first in relative brain size; the elephant's is 2–3× larger. The distinctive rule belongs to primates. Recorded as a published negative in the M15 sense — this is the credibility, not an embarrassment.
• "The brain uses 20% of your energy, therefore human cognition is special." — NEGATIVE / inference defect. The premise is real; the conclusion does not follow. Receipt: per-neuron cost is conserved across primates, so the 20% share follows arithmetically from neuron count.
• "~10⁹ heartbeats per lifetime is an invariant." — OBSERVED-CONTESTED, and loose. Receipt: CV ≈ 77% on Levine's own 15-species mean, defended only within an order of magnitude — and, per Levine's Figure 1 caption, "Most coordinates represent average values (4–6)", where refs 4–6 are a 1956 biological-data handbook, a 1971 Encyclopaedia Britannica, and a 1938 elephant ECG paper. This chapter's spine applies to this chapter's own comparator: a loose invariant whose mammalian coordinates are averaged values from an encyclopaedia is a fourth instance of the failure mode, not an exception to it. Quoting it as a constant — without the CV, without that provenance, and without noting humans sit ~4 SD outside — is the defect. Recorded because it is repeated constantly in exactly that stripped form.
• "The inverted retina is fine / optimal, because Müller cells fix it." — NEGATIVE / inference defect. Franze et al. (2007) is real and replicated, and it rebuts "the inversion is a disaster." It does not rebut "the inversion is a frozen accident" — a patch recovering the loss is evidence the defect was real and worth patching. The blind spot is still unfixed optically; the cephalopod built the other arrangement independently.
• The obstetrical dilemma as settled fact. — NEGATIVE as stated / genuinely open. Receipt: Warrener et al. (2015) measured no locomotor cost to a wider pelvis, removing the assumed trade-off; but EGG is also contested (metabolic rate exceeds the 2.1 × BMR ceiling in ~⅓ of studies reviewed). Three positions in print; printing any as settled is the defect.
• NOT-SOURCED in this pass — named, therefore not asserted: that brain metabolism is nearly flat with cognitive effort (closure: fetch Raichle & Mintun, Annu Rev Neurosci 29:449–476, 2006); a primary for the ~10-order visual luminance range; the 120 dB SPL upper endpoint of the hearing range (a convention, no source carried); that neocortical synapse count declines with age — Tang et al. (2001) measured five young male brains and reports no age series, so the claim has been removed from that row rather than carried on a citation that cannot bear it (closure: a stereological cross-sectional ageing series); a measured absolute walk→run transition speed in m/s with its n (Kram et al. 1997 full text paywalled — their dimensionless Fr ≈ 0.5 is sourced from the abstract); a source for the 2,000 kcal/day TDEE input (closure: a doubly-labelled-water reference with stated scope); the 2024 primate-retina paper (title level only).
• RESOLVED this pass — was NOT-SOURCED, now primary-read: the exact Tang et al. (2001) synapse figure. The primary abstract reports 164 × 10¹² (CV = 0.17) in five young male brains — i.e. 1.64 × 10¹⁴. This chapter had printed ~1.5 × 10¹⁴, which is neither of the two renderings it had itself named — a ~9% error adopted against the only source cited for it, while conceding the primary was unread. Recorded, not silently swapped: the row enacted the failure the chapter indicts. The sourced value now stands.
• WITHDRAWN this pass — the unsourced 40-year lifespan. See N_beats,human@40yr. The chapter's own headline reversal ("recompute at 40 years → comfortably inside the band") rested on a number with no source, no stated quantity, and no scope, whose row-falsifier ("Arithmetic error") could not test it because the arithmetic was clean. Sourcing the input (Gurven & Kaplan 2007) removed the reversal: at a modal adult lifespan of ~7 decades humans remain ~3.3 SD outside. The flattering reading was the unsourced one — recorded as the fourth instance of this chapter's own spine, this time against itself.

HONEST FENCE — MODELED

This chapter is fenced MODELED. Individual rows carry their own classes (many OBSERVED-REPLICATED, several OBSERVED-CONTESTED, several NOT-MEASURED), but the chapter as an artifact composes measured constants through stated assumptions — the caloric equivalent of oxygen at RQ 0.82; a 70 kg, 1.75 m reference body with a 0.9 m leg; full evaporation at the skin; 70 bpm and a chosen lifespan. The assumptions are the fence. Change the leg length and the walk–run speed moves; change the lifespan and the heartbeat "outlier" moves — but not out of the band. At a sourced evolutionary lifespan (modal adult death ≈ seven decades, Gurven & Kaplan 2007) humans still sit ~3.3 SD outside. An earlier pass of this chapter printed an unsourced 40 years that did move us inside; sourcing the input deleted the reversal. That is the fence doing its job on the chapter itself, and it is printed rather than buried.

Per Gould & Lewontin (1979), "The Spandrels of San Marco and the Panglossian Paradigm" and NA-01: nothing above establishes that any human feature is an optimum. The inverted retina is this chapter's own standing counterexample to the Panglossian reading — a frozen accident of neural-tube development, patched but not fixed, in the organism most prone to being described as nature's summit. Nature's authority is precisely and only this: it has already run a very long parallel search under real physical constraints in which the failures were deleted. That makes convergence evidence of a constraint-optimum and makes every number above a hypothesis generator — never a proof. Per repo rule M7, any design taken from this chapter must still beat a tuned conventional baseline on a pre-registered metric with a load-bearing discriminator, or it is recorded NEGATIVE.

Not claimed

• Not claimed: anything whatsoever about consciousness. The hard problem — why physical processing is accompanied by subjective experience at all (Chalmers 1995, J Consciousness Studies 2(3):200–219) — is named here and left exactly where it stands: unsolved. Not solved, not dissolved, not explained, and emphatically not hand-waved by any neuron count, synapse count, or metabolic figure in the table above. A number of neurons is not a theory of experience. Nothing in this chapter bears on the question in either direction.
• Not claimed: that first-person testimony is evidence of anything TRUE-class. It is HONEST — respected exactly as lived, never calibrated, never merged with the measured store. That both concern the same body makes the crossing tempting, not permissible.
• Not claimed: that humans are the endpoint, summit, or intended outcome of anything. "Full human" and "beyond human" appear nowhere here as a target, milestone, or deliverable — they are permanent OPEN QUESTIONS, and a chapter of measurements about Homo sapiens has no bearing on either.
• Not claimed: that the endurance-running hypothesis is established. Both sides of a live dispute are carried; neither is endorsed.
• Not claimed: that sweating makes humans thermally superior in general. The claim is narrow and mechanical — decoupling cooling from respiration raises the ceiling on heat dissipation during sustained locomotion — and that ceiling is an upper bound which humid air and dripping sweat collapse.
• Not claimed: that the 1.5× brain-energy discrepancy is resolved, or that either figure is the right one. It is printed as a spread with both methods named.
• Not claimed: that any citation above raises any UNI rung. A nature citation is NEVER a UNI gate. The NATURA twelve-value class (§NA-00; six of the twelve registered by amendment 2026-07-15-A) and the UNI four-value ledger describe different kinds of claim and never merge. This chapter contains zero UNI claims.