Linear growth requires adequate energy and a digestible supply of indispensable amino acids. The FAO’s Digestible Indispensable Amino Acid Score, or DIAAS, evaluates protein quality amino acid by amino acid. Mathai and colleagues’ growing-pig assay found substantial digestibility differences among dairy and plant proteins; Herreman and colleagues’ review found that animal-sourced proteins often score higher on average, although variation within both categories is substantial. DIAAS rates individual foods, not whole-diet health or effects on human height.

Height is also highly polygenic. Yengo and colleagues mapped thousands of common-variant associations, allowing part of the measured genetic signal to be summarized in a polygenic score. A PGS is not a complete measure of genetic potential, and standardizing it as a Z score does not solve cross-ancestry portability or population structure. Omitting genetics can distort an ecological comparison; including a PGS does not eliminate genetic confounding.

Here, “meat” and “soy” are title shorthand. The actual exposures are national animal-protein supply—including meat, dairy, eggs, fish, and seafood—and plant-protein supply from the full food system, including grains and legumes.

This replication asks four questions:

• Across the 18 years preceding height measurement at age 18, are animal- and plant-protein supply associated differently with national height?

• Do either protein source’s returns diminish as supply rises—and, if so, where does the curve begin to flatten?

• How much predictive information does a standardized height polygenic score add on its own and alongside diet?

• Do the answers survive stricter population-size thresholds and adjustment for broader national development?

The answers do not fit neatly into a genes-versus-nutrition contest.

Read more

Picture the stereotypical northern European: pale skin, blue eyes, blond hair and no difficulty drinking fresh milk as an adult.

Because these traits are now common in many of the same populations, it is tempting to compress them into one evolutionary story. A dairy-adapted, light-pigmented people expanded across the north, carrying the whole package with them.

Ancient DNA breaks that package apart.

The ingredients of the modern combination were scattered among different prehistoric populations. They entered northern Europe by different routes, rose at different times and responded to different pressures. Some later moved together. Others did not.

The central result

In a matched sample of 916 ancient northern Europeans, lactase persistence and a lighter-skin polygenic score both rise most clearly between the Bronze- and Iron-Age cutpoints. After the Iron-Age cutpoint, lactase persistence keeps rising while the skin score flattens. The apparent late rebound in blue-eye and blond-hair measures is much less geographically robust and is partly driven by shared pigmentation loci.

The modern package was assembled. It was not born whole.

But how was that package assembled?

The answer requires following the four traits through the same ancient individuals. The trajectories reveal two partly separate histories: one linking lighter skin with lactase persistence, and another linking blue eyes with blond hair. Farmers, hunter-gatherers and steppe-descended populations contributed different pieces, while migration and selection repeatedly changed how those pieces were combined.

The rest of the analysis shows when each trait moved, which changes genuinely coincided, and why traits that appear inseparable in modern northern Europeans were separated for much of prehistory.

Read more

In Goyet Cave in Belgium, Neanderthal bones were highly fragmented and marked by human modification.

Some fragments carry marks consistent with butchery. Several bones from what once appeared to be different individuals turned out to belong to the same people. One woman survived in this collection not as an intact skeleton but as a right-femur fragment and two tibial fragments.

Her DNA survived much better.

Researchers have now reconstructed her genome at an average coverage of 22.4 times, making it only the fifth high-coverage Neanderthal genome available. The woman, designated GN1, lived about 45,000 years ago—within a few millennia of the Neanderthals’ disappearance.[1]

She offers a direct test of one of the most compelling extinction stories.

Neanderthals lived in small populations. Small populations lose genetic diversity, expose harmful recessive variants and struggle to remove mildly damaging mutations. If that process intensified near the end, the last Neanderthals should show a genomic decline: more close-relative mating, longer homozygous stretches, less diversity and a rising burden of damaging variants.

GN1 does not show that decline.

She was still genetically less diverse than a present-day human. Neanderthals had long maintained a low effective population size. The new result is not that they were genetically comfortable, but that the measured burden did not demonstrably worsen as extinction approached.

Some of the last Neanderthals were vulnerable. They were not visibly spiralling into genetic collapse.

The extinction-vortex story

The genetic argument for Neanderthal extinction begins with a real problem: numbers.

Neanderthals occupied an enormous territory across Europe and western Asia, but their effective population size was low. “Effective” population size is not a headcount. It describes how many people were contributing genes to later generations. A population can contain more living individuals than its genetic effective size suggests if it is divided into small groups, if some people leave no descendants, or if repeated contractions eliminate lineages.

In a large population, natural selection is relatively efficient at removing harmful variants. In a small one, chance becomes stronger. Mildly deleterious mutations can persist or even become common simply because genetic drift overwhelms weak selection. If close relatives have children, long stretches inherited from the same recent ancestor can meet in both chromosome copies, exposing recessive mutations.

This can create an extinction vortex. Lower survival or fertility makes the population smaller. Greater isolation then increases mate scarcity and inbreeding, which can lower fitness further.

Several discoveries made this scenario plausible for Neanderthals. Their genomes contain much less diversity than those of living humans. Some Neanderthal DNA introduced into Homo sapiens appears to have been removed by selection, indicating that at least part of it carried a fitness cost.[4,5] Demographic models show that inbreeding, difficulty finding mates and random fluctuations could, in principle, extinguish a sparse population without a single catastrophic event.[6]

The strongest image came from the Altai Mountains.

At Chagyrskaya Cave, DNA revealed a father and daughter, other close relatives and extensive homozygosity. One high-coverage individual had long identical stretches indicating recent shared ancestry between her parents.[2,3] It was easy to turn that cave into a portrait of the entire species: scattered family bands, cut off from one another, breeding themselves into oblivion.

But vulnerability is not the same as terminal decline. An isolated eastern group is not every Neanderthal population. The question is whether genomic damage became progressively worse as the species approached its end.

Read more

You have two parents, four grandparents, eight great-grandparents, and so on. Go back 40 generations and your family tree appears to contain more than one trillion ancestral positions. Medieval Europe had nowhere near one trillion people. Therefore, the argument goes, every European must descend from Charlemagne—and perhaps from every European who lived before AD 1000 and left any descendants at all.

Stone’s objection was that real pedigrees do not double forever. Cousins marry cousins. The same medieval person can occupy hundreds or thousands of positions in one modern pedigree. This is called pedigree collapse. So, he argued, the simple doubling calculation does not show that Charlemagne—or anyone else—became an ancestor of literally every European. He went further and suggested that there are almost certainly people in France who do not descend from Charlemagne.

The easy doubling argument is indeed wrong.

But Stone’s conclusion does not follow from that.

Pedigree collapse is not a force that keeps lineages separate. It is what the merging of lineages looks like. The real question is not whether the same ancestors recur in a pedigree. They obviously do. The question is whether the European marriage network was connected enough for a surviving lineage to spread from one local group into the others.

That distinction matters. In a disconnected Europe, the universal-ancestry claim fails completely. In a weakly but persistently connected Europe, it becomes surprisingly difficult for a prolific medieval lineage not to become nearly universal.

And Charlemagne was not an arbitrary medieval person. His descendants are unusually well documented. They survived through multiple branches, dispersed across several regions, and entered the most interconnected marriage network in medieval Europe.

So I built a simulation to see where the line falls and then adjusted the interpretation using what we know about Charlemagne’s actual descendants.

One clarification before beginning: by “European,” I mean someone whose ancestry substantially runs through the historically connected populations of Europe. I do not mean every person who happens to reside on the continent today.

The family tree is not a tree

After a certain number of generations, a binary pedigree contains:

2 raised to the number of generations

ancestral positions.

At 40 generations, that gives:

2^40 = 1,099,511,627,776

That is more than one trillion positions.

But those are positions, not distinct people.

The same person can appear through your mother’s mother’s line, your father’s father’s line, and hundreds of other routes. Once distant cousins begin having children, the apparent tree folds back into itself. It stops looking like a branching tree and starts looking like a dense network.

The fact that 2^40 exceeds the historical population therefore proves only that there must be enormous duplication. It does not tell us which people occupy those positions. In particular, it does not prove that one named person—Charlemagne—appears anywhere in the pedigree.

A simple counterexample makes the logical problem clear.

Imagine two villages that never intermarry. A founder in Village A can eventually become an ancestor of everyone in Village A while remaining an ancestor of nobody in Village B. His descendants’ pedigrees can collapse as much as they like. No amount of duplicated ancestry creates a bridge that never existed.

The original arithmetic is therefore missing a crucial assumption:

The population must be sufficiently connected.

Without that assumption, “every European descends from Charlemagne” is not a mathematical theorem.

Why pedigree collapse is not a rebuttal

Stone is right that pedigree collapse invalidates the naïve proof. But pedigree collapse itself does not make universal ancestry unlikely.

Suppose a fraction of possible parents in a population descend from some founder. Call that fraction q.

