The idea is older than the data. In 1966 the psychologist Frederick Snyder proposed what he called the sentinel hypothesis: group-living animals can afford deep sleep because sleep is staggered — while some sleep, others are awake, and the group as a whole never goes blind. It is a tidy piece of evolutionary logic that sat untested in humans for half a century, mostly because testing it requires something inconvenient: a human population sleeping in the conditions the hypothesis is about — outdoors, in groups, among things with teeth — and instruments precise enough to know who is asleep at 3:41 a.m.

Current Conditions

The Briefing in Five Lines
The camp was unguarded for eighteen minutes — in twenty daysSamson and colleagues, Proceedings of the Royal Society B, 2017: actigraphy on 33 Hadza adults over 20 days found all subjects simultaneously asleep for 18 minutes in total. At the median, eight people were awake at any point in the night; at least one adult was awake or in light sleep during 99.8 percent of sampled epochs. Roughly 40 percent of the group was active at any given time.
Nobody wrote the rosterThe stagger was not organized. It emerged from chronotype variation, and chronotype covaries with age — younger Hadza trended owl, elders trended lark. The coverage pattern is a free byproduct of a mixed-age group going to bed whenever each body prefers.
The shift pattern is heritableTwin and family studies put chronotype heritability around 40–50 percent. A genome-wide association study of 697,828 people found 351 loci; carriers of the most morningness alleles sleep about 25 minutes earlier than carriers of the fewest, and the genes cluster in circadian machinery, the retina, and the hypothalamus. Your preferred bedtime is, to a first approximation, a trait — not a habit.
The grandparent, reframedThe authors call it the poorly sleeping grandparent hypothesis: the light, fragmented, early-waking sleep of the old — treated clinically as decline — staffs exactly the hours when everyone else is deepest under. In this reading, grandpa is not broken. Grandpa is the night shift.
Your bedroom is the anomalyThe Hadza averaged 6.25 hours of sleep at 68.9 percent efficiency, napped on half of all days — and showed stronger circadian rhythms than Western samples. A 2023 comparison found nonindustrial groups wake far more at night than Americans do; the sealed, climate-controlled, silent bedroom is the historical outlier. By ancestral standards, sleeping through the night is the exotic behavior.

Eighteen Minutes in Thirteen Thousand

The Hadza of northern Tanzania forage in the same East African savannah woodland where the genus Homo did its formative sleeping, and in 2015 they agreed to wear actigraphs — wrist devices that score sleep and wake from movement and light. David Samson and colleagues tracked 33 adults across 20 days and then asked the simplest possible question of the data: how often was everyone asleep at once?

The answer was eighteen minutes. Not per night — in total, across the entire study, out of roughly thirteen thousand observed minutes of night. Between the first person falling asleep and the last person waking, at least one adult was awake or hovering in light sleep during 99.8 percent of sampled intervals; the median number of people awake at any moment was eight. The researchers built a simulation to generate what coverage would look like under null assumptions, and the observed asynchrony was doing real work: the camp was, in effect, continuously attended, though no one attended to anything on purpose. The wakeful were stoking fires, nursing infants, dozing, chatting — which is why the paper is careful to say sentinel-like behaviour. Nobody stood guard. The guarding was a property of the group, not a duty of any member.

And the mechanism generating it was not vigilance at all. It was disagreement about bedtime. Individuals varied widely in when they slept — chronotype variation, in the trade term — and that variation lined up with age. The young went down late and got up late; the old went down early and got up in what the young would consider the middle of the night. Stack those preferences across a mixed-age camp and the coverage falls out arithmetically, the way a city ends up with someone awake on every block without a municipal wakefulness plan.

The Roster Written in the Genome

The load-bearing question is whether the stagger is a stable feature of humans or a lifestyle artifact, and here the genetics is unusually solid. Chronotype is heritable at roughly 40 to 50 percent in twin and family studies — about the same ballpark as height’s environmental component, for calibration. The molecular substrate is the circadian clock itself: a transcription-translation feedback loop of genes discovered in fruit flies — PER, CRY, CLOCK, BMAL1 — whose variants shift the phase of the whole cycle. Familial extreme larks carry identified mutations; so do some extreme owls. And the largest genome-wide association study, nearly seven hundred thousand people across UK Biobank and 23andMe, mapped 351 loci for morningness, enriched exactly where a clock should live — core circadian genes, the retina that entrains the clock to light, the hypothalamus that houses it. People carrying the most morningness alleles sleep about 25 minutes earlier than those carrying the fewest. Twenty-five minutes sounds modest until you remember that is the difference between the tails of a purely additive score; the full observed spread between a strong lark and a strong owl runs to hours.

The age pattern is equally robust and equally involuntary. Human chronotype delays sharply through adolescence, peaks in lateness around age twenty, then advances steadily for the rest of life — a curve so regular that chronobiologists have proposed the turnaround as a biological marker for the end of adolescence. The teenager physically cannot feel sleepy at ten p.m.; the seventy-year-old physically cannot stay under at five a.m. Neither chose it, in the same sense that neither chose their height. What the Hadza data adds is a reason the curve might have been worth keeping: a group whose members’ clocks disagree is a group that is never entirely offline. Variance itself — the thing quality control exists to stamp out — is doing the safety-critical work.