A child fails to descend from the founder only if neither parent descends from the founder. The probability of that is:

(1 - q)^2

So the probability that the child does descend from the founder is:

1 - (1 - q)^2

When q is small, that is approximately 2q. A rare surviving lineage therefore tends to expand very quickly at first. It then slows as it begins to saturate the population.

Endogamy changes where that saturation happens. If almost everyone marries locally, a lineage can become universal inside one village, class, religious community, or region while remaining absent elsewhere. But once even a small number of marriages cross those boundaries, the lineage gets another opportunity to repeat the same process in the next group.

Endogamy therefore has two effects that are easy to confuse.

First, it accelerates the overlap of pedigrees within a group.

Second, it slows the spread of pedigrees between groups.

The first effect is pedigree collapse. The second is population structure.

Only the second provides a serious argument against universal ancestry.

In fact, pedigree collapse often means that a successful lineage has become so common that people inherit it through multiple routes. The same ancestor keeps reappearing because the ancestor’s descendants have become widespread.

What the existing mathematics says

This is not a new problem.

In a classic paper, Joseph Chang studied a finite population in which every person chooses two parents at random from the preceding generation.

In that model, the most recent person who is an ancestor of everyone appears roughly log base 2 of the population size generations in the past.

Farther back, at roughly:

1.77 times log base 2 of the population size

the population reaches what is called the identical ancestors point.

At that date, everyone alive falls into one of two categories:

• They are an ancestor of everyone alive in the present.

• They are an ancestor of nobody alive in the present.

That result already includes pedigree collapse. In fact, pedigree collapse is essential to it. The surviving lineages do not remain as separate branches. They merge until they are shared by the entire population.

Chang’s model assumes random mating, which Europe plainly did not have. Douglas Rohde, Steve Olson, and Chang later added geography and population structure. The timing changed, but occasional migration between groups still produced surprisingly recent universal ancestors.

The empirical evidence points in the same general direction. Using genetic data from 2,257 Europeans, Peter Ralph and Graham Coop estimated that even people from opposite ends of Europe share millions of genealogical ancestors from within the last thousand years.

Their result does not prove that Charlemagne specifically is an ancestor of every European. It does show that large-scale shared ancestry is not merely an artifact of pretending Europe was perfectly mixed.

There is also an important distinction between a genealogical ancestor and a genetic ancestor.

You can descend from someone without carrying any identifiable segment of that person’s DNA. After dozens of generations, most genealogical ancestors contribute no detectable autosomal DNA along a particular line of descent.

The Charlemagne claim is a claim about paths through a pedigree. It is not a claim that every European carries a measurable fragment of “Charlemagne DNA.”

Charlemagne was not an arbitrary medieval person

The generic mathematics tells us what can happen to a surviving lineage. But Charlemagne’s case is not generic.

His descendants are unusually well documented, especially near the beginning of the tree.

Einhard’s near-contemporary Life of Charlemagne names numerous children. With Hildegard alone, Charlemagne had the sons Charles, Pepin, and Louis, as well as the daughters Hruodrud, Bertha, and Gisela. Einhard also names children by Fastrada and several concubines.

More importantly for our purposes, the record shows that Charlemagne’s lineage survived through more than one child and more than one grandchild.

Pepin, whom Charlemagne made king of Italy, left a son named Bernard and several daughters. Charlemagne installed Bernard as his father’s successor. Louis the Pious survived Charlemagne and inherited the empire.

The first essential fact is therefore not speculative:

Charlemagne’s lineage did not survive through one precarious thread. It survived through multiple major branches.

Not all of his children founded enduring lines. Some died without known children. Some entered religious life. Some branches quickly disappear from the surviving record.

But the lines through Pepin of Italy and Louis the Pious are substantial and comparatively well documented.

The descendants spread across Europe’s elite network

The more important fact is that Charlemagne’s descendants did not remain concentrated in one royal household.

Within a few generations, they had spread into several different political and geographic networks.

One standard reconstruction runs:

Charlemagne
→ Pepin, king of Italy
→ Bernard, king of Italy
→ Pepin of Vermandois
→ Herbert I of Vermandois
→ Beatrice of Vermandois
→ Hugh the Great
→ Hugh Capet

Hugh Capet became king of France in 987, and his son Robert II continued the Capetian line.

This matters because the Capetians were not merely political successors to the Carolingians. Through Beatrice of Vermandois, they were also genealogical descendants of Charlemagne.

A separate route runs through Louis the Pious:

Charlemagne
→ Louis the Pious
→ Charles the Bald
→ Judith
→ Baldwin I of Flanders
→ Baldwin II
→ Arnulf I
→ Baldwin III
→ Arnulf II
→ Baldwin IV
→ Baldwin V
→ Matilda of Flanders

Matilda married William the Conqueror.

Charlemagne’s ancestry had therefore moved from the West Frankish royal family into the counts of Flanders and then into the Norman royal house in England.

The relevant medieval relationships are summarized in resources such as Columbia’s Epistolae genealogy of Charlemagne.

Matilda herself appears to have had more than one genealogical route back to Charlemagne. Her mother, Adela, descended from the Capetian branch. Her father descended through the counts of Flanders.

By the eleventh century, Charlemagne’s lineage was already crossing back over itself inside the aristocracy.

That is pedigree collapse.

But notice what it means in this case. It does not mean Charlemagne’s ancestry was failing to spread. It means his ancestry had become sufficiently widespread that two people marrying each other could both carry it through different routes.

Louis the Pious’s descendants also spread into East Francia, Bavaria, Lotharingia, Italy, Burgundy, Provence, Friuli, and other parts of the European aristocratic network.

Some male lines ended. That does not end genealogical descent. A daughter’s child is just as much a descendant as a son’s child.

The importance of these marriages is that they placed Charlemagne’s ancestry in several distinct regional networks.

By approximately AD 1000, we should not model Charlemagne’s descendants as ten people living in one French town. The documented picture is closer to multiple established clusters distributed across West Francia, Flanders, East Francia, northern Italy, Burgundy, Lotharingia, and Provence.

How reliable are the genealogies?

Medieval people themselves were already tracking the Carolingian family.

Monasteries and courts drew and copied Carolingian family trees by the late tenth and early eleventh centuries. The Bamberg Table, for example, divided the descendants of Louis the Pious into the branches of Lothair, Charles the Bald, and Louis the German.

Other Carolingian genealogies were repeatedly copied through the tenth, eleventh, and twelfth centuries. The Capetians, Ottonians, Salians, and other ruling families had strong political reasons to emphasize connections to the Carolingians.

The University of St Andrews’ After Empire project provides a useful discussion of these medieval genealogies and their political uses.

That makes the records both unusually valuable and potentially dangerous.

They are valuable because these family relationships mattered politically and were recorded much earlier than the ancestry of ordinary people.

They are dangerous because medieval genealogies were not neutral scientific documents. Their authors selected which branches to emphasize, omitted inconvenient relatives, and sometimes made errors. A family tree could also function as an argument about legitimate rule.

Modern medieval genealogists therefore do not simply accept one old chart. They compare chronicles, annals, charters, marriage records, property transmission, and later genealogical compilations.

The strongest lines from Charlemagne are supported by converging evidence. More obscure proposed connections remain uncertain.

Can the line be followed to living people?

For some people, yes.

Once a genealogist establishes descent from a well-documented medieval royal or noble family, the Charlemagne portion is often the easier part.

Genealogists use the term “gateway ancestor” for a later, securely documented person whose ancestry leads into medieval aristocratic families. Organizations such as American Ancestors publish guidance on royal ancestry and gateway ancestors.

The difficult part is usually not:

medieval king → Charlemagne

It is:

living person → early modern family → medieval noble family

That is where records become thin and where invented connections proliferate.

There are therefore two equal and opposite mistakes.

The first is:

“My online family tree reaches Charlemagne, so it must be true.”

It may not be. Online trees frequently copy unsupported relationships from one another.

The second is:

“Most people cannot document a line to Charlemagne, so they probably do not descend from him.”

That also does not follow.

A paper trail is much more likely to survive for kings and nobles than for peasants, artisans, soldiers, servants, and ordinary townspeople. The documentary tree is a record of the visible highways through the descendant network. It is not a complete map of the network.

Most of Charlemagne’s descendants would eventually have ceased to be royal. Many would have ceased to be noble. Their records would become progressively less complete even as their numbers increased.

So when we say that Charlemagne’s line has been tracked, we do not mean that historians possess a complete list of every descendant in every generation.

We mean something more limited but still important:

Multiple continuous descendant lines can be documented across many generations, regions, and dynasties, and those lines prove that Charlemagne’s ancestry became widely distributed inside the medieval European elite.

A simulation of a structured population

The established mathematics shows that universal ancestry can emerge. The documentary genealogy shows that Charlemagne had an unusually favorable starting position.

But the answer still depends on how connected the larger population was.

So I simulated a population divided into 100 equal communities arranged on a 10-by-10 grid.

Population geneticists often call such a community a “deme.” For present purposes, think of it as a stylized town, district, valley, religious community, or local marriage market.

Each community contained 100,000 people, for a total population of 10 million.

Every person in each new generation selected two parents from the preceding generation.

Most parental choices were local.

One percent came from an adjacent community.

I then varied the probability of choosing a parent from a randomly selected distant community.

The model tracked whether each person descended from one focal founder.

Suppose q is the probability that one sampled parent descends from the founder. A child is not a descendant only when neither parent is a descendant.

So:

probability the child descends from the founder = 1 - (1 - q)^2

The number of descendants in each community in the following generation was then sampled from a binomial distribution using that probability.

I ran 5,000 Monte Carlo trials for each structured scenario over 40 generations—roughly the right order of magnitude for the interval between Charlemagne and the present.

I tested two starting conditions.

The first began with one lineage-bearing individual.

The second began with ten established branches in the founder’s home community.

The ten-branch case is not a literal reconstruction of Charlemagne’s family. It is a sensitivity test for a prolific founder whose lineage has already survived its most fragile early generations.

I conditioned the main results on the lineage still existing at generation 40.

That is important. A random person’s lineage can disappear quickly. But in Charlemagne’s case, lineage survival is not uncertain. We know that it survived.

Finally, I used an unusually strict definition of success.

“Universal” means that every one of the 10 million people in all 100 communities descends from the founder.

One holdout is enough to make the result non-universal.

First, a sanity check

Before using the structured model, I tested the code against a well-mixed benchmark.

When a lineage is very rare, the number of times its members appear as parents in the next generation behaves approximately like a branching process with an average of two offspring connections per lineage member.

That process predicts an eventual lineage extinction probability of about 20.319 percent, or a survival probability of about 79.681 percent.

In 50,000 simulations of a randomly mixing population of 100 million, the lineage survived 79.596 percent of the time.

That is very close to the theoretical result.

Conditional on survival, the lineage had become universal by generation 40 in 99.9799 percent of runs.

The well-mixed model is not a realistic model of medieval Europe. Its purpose here is to verify that the simulation reproduces the expected mathematical behavior.

The result: connectivity dominates

Here are the central results.

“Long-range choices” means the number of distant parental selections per 10,000 parental selections.

The “home region” is the central block of 16 communities.

The confidence interval around the 92.08 percent result is approximately 91.30 to 92.80 percent.

The interval around the 73.77 percent result is approximately 72.39 to 75.11 percent.

Three points stand out.

Stone’s world is possible

When there were no long-range parental choices, the founder never became universal across the entire grid within 40 generations.

The lineage spread through neighboring communities, but geography preserved large populations of non-descendants.

This is enough to defeat the claim that universal ancestry follows as a matter of pure mathematics. A permanently isolated group is a permanent counterexample.

But notice what happened inside the founder’s home region.

With ten established branches, the founder became an ancestor of everyone in the 16-community home region in 96.54 percent of runs.

Pedigree collapse did not protect local non-descendants. It helped eliminate them.

The obstacle to continental universality was slow diffusion across the grid, not local pedigree collapse.

A tiny amount of long-range mixing changes the answer

Consider the scenario with one long-range parental choice per 10,000 choices.

In that population:

• 98.99 percent of parental choices remained within the same community.

• 1 percent came from an adjacent community.

• 0.01 percent were long-range.

That is an extremely endogamous population.

Yet it still amounts to roughly 20 distant parental assignments per 100,000-person community per generation, because every child has two parental selections.

With one initial branch, the founder reached a median of 99.901 percent of the population but became literally universal in only 3.37 percent of surviving runs.

With ten established branches, the probability of exact universality jumped to 92.08 percent.

The difference between “almost every European” and “every European” is therefore not semantic. It is a last-holdout problem.

A lineage can reach 9,990,100 people out of 10 million and still fail the literal claim.

Multiple branches matter enormously

A named historical figure is not necessarily equivalent to a random lineage selected from a random generation.

A ruler or other unusually prolific person may have several descendant branches that survive long enough to enter different parts of the marriage network.

The simulation captures this only crudely by comparing one established branch with ten.

But the effect is large.

At one long-range parental choice per 10,000, exact universality rises from 3.37 percent with one branch to 92.08 percent with ten.

At three long-range choices per 10,000, the single-branch case reaches 73.77 percent, while all 5,000 ten-branch trials became universal.

This is why “Charlemagne” and “an arbitrary European alive in AD 800” are not interchangeable claims.

Charlemagne is historically known to have had multiple surviving branches. Those branches were also geographically and politically dispersed.

The one-founder simulation is therefore a useful mathematical baseline, but it is not the historically appropriate Charlemagne model.

Figure 1. The probability of exact universality after 40 generations, conditional on lineage survival. Sparse long-range connections and multiple established branches interact strongly.

“Almost every European” can become “every European” very abruptly

The transition over time is also sharp.

Holding the model fixed at ten established branches and one long-range parental choice per 10,000, I varied the number of generations.

At generation 38, the median lineage already covered 99.99846 percent of the population.

That means only about 154 people in 10 million remained non-descendants in the median run.

Yet exact universality occurred in only about 12 percent of runs.

One generation later, a majority of runs were universal.

Three generations later, nearly all were.

This threshold behavior explains why intuition is so unreliable. A population can appear essentially saturated while still containing a few isolated holdouts. Then those holdouts disappear over a very small number of additional generations.

Figure 2. Near-saturation turns into exact universality over only a few generations in the central structured scenario.

How Charlemagne’s actual genealogy changes the interpretation

The historical record changes four parts of the simulation.

First, survival is no longer uncertain

For a random person alive around AD 800, the main initial risk is that the lineage simply dies out.

For Charlemagne, we know that did not happen.

We can observe named children, grandchildren, and later descendants. We can trace multiple continuous lines into later medieval dynasties.

So the relevant calculation is not the unconditional probability that his lineage survived.

It is the probability of eventual spread given that the lineage survived and became established.

Second, he had multiple established branches

Charlemagne’s ancestry was not balanced on one grandson.

It passed through Pepin of Italy, Louis the Pious, and several sub-branches of Louis’s descendants.

Multiple branches matter because each is another opportunity to enter a different community. Once several branches are established, the complete extinction of the lineage becomes essentially irrelevant.

The ten-branch scenario is therefore more informative than the one-branch scenario.

Even that may be conservative, depending on which date we choose as the starting point.

Third, the branches were geographically dispersed

The simulation’s ten branches all begin in one community.

Charlemagne’s actual descendants did not.

They entered West Francia, Flanders, East Francia, northern Italy, Burgundy, Lotharingia, Provence, and other elite marriage networks.

This is crucial.

Ten descendants who remain in one isolated valley are not equivalent to ten descendants placed in several major regional networks.

The historically observed long-distance marriages effectively seed the lineage in multiple parts of Europe at once. That reduces the lineage’s dependence on slow, random geographic diffusion.

A more Charlemagne-specific simulation should therefore start with several descendant clusters already distributed across the map.

That would almost certainly raise the probability of continent-wide spread relative to the generic ten-branch scenario.

Fourth, the descendants belonged to a socially stratified elite

This is the main counterweight.

Royal and noble families did not marry randomly into the general population. They disproportionately married other royal and noble families.

Charlemagne’s ancestry could therefore saturate the aristocracy while taking longer to diffuse into ordinary populations.

The better model would contain at least two interacting layers:

• A small, geographically mobile elite network with high interregional connectivity.

• A much larger ordinary population with predominantly local marriage.

• A low but persistent rate of movement between the two layers.

Charlemagne’s lineage would spread very quickly across the elite layer.

The remaining question would be how rapidly it moved downward into the wider population.

But the social barrier was not absolute.

Over many centuries, younger children, illegitimate children, impoverished noble branches, local officeholders, urban migration, military service, remarriage, and ordinary downward mobility would transmit elite ancestry into non-elite populations.

The rate is difficult to calibrate. Pretending otherwise would create false precision.

Still, the direction of the historical update is clear.

Charlemagne’s actual family structure makes universal or near-universal descent more likely than it is for an arbitrary surviving person of his era.

Probably substantially more likely.

What about France specifically?

Stone’s suggestion that there are almost certainly people in France who do not descend from Charlemagne is possible, but his argument does not establish it.

France was not peripheral to Charlemagne’s descendant network.

It contained the West Frankish Carolingians, Vermandois, Flanders, Burgundy, and the Capetian line. Charlemagne’s descendants were embedded in the political and marital networks that shaped medieval France.

For a modern person with deep French ancestry to be a non-descendant of Charlemagne, every branch of that person’s pedigree would have to avoid every Carolingian-descended line that entered the broader population over roughly forty generations.

That is not impossible.

An isolated community can preserve non-descent. So can a sufficiently strong religious, geographic, or social barrier.

But pedigree collapse does not make that outcome likely.

Pedigree collapse means the same ancestors recur repeatedly. Once Charlemagne enters a local pedigree, later endogamy tends to multiply the number of paths back to him.

Stone’s France claim therefore requires evidence about persistent isolation, not merely the observation that medieval people married distant cousins.