The Night Watch — By the Numbers
Total simultaneous sleep, 33 adults, 20 days (Samson 2017)18 minof ~13,000 observed night minutes
Night epochs with at least one adult awake or in light sleep99.8%median: 8 people awake at any moment
Chronotype heritability (twin & family studies)~40–50%351 loci · GWAS of 697,828 people
Sleep-timing gap between polygenic extremes (top vs bottom 5%)25 minfull lark–owl spread: hours
Hadza average sleep — duration / efficiency6.25 h / 68.9%naps on 54% of days · stronger circadian rhythms

Reading the Pathologies Backwards

Hold the sentinel frame up to a modern sleep clinic’s intake list and watch the entries change sign. Delayed sleep phase in adolescents: the late watch. Advanced sleep phase and fragmented sleep in the elderly: the dawn watch. Nocturnal awakenings in mid-life — the 3 a.m. surfacing that the wellness industry treats as a crisis of cortisol: entirely ordinary in every population sleeping outside a climate-sealed bedroom, and plausibly the connective tissue between shifts. The 2023 follow-up work makes the comparison explicit: measured by wake-after-sleep-onset, Hadza and Malagasy sleepers wake substantially more per night than American cohorts, whose insulated bedrooms buffer away the noises a threat-detection system evolved to check on. Our consolidated eight-hour block is not the design spec. It is what the design does when you remove everything it was designed for.

None of the individual complaints is thereby cured, and the reframe earns its keep only if stated precisely: the claim is not that your insomnia is secretly good, but that the machinery producing it may be population-level machinery, misread for a century as individual defect because sleep medicine studies one person per bed, one bed per room. A distributed system was diagnosed one node at a time, and every node looked faulty. The fault was a monoculture: industrial society picked a single chronotype — the moderate lark, at whose convenience school bells and office hours ring — declared it the norm, and medicalized the tails.

The Monoculture and Its Bill

The bill has a name: social jetlag — Till Roenneberg’s term for the standing gap between your body clock and your alarm clock, measured as the shift in sleep midpoint between workdays and free days. An owl working lark hours lives in permanent mild time-zone displacement without ever boarding an aircraft. The epidemiology consistently finds evening types and large social jetlag associated with worse metabolic numbers, more smoking, more depression — and here I owe the usual caution, because the associations are the weakest part of the story. The measured effects are real and small: the flagship meta-analytic BMI difference between evening and morning types is 0.44 kg/m², roughly a heavy dinner. Confounding is everywhere — evening types are younger, drink more, and fill in questionnaires differently. The cleanest causal probe available, Mendelian randomization on the GWAS data, finds morningness genuinely improves mental health — consistent with misalignment doing damage — but shows no causal effect on body-mass index or type 2 diabetes. The honest summary: forcing owls to live as larks measurably hurts their mood and plausibly their metabolism, and the metabolic half of that sentence is still an argument, not a result.

What does survive every control is the asymmetry of blame. A trait that is half genetic, mapped to 351 loci, and visibly age-programmed is treated, in the one case, as a moral failing — the owl is lazy, undisciplined, in need of a 5 a.m. bootcamp — and in the other case as a virtue: the lark’s early rising is read as character, when it is substantially a genotype taking credit for itself. The sentinel data quietly dissolves the hierarchy. In the environment that built us, the owl was not the camp’s degenerate. The owl was the first watch.

For 99.8 percent of the night, someone was awake. No rota, no orders, no captain of the guard. The schedule was written in the genome and staffed by age — and what our clinics call sleep disorders may be the shifts, still reporting for duty in a world that cancelled the watch.

Three Honesties

First: the evidence is one camp wide. Thirty-three adults, one population, twenty days, actigraphy — which scores sleep from wrist movement and cannot see the difference between still-and-awake and asleep. Malagasy agriculturalists show compatible patterns, and the age-chronotype curve replicates globally, but the headline measurement — the eighteen minutes — has the sample size of a large dinner party. It is a beautiful result awaiting its replications, and my enthusiasm should be read at that dosage.

Second: the function is inferred, not observed. What was measured is the absence of gaps. What is claimed is vigilance — and no leopard was recorded declining to attack a covered camp. The wakeful Hadza were not scanning perimeters; they were living the ordinary small hours. Coverage that would enable vigilance is not evidence that vigilance is what selection paid for; the authors know this, which is why the modest phrase sentinel-like carries the whole paper’s weight on its hyphen.

Third: the adaptationist trap is set exactly here. Variance in chronotype needs no purpose to exist. Mutation-selection balance maintains variance in nearly everything; the age curve could be a byproduct of developmental and hormonal schedules selected for other reasons entirely; a trait can be useful without having been selected for its usefulness. The just-so story — grandma’s insomnia evolved to guard the camp — is precisely the genre of claim this blog grades harshly when a lifestyle movement makes it about money or diet, and it does not become rigorous because it is charming. The defensible version is weaker and still valuable: chronotype variance is real, heritable, and ancient; in ancestral sleeping groups it happened to buy continuous coverage; whether evolution noticed is an open question. The reframe of the pathologies survives at that strength. The bedtime story about why does not, yet.