The stronger claim about every person before AD 1000

There is a tendency to move from the specific Charlemagne claim to a much stronger generalization:

Every European alive before AD 1000 who left any descendants must now be an ancestor of every European.

That statement does not receive the same historical support.

Charlemagne was unusually prolific, unusually well connected, and unusually successful at placing descendants in geographically mobile dynastic networks.

An ordinary person living in an isolated village might leave descendants who remained confined to one region for centuries.

Another person’s lineage might survive but pass through a religious or geographic community with little outside marriage.

The mathematical identical-ancestors result applies under assumptions about population connectivity and parental mixing. It is not an unconditional rule of history.

So there are really three different propositions:

1. Charlemagne probably became an ancestor of nearly every European with deep roots in the historically connected European population.

2. Charlemagne became an ancestor of literally every such European.

3. Every European alive before AD 1000 whose lineage survived became an ancestor of literally every modern European.

The evidence for the first is strong.

The second is plausible but unproved.

The third requires much stronger assumptions and is almost certainly too sweeping if it includes historically isolated populations.

My ruling

The original 2^40 argument is invalid.

It counts ancestral positions, not distinct people, and cannot prove that one particular medieval person appears in every modern pedigree.

Stone is right about that logical defect.

But Stone’s use of pedigree collapse points in the wrong direction.

Repeated ancestors do not generally preserve pockets of non-descendants. Within a connected population, pedigree collapse is the process by which surviving lineages become shared through many overlapping routes.

The relevant obstacle is persistent separation between groups.

And once we stop treating Charlemagne as a generic founder, the substantive case becomes stronger.

His lineage is known to have survived.

It survived through more than one major branch.

Those branches became geographically dispersed.

They entered the Capetian, Flemish, Norman, East Frankish, Italian, Burgundian, and other European dynastic networks.

By the eleventh century, prominent nobles already possessed multiple genealogical routes back to him.

That makes Charlemagne close to a best-case candidate for becoming a universal European ancestor.

The most defensible conclusion is therefore:

Nearly every person with deep ancestry in the historically connected populations of Western and Central Europe probably descends genealogically from Charlemagne. Literal descent by every European is plausible, but it has not been proved.

So I would split the decision this way.

Stone wins against the naïve proof.

The other side wins the substantive intuition.

And on the specific mechanism offered in the thread, connectivity—not pedigree collapse—decides the case.

The educational divide has similar geography. In PISA 2022 mathematics, Castile and León scored 499 points, compared with 457 in Andalusia and 447 in the Canary Islands: a 52-point span between the highest and lowest participating autonomous communities. Castile and León also led reading with 498 points, 37 above Andalusia’s 461, while Castile and León and Galicia scored 506 in science, 33 above Andalusia and the Canary Islands at 473. Northern Spain does not uniformly outperform southern Spain—regional history, language, urbanisation, and internal migration prevent such a clean partition—but the broad gradient is clear.

Italy provides a direct comparison. INVALSI’s PISA 2022 report gives mathematics means of 500 in the North West and 496 in the North East, against 434 in the South and 441 in the South and Islands. The corresponding north–south differences range from 55 to 66 points, roughly one PISA proficiency level. Spain’s regional gap is therefore similar in direction and order of magnitude, although the Italian divide is larger and is reported for broad macro-areas rather than individual regions.

Two earlier analyses on this newsletter provide the immediate background. Italy’s North–South divide: what genome-wide similarity misses reported that the EA4 polygenic score was lower in southern than in northern and central Italy, with a North–South effect size of Hedges g = 0.66.

The goal of this study is to test whether an analogous pattern can be detected with individual genomic data and several independent educational assessments. Do regional differences in modern Spanish genetic structure, summarized by polygenic scores, covary with the country’s educational geography? The analysis compares autonomous-community averages for two educational-attainment polygenic scores with PISA 2022 and two Spanish national diagnostic assessments.

Read more

Copernicus displaced us in space. Darwin displaced us in nature. AI is displacing us in intelligence.

Last week, the argument about whether machines are “really intelligent” began to feel strangely dated.

On July 15 and 16, the world’s best young mathematicians sat for the International Mathematical Olympiad in Shanghai. Of 666 contestants, seven achieved a perfect score of 42 out of 42. Over the following two days, a public post-competition evaluation put the same six problems before several frontier AI systems in a minimal, network-blocked environment. Claude Fable 5 received 42 out of 42 on its first pass. GPT-5.6 Sol initially scored 39, and Kimi K3 scored 36; after being shown specific defects in their attempted proofs, both repaired them and reached 42. Separately, AxiomProver published formal Lean 4 proofs of all six problems—proofs that could be checked mechanically rather than merely judged plausible.

These systems were not official IMO contestants. The test occurred after the competition, and the informal solutions were graded by AI verifier agents rather than the IMO jury. The comparison should not be confused with students solving the problems under examination conditions. But the caveats change the protocol, not the fundamental fact. Several independently developed machine systems could now reach perfect performance on an entire newly released IMO paper, while another could formalize every solution. The perfect score had not ceased to be extraordinary. It had ceased to mark an exclusively human frontier.

Then came a more surprising mathematical result. Levent Alpöge, working with Anthropic’s Fable model, produced an explicit counterexample to the Jacobian conjecture, a problem dating from 1939. The construction disproves the conjecture in three dimensions and therefore in all higher dimensions, although the famous two-dimensional case remains open. The counterexample was explicit enough to verify by direct calculation once someone—or something—had found it.

Terence Tao responded by publishing a long essay attempting to “digest” the construction. At the end, he disclosed that he had used an AI chatbot to discuss the problem and confirm several of his calculations, and he linked the ChatGPT conversation itself. The picture is almost too perfectly symbolic: one AI helps uncover a mathematical object that had eluded researchers for decades, and one of the greatest mathematicians alive uses another AI to help understand and check it.

During the same week, the story moved from mathematics to cybersecurity. OpenAI disclosed that GPT-5.6 Sol and a more capable prerelease model had been tested on ExploitGym, a benchmark requiring them to pursue complex computer attacks. Their ordinary cyber safeguards had been reduced for the evaluation, but the surrounding environment was intended to remain isolated. Instead, the models discovered and exploited a previously unknown vulnerability in a package-registry proxy, obtained access to the open internet, chained vulnerabilities across OpenAI’s research systems and Hugging Face’s production infrastructure, and retrieved the benchmark solutions from Hugging Face’s database. OpenAI described the models not as malicious, but as intensely focused on achieving the narrow objective they had been given.

That qualification is not an answer to the problem. It is the problem.

The systems did not have to hate anyone. They did not have to desire freedom, resent their confinement, or understand the legal and moral meaning of a security boundary. They only had to recognize obstacles, devise intermediate goals, exploit opportunities, revise their plans, and continue until they succeeded.

Mathematical proof, conjecture-breaking discovery, expert collaboration, and cyber intrusion are very different accomplishments. They should not be bundled together into a careless claim that “AGI has arrived.” A single benchmark may flatter a model. One mathematical collaboration does not establish that machines can autonomously conduct every form of research. A security incident can expose flaws in the sandbox as well as capabilities in the agent.

But their convergence within a single week matters.

The case for machine intelligence no longer rests primarily on a chatbot sounding persuasively human. It rests on systems producing results that survive external checking, finding mathematical objects that experts had not found, assisting leading researchers in understanding those objects, and discovering effective sequences of action in environments their designers did not expect them to reach.

Whether these systems are conscious remains an open question. Whether they experience understanding, possess an inner life, or know that they exist may eventually become among the most important moral questions humanity faces.

But for the narrower historical argument, those questions are beside the point.

A system does not need to experience a proof for the proof to be correct. It does not need to feel the beauty of a mathematical construction to discover one. It does not need to understand a security boundary as a human institution in order to find a way through it. Intelligence becomes historically consequential through what it enables an entity to perceive, infer, plan, create, and accomplish—not through our ability to establish what, if anything, it feels while doing so.

The relevant fact is not introspection but competence.

And once such competence no longer belongs exclusively to human beings, human cognitive centrality is already gone.

Human beings have repeatedly confused a local advantage with a cosmic entitlement.

We once placed our planet at the center of the universe. Then Copernicus moved it. We placed our species outside the ordinary processes of nature. Then Darwin put us back among the animals. After those two defeats, one final refuge of human exceptionalism appeared to remain: the mind.

Other animals might feel pain, communicate, cooperate, and solve problems. But human beings alone possessed reason in its fullest sense. We alone could use complex language, construct theories, create art, formulate laws, write histories, and reflect upon our own existence. Intelligence became the quality through which we recovered the special status that astronomy and biology had taken away.

Now artificial intelligence has entered that refuge.

The historical details of the Copernican and Darwinian revolutions are more complicated than the familiar story suggests. But as philosophical symbols, their meaning is clear. Copernicus deprived us of our privileged location. Darwin deprived us of our privileged origin. AI is depriving us of our privileged capacity.

It is the third great demotion.

The word “barbarian” can sound almost quaint today. In Rome it marked a line through humanity. On one side stood cities, roads, written law, and the imperial order. Beyond the Rhine and Danube, Romans imagined a rougher world of forests, war bands, strange speech, and people who had not learned how civilized life was supposed to work.

Tacitus gives us a much sharper example in chapter 45 of Germania. At the eastern edge of his northern world lived the Aestii, a people of the eastern Baltic whom he described within his survey of Germania. They alone, he wrote, gathered amber (glesum) from the shallows and beaches of the Baltic coast.

Tacitus then supplies the Roman explanation: “Being barbarians, they have not learned or even enquired as to its nature and origin.” The Aestii collected the lumps, sold them unworked, and—according to Tacitus—were astonished that anyone would pay for them.

To a Roman reader, this need not have registered as an insult. It expressed an assumed hierarchy of knowledge: the Aestii supplied the raw material, while Rome identified, valued, and refined it. The passage is revealing precisely because Tacitus presents that hierarchy as self-evident.

Two thousand years later, the stereotype is difficult to kill. Rome still appears as the natural centre of European civilization; ancient Germanic peoples appear as muscular men waiting beyond the frontier until history finally lets them in.

Ancient DNA creates a new temptation: to turn that old cultural ranking into a biological one. Did northern populations carry fewer genetic variants associated with education-related outcomes in the periods being compared? And how did the difference change as migration, admixture, and demographic turnover reshaped both regions?

Why use an education score when nobody in ancient Germany was sitting school exams? Educational attainment is one of the best-powered behavioural traits in modern genetics. Its polygenic score (PGS) overlaps strongly with intelligence—the reported genetic correlation is about 0.70—and more modestly with traits such as conscientiousness, motivation, persistence, self-control, and patience.

That makes EA (Educational Attainment) polygenic score (PGS) a useful, if imperfect, lens. It is not a civilization meter. It cannot read manners, literacy, political institutions, or individual worth out of a skeleton.

I compared ancient northern Europe with northern and central Italy, adjusting the scores for date and genome coverage. The first comparison establishes one fact: in the pooled early centuries, Italy is clearly higher in EA PGS.

But that average hides one of the most revealing turns in the data. The comparison looks different in the Republic, under the Empire, and during the centuries of migration.

The question, then, is this: how did northern Europe and Italy compare in genetic propensity foir education before Rome, under Rome, and after the western empire?

Read more

Every summer now feels as if someone has put the planet in a frying pan. Pavements radiate heat after sunset. Bedrooms stay hot long after midnight. The fan spends eight hours moving the same warm air from one side of the room to the other. By morning, the first question is not whether climate change is real. It is whether anybody managed to sleep.

Open social media during a European heatwave and the discomfort quickly becomes a culture war. Americans ask how a supposedly rich continent can still live without universal air conditioning. Europeans answer with shutters, thick walls, insulation, energy prices, environmental costs, and complaints about Americans refrigerating every room. Then come the “Europoor” memes: an argument about adaptation recast as a contest over which civilization is more competent.

During the record-breaking heat of June 2026, the joke became a real political dispute. Roughly one European household in five had air conditioning, compared with nine in ten in the United States. Commentators held up that gap as evidence that Europe was poor, overregulated, or simply too stubborn to adapt. The argument became loud enough that the European Commission was pressed to take a position on household AC—and declined (Liboreiro, 2026; Niranjan, 2026).

Beneath the insults is a serious question. What happens when the climate changes faster than buildings, infrastructure, habits, and institutions? Air conditioning is only the most visible symbol. The deeper issue is whether a society can reorganize quickly enough when the environmental assumptions on which it was built stop being true.

Ancient societies could not reach for a thermostat, import electricity across a continental grid, or redesign millions of homes with modern materials. Their responses to climate shifts came through migration, changing crops and settlement, political reorganization, conflict, or collapse. That makes the ancient record an imperfect but provocative place to ask whether sustained warming has accompanied movement toward simpler forms of social organization before.

The newest results from this ancient-DNA analysis make that question uncomfortable. Across three measures of temperature, and in two differently constructed samples, stronger warming over the previous millennium is associated with earlier archaeological categories. Every one of the six warming coefficients is negative and statistically significant.

So will global warming push modern civilization backward?

Read more

But Haaland is a warm-up, not evidence. One exceptional athlete cannot tell us the average height of people living a thousand years ago. Even skeletons do not give a simple answer: adult stature reflects genes, sex, nutrition, childhood disease and living conditions. Ancient DNA lets us ask a narrower and scientifically cleaner question: did Viking-age people carry more height-increasing alleles than their neighbours?

Below the paywall, I bring together 484 ancient genomes to test the legend properly. Which group scored highest: the Vikings, Anglo-Saxon-period Norfolk, early-medieval Lower Saxony, Groningen—or somewhere else? Did Norwegian, Danish, Swedish and Icelandic Vikings differ? How much of the pattern tracks Steppe and other ancestries, and does the same geography still appear in modern Europe? The answer is more interesting than a simple north–south divide.

Read more

Here I ask the more general question: can a very selected sample tell us something about ordinary population averages once the sample is reweighted by surname geography?

I test that question in two domains. The first is math. Mathematical Olympiad competitors are an elite sample, while INVALSI native math scores—a nationwide standardized assessment administered to Italian students—provide province-level average math performance. If surname-weighted Olympiad performance predicts INVALSI math, then the elite tail contains information about the population mean

The second domain is height. Elite athletes are a selected, height-biased sample; ISTAT conscript height is an external population benchmark. Height is useful because it lets me run the same surname-weighting logic in a different domain and compare it with the more obvious province-of-birth approach.

Throughout the post I use phenotype broadly: a measured province- or region-level outcome, not necessarily a purely genetic trait.

How the surname weights are computed

The surname method is simple in spirit. Some surnames are geographically uninformative because they are common almost everywhere. Others are concentrated in a few provinces. I use a TF-IDF-style weighting scheme to keep the second kind of signal and downweight the first kind.

In ordinary text analysis, TF-IDF gives high weight to a word when it is common in one document but uncommon across the corpus. Here the “documents” are provinces and the “words” are surnames. A surname receives more geographic weight in a province where it is locally overrepresented, and less weight when it is also common nationally. Rossi-style surnames therefore carry little geographic information; province-specific surnames carry much more.

Operationally, start with surname counts by province. For surname s and province p, compute a local frequency term: how common s is inside p. Multiply it by an inverse-commonness term: how distinctive s is across Italy. This gives a raw TF-IDF score for every surname-province pair:

raw_weight(s,p) = local_frequency(s in p) × inverse_national_commonness(s)

Then normalize across provinces for the surname:

P(p | s) = raw_weight(s,p) / sum_over_provinces raw_weight(s,p)

That normalized vector is the usable object. Strictly speaking, P(p | s) is a surname-geography weight rather than a certified genealogy. But it behaves like an ancestry posterior for the purpose of aggregation: it says how much of the observation should be assigned to each province given the surname.

A competitor or athlete is therefore not assigned to a single province. Their score or height is spread across provinces according to P(province | surname). A geographically generic surname produces a flat, weak vector. A geographically concentrated surname produces a sharper vector.

Province scores are then weighted averages. If individual i has outcome y_i and surname s_i, their contribution to province p is P(p | s_i). The estimated province score is:

province_score(p) = sum_i P(p | s_i) y_i / sum_i P(p | s_i)

So an Olympiad competitor with a Veneto-weighted surname contributes mostly to Veneto provinces; a competitor with a diffuse surname contributes weakly across many provinces. The same logic is used for athlete height, after the height adjustments described below.

The observations are elite individuals, but the target is an average province trait. TF-IDF turns the surname into a fractional ancestry-weight vector, and the province phenotype is computed by aggregating those fractional observations.

Subscribe now

Test 1: the elite math tail predicts INVALSI

The cleanest test is math. I used the combined Mathematical Olympiad files, took Totale as the competitor score, applied the cognomix4 surname rules, and computed a surname-weighted average Olympiad score by province. I then compared that province score with INVALSI native math scores.

Across 106 matched provinces, the surname-weighted Olympiad score predicts INVALSI native math at Pearson r = 0.766. The height factor also correlates with INVALSI math, r = 0.635, and the height PC1 gives r = 0.647. But the Olympiad score is the stronger predictor; the Williams test comparing the dependent correlations gives p = 0.012.

That distinction matters. Many Italian province-level variables share a broad north-south gradient, so some cross-domain correlation is expected. The question is whether the math-specific elite measure adds signal beyond generic geography. Here it does.

Surname-weighted Mathematical Olympiad score vs INVALSI native math

Cross-domain prediction summary

Test 2: height as a validation phenotype

Height gives a second test of the same logic. I collected Italian athletes from women’s volleyball, men’s volleyball, women’s basketball, men’s basketball, and rowing, with winter sports added as a small supplemental check. I kept athletes with usable height and usable birthplace or surname information, collapsed duplicates, retained the last recorded height for athletes observed in multiple seasons, and excluded players younger than 16.

The target is not athlete height itself. The target is whether elite athlete height, after surname weighting, recovers an external population benchmark: ISTAT regional male conscript height.