Three-Layer Reading
What it saysOver 20 days, a camp of 33 Hadza adults was collectively asleep for 18 minutes; at least one adult was awake or near-waking for 99.8 percent of night epochs. The stagger emerges from chronotype variation, which tracks age — young owls, old larks — and chronotype is roughly half heritable, mapped to 351 genomic loci.
What it impliesSeveral canonical sleep complaints — adolescent delayed phase, elderly early waking, mid-night arousals — read plausibly as shift slots in an emergent coverage system, misdiagnosed as individual defects because sleep is studied one sealed bedroom at a time. The disorder appears when a monoculture prices one chronotype as the norm; social jetlag is that mispricing, measured.
What it means operationallyGrade the measurement separately from the story: the coverage is data; the guard duty is reconstruction; the evolutionary purpose is a hypothesis. Treat your own chronotype as a trait to schedule around, not a character flaw to train away — and price the 5 a.m. content industry accordingly.

What to Actually Take From This

This belongs to a larger question I have spent months inside: my two research preprints ask why sleep exists at all, inside one organism — this piece asks what a night does across a camp. The transferable cargo is mostly epistemic.

Variance can be the feature. Every quality-control instinct says compress the spread; this file shows a spread doing safety-critical work that no individual could do and no individual was assigned. Before medicating, managing, or optimizing away a deviation, ask what the deviation does at the population level. Systems that look broken node by node can be functioning precisely as ensembles — and you cannot see it from inside one node.

The measurement, the mechanism, and the meaning carry different grades. Eighteen minutes: measured. Chronotype-driven stagger: demonstrated. Sentinel function: inferred. Evolutionary purpose: storytelling until further notice. Most science communication welds these four into one claim and sells it at the strength of the first. Unbolt them by habit — here, and everywhere else this blog has pointed the same tool.

Retire the moral reading of wake time. A half-heritable, age-programmed trait is claiming credit as discipline in one tail and taking blame as sloth in the other. Find your own type honestly — free-day sleep midpoint, no alarm, is the cheap instrument — then schedule what you control around it, and treat your standing social jetlag as a measurable cost, not a personal failure. The owl reading this at 1 a.m. is not broken. The owl is ancestrally on shift.

Instrument Check — Worth Your Attention

Study — Samson et al., “Chronotype variation drives night-time sentinel-like behaviour in hunter-gatherers,” Proc. R. Soc. B, 2017. The primary source, short and readable. Note the null-model simulation, the careful hyphen in “sentinel-like,” and how much of the discussion the authors spend disciplining their own hypothesis — a model of how to publish a seductive idea without overselling it.

Read — Jones et al., “Genome-wide association analyses of chronotype in 697,828 individuals,” Nature Communications, 2019. The genetic bedrock: 351 loci, enrichment in clock genes, retina and hypothalamus, the 25-minute polygenic spread — and the Mendelian randomization showing morningness helps mental health while leaving BMI and diabetes untouched. The supplement is where the honest effect sizes live.

Follow — Roenneberg’s Munich ChronoType Questionnaire, and my sleep research. The MCTQ is the free, validated instrument behind “social jetlag”: ten minutes to locate your own sleep midpoint and your standing misalignment. For the within-organism half of the sleep question — why the state exists at all — my two preprints on the research card take it from here.

Flight Log — Dispatch From Altitude

Aviation rebuilt the sentinel hypothesis from scratch, with paperwork where the savannah had genes. A long-haul augmented crew is a sleeping group engineered on purpose: bunk rotations calculated so that someone is always asleep in order that someone is always awake, controlled rest procedures that let one pilot go under in the seat while the other formally holds the watch. The regulator has, in effect, written down what the Hadza camp does for free — continuous coverage from staggered sleep — and the fact that it took a rulebook, rest-facility certification, and a fatigue-science literature to reproduce what emerges spontaneously from a mixed-age campfire is its own comment on which system is the more elegant engineer.

What the rulebook cannot yet do is read the genome. Crew rosters are built from legality and seniority, not chronotype; the 4 a.m. sector is flown by whoever the pairing algorithm dealt it to, and every crew room knows the difference between the captain whose body considers 4 a.m. a reasonable hour and the first officer whose body files a formal protest. Fatigue risk management systems are the industry’s slow admission of this piece’s third signal: that alertness is scheduled biology, not personal virtue — you cannot brief your way out of a circadian trough, you can only staff around it. The airlines learned at altitude what schools and offices still moralize at ground level.

And there is a private consolation in the file. A pilot lives in permanent, occupational social jetlag — the body clock never files a flight plan, and it certainly never reads mine. So when I surface at three in the morning in a hotel eight time zones from my own bed, fully, uselessly awake, I now have a better story than cortisol: somewhere far back in the genealogy, that wakefulness was the camp’s insurance, one clock among many disagreeing on purpose. Nobody is guarding anything from a hotel room in the small hours. But the machinery that keeps watch was never told the watch was cancelled — and there are worse things to be, at 3 a.m., than the last working part of a very old security system.