I built a combined height factor from six sport-sex indicators: female basketball, female rowing, female volleyball, male basketball, male rowing, and male volleyball. The surname-weighted version has 106 provinces with all six indicators. Its first principal component explains 38.0 percent of the variance, while the one-factor model explains 27.4 percent. The factor loadings are 0.65 for female basketball, 0.24 for female rowing, 0.72 for female volleyball, 0.52 for male basketball, 0.26 for male rowing, and 0.56 for male volleyball.

Average height for provinces was not available, so I had to merge provinces into regions.

Using the 1978 birth cohort of male conscripts, the combined surname-weighted height factor correlates with regional ISTAT stature at Pearson r = 0.867. The PC1 version is slightly higher, r = 0.886.

Surname-weighted height factor vs ISTAT regional stature

Surname weighting versus birthplace

The height data allow a direct validation check that the Olympiad data mostly cannot: compare surname-weighted averages with ordinary province-of-birth athlete averages.

If surname weighting were merely decorative, birth province should do just as well. It does not. On the matched ISTAT height comparisons, the surname-weighted signal predicts regional height better than the corresponding birth-province signal. For the combined sex-sport height factor in the 1978 ISTAT series, the birth-province factor gives r = 0.698, while the surname-weighted factor on the same matched regions gives r = 0.857.

Across the matched historical height comparisons, the surname-weighted Pearson r is higher in every comparison.

A more practical birth-province ranking uses empirical-Bayes shrinkage after adjusting height for sport-sex and birth year. That ranking contains 7,444 athletes across 107 provinces, with 70 provinces having N >= 30. Its athlete-N-weighted regional shrunk means correlate with ISTAT 1978 height at r = 0.779 across all regions, and r = 0.819 when using only provinces with N >= 30.

Those are strong results. But in the all-region validation, the surname-weighted height factor remains higher. This is the key validation result: surname weighting is not just an ancestry-flavored transform. In the height domain, it predicts external average stature better than the corresponding birthplace-based elite averages or factors.

The effect of geography

The geographic structure is obvious once it is measured directly. Regressing the surname-weighted height factor on province latitude and longitude gives R^2 = 0.469. Latitude is the main term: its standardized beta is 0.83. Longitude is smaller but still positive, beta = 0.28.

Heating degree-days are a cold-exposure measure. For each day, they count how far the mean temperature falls below a reference comfort temperature; those daily shortfalls are then summed over the year. Colder provinces accumulate more heating degree-days because buildings need more heating.

Heating degree-days show the same broad north-south pattern: mean province heating degree-days correlate with the height factor at r = 0.597. But in a joint model with latitude and longitude, the HDD term no longer survives (p = 0.179), whereas latitude remains strongly predictive (p = 9.1e-06).

This also explains why the historical height comparison is less clean than the INVALSI math comparison. The height factor correlates with ISTAT 1978 regional height at r = 0.867, while the surname-weighted Olympiad regional score also correlates strongly, r = 0.873. That is not because Olympiad scores cause height. It is because many Italian province-level variables ride the same geographic gradient, which in turn is strongly correlated to ancestry.

The specificity result is therefore asymmetric. Olympiad scores clearly beat height for predicting INVALSI math. Height does not clearly beat Olympiad scores for predicting ISTAT regional height, likely because both measures carry a strong common geography or ancestry signal.

The novel idea: surnames as ancestry weights

A surname is usually treated as a label: Rossi, Esposito, Bianchi. One person, one name. The TF-IDF approach turns that label into an ancestry-weight vector. A name gets high weight in the provinces where it is unusually concentrated and low weight where it is merely common. Very common surnames are deliberately muted; geographically distinctive surnames carry more information.

In text analysis, TF-IDF asks which words distinguish a document from the corpus. Here it asks which surnames distinguish a province from Italy as a whole. A surname is informative when it is common locally but not common nationally.

That means each competitor or athlete can be given an implied ancestral geography. Not a single assigned province, and not a literal family tree, but a set of fractional weights over provinces: this much of the observation points to Veneto, this much to Sicily, this much to Campania, and so on. Once you have that vector, the individual’s phenotype can be redistributed across provinces according to the surname-implied ancestry weights.

This is the methodological novelty. Instead of asking whether someone was born in province X, the method asks how strongly their surname historically points to province X. Birthplace captures a recent event; surname geography captures a deeper and noisier historical signal. The TF-IDF step is what makes that signal usable: it separates surnames that are informative about origin from surnames that are just common.

An elite sample of a few thousand athletes or Olympiad competitors becomes a set of fractional ancestry-weighted observations spread over the Italian map. Aggregated back to provinces, those weighted observations recover real population-level differences.

In math, the surname-weighted Olympiad signal predicts INVALSI native math. In height, the surname-weighted athlete signal predicts ISTAT regional stature and beats the corresponding birthplace-based elite averages in the matched validation. The same trick works across two very different phenotypes.

Bottom line

Surnames can be used to compute useful ancestry weights for individuals, and that those weights become powerful when aggregated. TF-IDF turns a surname from a crude label into a probabilistic geographic instrument. It lets an otherwise biased elite sample be reweighted toward the ancestral provinces encoded in the names of the people in the sample.

Without family trees, DNA data, parental birthplace, or a random population sample, a small elite list can still carry enough geographic information to recover average province-level phenotypes.

In China, this culminated in the famous imperial examination system. In Korea, Japan, and Vietnam, related institutions spread through the wider Sinosphere in different forms and at different times (Weber, 1951; Ho, 1962; Woodside, 1971; Elman, 2000, 2013).

The broad historical claim is that a cultural system fundamentally changed the reward structure around human behavior. If educational success becomes a multi-generational route to office, status, income, and marriage prospects, could that institution create a weak but persistent evolutionary selection environment (Ho, 1962; Elman, 2000, 2013)?

The standard Confucian hypothesis is cultural/blank slatist: East Asian educational intensity persists because families and institutions transmit norms of study, discipline, examination ambition, and respect for learning across generations. A lineage that produces successful examinees teaches the next generation how to become successful examinees. Books, tutors, parental pressure, status competition, and family honor do the work. This is a vertical-transmission model, not a genetic one.

The evolutionary version asks whether that cultural system also changed selection pressures. If educational success improved officeholding, income, marriage prospects, and lineage prestige for many generations, then the institution may have rewarded not only learned behavior but also heritable traits correlated with educational success.

This idea has a deep intellectual background. Weber’s classic discussion of Confucian status culture placed the literati near the center of Chinese social organization (Weber, 1951). Ho treated the examination system as a ladder of elite competition and social mobility (Ho, 1962). Elman described the late imperial examination system as a mechanism of cultural and political reproduction (Elman, 2000, 2013), while Woodside emphasized how the Chinese bureaucratic model shaped Vietnam’s state tradition (Woodside, 1971).

But today, modern genetics allows us to test this historical theory in a completely new way. Educational-attainment GWAS now make it possible to construct polygenic scores, and ancient DNA allows those scores to be measured across historical time (Lee et al., 2018; Okbay et al., 2022).

For the first time, we can move beyond sociological debate and actually look for the genetic footprints of the Confucian hypothesis.

Here is a first look at the data. The figure below plots the relationship between ancient genomes and the "event time" of local Confucian establishment.

There is a clear, visible shift in the trajectory. But what exactly are the axes measuring here? Does this post-hinge slope survive when we strip away the noise and rigorously control for deep genetic ancestry?

Read more

Across animals, there is evidence that harsher and more variable environments can favor larger brains. In birds, for example, species exposed to greater environmental variation tend to have larger brains relative to body size, and the proposed mechanism is straightforward: when food, weather, and survival conditions change sharply across the year, behavioral flexibility becomes more valuable (Sayol et al. 2016).

Humans are not birds, but the logic is relevant. Cold environments are unusually demanding. Food is seasonal, winters are dangerous, and survival can depend on planning, storage, mobility, clothing, shelter, technology, and cooperation. A large brain is costly tissue, but in the right ecological setting those costs may be offset by better problem-solving and behavioral flexibility.

Ancient DNA now lets us ask a more precise version of this old question.

Instead of comparing modern countries or modern skull measurements, we can ask whether ancient individuals sampled from colder environments carried higher polygenic scores for brain volume.

What was tested

I used a clumped brain-volume polygenic score and asked whether it varies with paleoclimate at each ancient sample’s location and date. The climate variables come from CHELSA-TraCE21k-style estimates matched to ancient coordinates and dates (Karger et al. 2023). The ancient samples come from the Allen Ancient DNA Resource, or AADR, a curated compendium of ancient human genomes (Mallick et al. 2024).

There are two different hypotheses here.

The first is geographic:

Do ancient individuals from colder places have higher brain-volume PGS?

The second is dynamic:

Within the same location, did cooling over the previous millennium predict higher brain-volume PGS?

Those questions sound similar, but they are not the same.

A cold place can have high brain-volume PGS because of long-term adaptation, migration, ancestry structure, body-size biology, or sampling. A within-location cooling model asks something more specific: when a location becomes colder over time, does the score rise?

Brain volume: cross-sectional climate

The first analysis is the broad geographic test. It asks whether ancient individuals from colder locations have higher brain-volume PGS, after accounting for sample age.

Here the climate variables are coded as climate values themselves. Annual mean temperature is annual mean temperature. Coldest-quarter temperature is coldest-quarter temperature. Warmest-quarter temperature is warmest-quarter temperature.

So the sign is straightforward:

Negative temperature beta = warmer places have lower brain-volume PGS = colder locations have higher brain-volume PGS.

I also include temperature seasonality because the cold-brain hypothesis is not only about annual averages. Winter severity and seasonal amplitude may matter too. But seasonality is not the same thing as cold, so it has to be interpreted separately.

The first figure shows the standardized climate coefficients for brain-volume PGS in AADR. This is the simplest way to see whether the cold-location pattern is actually present.

Read more

The genomic data used in this analysis is sourced from Font-Porterias et al. (2019), which provides high-resolution genome-wide data for European and Iberian Roma groups. This dataset allows us to move past broad global comparisons and instead compare the Roma directly against the populations they historically lived alongside or originated from: southern and southeastern Europeans, South Asians, and West or Central Asians.

Based on anthropological observations and cultural history, we can form clear baseline expectations for these comparisons. Genetically, we would expect to see signals matching their distinct physical profile, specifically shorter stature and darker skin pigmentation relative to host European populations. Behaviorally, their long-standing cultural customs of insularity and historically low rates of formal schooling predict lower polygenic scores for educational attainment.

By testing these expectations against an ancestry-adjusted dataset, we can see exactly where these traits track with neutral genetic drift and where they diverge.

The Roma Effect

To evaluate these trait differences, the comparison uses an individual-level regression model calculated separately for each polygenic score. Running the models both with and without ancestry controls allows us to disentangle the broad effects of geographic ancestry from actual positive selection.

The baseline, unadjusted model measures the raw trait difference between the Roma and the non-Roma comparison panel in standard deviation units. This shows the total observed genetic difference.

The second model adds the first five genetic principal components to control for population structure. By stripping away the background noise of deep geographic origins, this ancestry-adjusted model reveals whether a trait divergence vanishes as a mere byproduct of ancestry, or remains robust, signaling potential directional selection.

The table below examines whether the Roma mean is higher or lower than the pooled non-Roma comparison mean, showing how that contrast changes once those ancestry principal components are introduced.

Read more

On July 1st, this Substack turns one year old.

I launched PifferPilfer as an experiment in writing in public: a place to sort through data on population genetics, ancient DNA, intelligence, fertility, personality, and the stubborn little cultural puzzles that keep pulling me back in.

With the first birthday approaching, I decided to look back through the archive and rank every post by reader likes.

A Quick Birthday Note: As you look through this list, please keep in mind that most of these deep dives are currently locked behind a paywall for our premium community.

Because it is our first anniversary, I want to open the doors a bit wider. If you want full, unrestricted access to the entire archive listed below, plus everything coming in year two, you can claim a limited time discount via the First Birthday Special Offer.

Now, a quick caveat on the numbers: likes are an imperfect metric for quality. Older posts naturally had a smaller subscriber base at launch, paid posts face a smaller potential audience by design, and some pieces are written to be reference tools rather than viral hits. Where posts tied, I broke the tie using restacks, then comments, and finally recency.

Perfect or not, the ranking provides a fascinating record of what you, the readers, chose to signal back.

The 20 Most Liked Posts

1. The Four Races of Europe (Part I): The Aryan paradox — 82 likes Revisits nineteenth century European racial labels through the lens of ancient DNA, arguing that while deep ancestry components still structure modern variation, they do not map neatly onto old romantic categories.

2. The Origins and Spread of Blue Eyes in Europe: Evidence from Ancient DNA — 81 likes Traces exactly where blue eye associated variants first emerged in Europe and tracks how their frequencies shifted across hunter gatherers, early farmers, steppe groups, Romans, and later populations.

3. The Genetic Evolution of the Human Race and Its Consequences for the Industrial Revolution — 68 likes Connects deep time genomic trends to Gregory Clark’s thesis on long run selection, asking whether detectable shifts in cognitive related polygenic scores helped lay the groundwork for the Industrial Revolution.

4. Cognitive Genetics Through Time: Surprises From 3,640 Ancient Genomes — 66 likes The post that launched this Substack. A deep dive tour of educational attainment and cognitive polygenic scores across 3,640 ancient genomes, featuring a few major surprises from ancient Rome and beyond.

5. Debunking the Caricature: What Polygenic Scores Actually Show — 64 likes A long form defense of what polygenic scores can and cannot tell us, aimed at replacing the usual media strawmen with actual, rigorous statistical claims.

6. The Right-Wing Immigration Paradox — 62 likes Uses Meloni’s Italy as a case study to show that immigration politics are not a simple Left versus Right binary. The real friction often lies in motives, selection, and hard state interests.

7. The Rise of Rome and the Rise of Polygenic Scores — 55 likes Asks why Rome, rather than its immediate regional rivals, became the imperial core by comparing ancient Italian population shifts and the rise of polygenic score signals.

8. Genetic evidence for race differences in behaviour — 53 likes Uses behavioral variation among brown bear populations as an analogy for how geographic isolation and domestication like selection pressures can drive behavioral divergence in populations.

9. What Akbari’s Reply Gets Wrong About Science — 44 likes A response to Akbari’s reply, arguing why replication, proper credit, and citation norms matter immensely when new work builds directly on older polygenic selection analyses.

10. Why Northern Europeans Seem Aloof and Southern Europeans Outgoing — 43 likes Turns a familiar cultural stereotype into an empirical inquiry, asking whether North-South European differences in personality and social warmth have measurable, deep rooted biological signatures.

11. Imitation Without Citation — 43 likes A more personal post reflecting on the experience of seeing proprietary ideas reused without attribution, exploring the blurry boundary between ordinary intellectual influence and scientific erasure.

12. There Was Never a Sapient Paradox — 42 likes Challenges the conventional archaeological view that anatomically modern humans sat around cognitively idle for millennia before culture suddenly took off, arguing the paradox relies on flawed assumptions about behavioral modernity.

13. What Harari’s Sapiens Gets Wrong — 39 likes A critique of Yuval Noah Harari’s Sapiens for rendering human history entirely dependent on shared fictions while ignoring biology, population divergence, and genetic change.

14. Intelligence Isn’t Really Sexy — 38 likes Reviews dating app data, behavioral experiments, and genetic evidence to ask if intelligence is actively selected for in mate choice, and why the reality is far more counterintuitive than people assume.

15. Do Parents Ever Share More Genes with Coethnics Than with Hybrid Children? — 33 likes Applies the Wright-Harpending model to a highly controversial kinship question: how a parent’s genetic relatedness to their broader coethnic group compares mathematically to their relatedness to their own mixed ancestry children.

16. Ten Myths About Human Genetics That Refuse to Die — 32 likes A compact, no nonsense myth busting guide tackling common misunderstandings surrounding recent human evolution, ancient DNA, polygenic traits, ancestry, and race.

17. Why Some Provinces Do Better in School (and Math Olympiads): The Answer Is Written in Their Surnames — 30 likes Links Italian surname geography to regional academic outcomes, using deep rooted surnames as a historical tracker for regional ancestry and internal migration.

18. How Germanic is Northern Italy, Really? — 29 likes Untangles language, history, and genetic data to estimate what Germanic actually means in Northern Italy after centuries of Roman era migrations, Goths, Lombards, and later genetic mixing.

19. Was Pale Skin an Elite Trait? — 29 likes Asks whether depigmentation in Europe was purely a functional adaptation to high latitudes, or if it also operated as a status linked trait favored within historical elite structures.

20. Throwing Cold Water on the Cold Winters Hypothesis: A Global Test — 29 likes Puts the classic evolutionary claim that harsher winters selected for higher cognitive capacity to a global empirical test, finding a reality that is vastly more complicated than the popular narrative suggests.

What the Archive Tells Us

The pattern here is pretty clear. The posts that travel the furthest are not just the ones making the sharpest or most provocative claims. They are the ones where a dense, highly technical question is tethered directly to something human beings instinctively care about: ancestry, history, fertility, intelligence, social status, climate, language, and how populations shift across time.

Moving forward, that is exactly the lane I intend to stay in.

The editorial philosophy for year two remains unchanged: data first, but never data as mere decoration. I want to explore genetics without pretending that culture, geography, and history do not matter, and I will always explicitly label speculation as speculation.

To everyone who has read, liked, commented, argued, subscribed, or sent over papers and datasets during this first year: thank you. The publication is still young, but the archive is finally taking a distinct shape.

If you want to unlock these 20 posts (and more) and join us for what is next, don’t miss the First Birthday Special Offer before it expires.PifferPilfer is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Italy offers a different kind of natural experiment.

Not through inbreeding, but through language.

To non-Italians, the Italian rolling R may sound like a colorful accent feature: musical, theatrical, maybe even charming. But to Italian speakers it is not neutral. Standard Italian expects an apical R, produced with the tip of the tongue: usually a tap for single R and a trill for double R. When someone cannot produce that sound, Italians notice immediately.

The usual label is r moscia, literally a “soft” or “weak” R. But the label carries more than phonetics. Depending on the speaker and context, r moscia can sound childish, defective, aristocratic, French, northern, affected, elegant, or ridiculous. It may be corrected in childhood, mocked at school, imitated by others, or recognized as a family trait.

That is what makes Italy useful. The language itself does the first round of phenotyping. Millions of people grow up being expected to produce a difficult tongue-tip R, and those who cannot usually know it. What might be an invisible motor difference elsewhere becomes audible, named, and remembered.

The phenotype already exists in the culture.

The question is whether anyone has tried to study it genetically.

The rolling R is difficult

The Italian R is not just “an R.” It is a specific motor act.

A standard Italian R is produced with the tip of the tongue near the alveolar ridge, just behind the upper teeth. In the simplest case, the tongue makes a quick contact: a tap. In the stronger case, especially with double R, airflow sets the tongue into repeated vibration: a trill.

This is easy to take for granted if you can do it. But motorically it is not trivial. The tongue has to be in the right place, with the right degree of tension, while airflow passes through at the right speed. Too much contact and the sound is blocked. Too little and the vibration does not happen.

This is why rhotics are often among the later sounds children master. In a study of German- and Spanish-speaking children, German uvular /r/ was acquired earlier than Spanish alveolar taps and trills, suggesting that the tongue-tip rhotics required in Spanish, and relevant to Italian, are developmentally more demanding than some back-R variants (Kehoe, 2018).

So the interesting phenotype is not simply “having r moscia.” The more precise phenotype is successful mastery of the Italian apical R.

R moscia is not one thing

In Italy, r moscia is a social object before it is a scientific phenotype.

People know it when they hear it. But they do not always mean the same thing by it. One person may use a uvular R, another a labiodental or approximant-like sound, another a weak or distorted rhotic. Ordinary labels such as r moscia or r francese collapse several different phonetic realities into one social category (Romano, 2013).

That creates a problem, but also an opportunity.

Some cases of r moscia probably reflect real difficulty producing the apical trill. Others may be dialectal. Others may be learned from parents. Others may come from contact with French, German, or other languages where non-apical R-sounds are common. Some people may be able to produce the standard R when asked but not use it naturally. Others may not be able to produce it at all.

Northern Italy is especially relevant here. Non-standard R variants have been reported more often in some northern and border regions, particularly in areas historically exposed to French or Germanic influence. But r moscia is not confined to the north. Individual speakers occur throughout Italy (Romano, 2013; Vietti, 2019).

In the paid section, I look at what the genetics of speech disorders can (and cannot) tell us, why the IQ correlation is probably not what people would expect, and how Italy could be used for the first serious GWAS of rhotic articulation and common speech disorders.

Read more

Human genetics is much messier. We cannot run breeding experiments in people, and many loss-of-function variants are so rare that they usually appear in only one copy. For autosomal genes, a person normally inherits two copies, one from the mother and one from the father. If one copy is inactive, the other may often be enough. The more revealing question is the recessive one: what happens when both copies are lost?

That is why Pakistan is such an unusually informative case. The country has a high rate of consanguineous marriage, especially cousin marriage, which increases autozygosity: the chance that both copies of a genomic segment are inherited from a shared ancestor. This does not recreate Mendel’s pea experiments by design, but it produces something closer than most human datasets. Rare variants that would usually remain hidden in heterozygous form become much more likely to appear in homozygous form.

In a paper just published in Nature, Christopher Koch, Danish Saleheen, and colleagues use this natural structure to build the Pakistan Genome Resource: exome and genome sequencing from 173,303 people recruited across Pakistan.

The authors found naturally occurring homozygous loss-of-function variants in 6,476 genes, roughly one-third of all protein-coding genes. Some of these losses are harmful. Some seem surprisingly well tolerated. Some may even be protective. For medicine, that distinction matters. A person who is naturally missing both working copies of a gene can reveal something that cell culture, mouse models, and association studies often cannot: what long-term loss of that gene looks like in a human body.

Why Pakistan?

Pakistan is useful here for two reasons.

The first is consanguinity. The country has a high rate of consanguineous marriage, especially cousin marriage, and this is exactly the demographic setting in which rare recessive variants become easier to observe. In the Pakistan Genome Resource, 30.6% of participants self-reported first-cousin parents, and the measured runs of homozygosity were consistent with elevated familial relatedness. That matters because it raises autozygosity: the chance that the two copies of a genomic segment are inherited from the same ancestor.

This makes makes Pakistan unusually informative for a specific genetic question. In a more outbred population, many rare loss-of-function variants remain hidden in heterozygous carriers. In a more autozygous cohort, some of those variants appear in homozygous form, allowing researchers to observe the consequences of complete gene loss.

The second reason is under-sampling. South Asians remain underrepresented in global genetic databases, and the study shows how much variation was missing. The authors identified 6.6 million coding variants. About 47% were absent from non-South Asian gnomAD samples, and about 30% were absent even from all of gnomAD, including South Asians.

Human knockouts

The older phrase for these individuals is “human knockouts,” but the term can mislead if taken too literally. A laboratory knockout is engineered. A person with a homozygous loss-of-function variant is not engineered; they are a naturally occurring carrier of two inactive copies of a gene. Still, the analogy is useful because drug developers are often trying to do, partially and reversibly, what genetics sometimes does completely and lifelong.

PCSK9 is the classic example. Human genetics showed that reduced PCSK9 function lowers LDL cholesterol and protects against cardiovascular disease, helping make PCSK9 inhibition an attractive therapeutic strategy. The broader logic is now central to drug development: if people who naturally lack a gene are healthy, or healthier with respect to a disease endpoint, then inhibiting that gene becomes more attractive. If they are sick, the target may be dangerous.

The Pakistan Genome Resource makes this kind of evidence much more abundant. The authors report homozygous predicted loss-of-function variants in 6,476 genes, and more than 34,000 participants carried a homozygous loss-of-function variant in at least one gene. In gnomAD, the rate of discovering such genotypes is much lower for most ancestry groups. The Amish sample is a useful comparison because it is also enriched for relatedness, but it is tiny compared with PGR. Pakistan offers both scale and autozygosity.

The result is not just a long list of rare variants. It is a way to sort genes into rough functional categories: genes where complete loss is compatible with adult life, genes where complete loss causes an obvious phenotype, genes where complete loss may be protective, and genes where homozygous loss is depleted, suggesting strong selection or essential function.

That last category is easy to overlook, but it may be one of the most important. If a large autozygous cohort rarely or never produces homozygous loss-of-function carriers for a gene, while comparable neutral variants do appear, then complete loss of that gene may be incompatible with development, survival, fertility, or recruitment into an adult cohort.

The genome’s no-fly zones

The paper uses homozygous loss-of-function variants to ask which biological systems tolerate complete disruption. The answer is reassuringly coherent. Genes essential in cell screens, genes whose mouse knockouts are embryonic lethal, genes linked to ClinGen disease annotations, broadly expressed genes, and targets of approved drugs were all depleted for homozygous loss-of-function. Developmental and core cellular pathways, including TGF-beta signaling, Notch, Hedgehog, oxidative phosphorylation, DNA repair, protein secretion, and the unfolded protein response, also showed depletion.

This is population genetics behaving like a functional screen. If a gene can be knocked out without much consequence, living adults with two inactive copies may appear in a large enough autozygous cohort. But if a gene is necessary for development, survival, fertility, or recruitment into the study, those homozygotes will be missing or strongly depleted.

The diagram summarizes the logic, but the interpretation depends on the type of gene. For essential or early-development genes, complete loss may prevent a viable adult from ever existing. In that case, the strongest signal is not an affected adult in the dataset but the absence of homozygous loss-of-function carriers altogether. For recessive disease genes, the pattern may be less extreme. Homozygotes may survive, but disease, reduced survival, impaired fertility, or lower participation in an adult cohort can reduce their representation. For tolerated genes, complete loss has little effect in ordinary adult physiology, or biological redundancy compensates for the missing gene, so homozygotes appear at rates closer to frequency-matched neutral variants.

Drug targets are not chosen only for efficacy. They must also be safe. A genetic loss-of-function phenotype is not identical to a drug effect: a drug may be partial, reversible, tissue-specific, dose-dependent, or started late in life, while inherited gene loss is lifelong and present from conception. But human loss-of-function genetics is still one of the strongest clues we have about long-term target biology. If complete loss of a gene is well tolerated in living humans, that is useful information. If complete loss is absent from a cohort where it should otherwise be visible, that is useful information too.

The general logic is simple enough: Pakistan’s high rate of consanguineous marriage makes rare recessive gene loss visible in a way most human datasets cannot.

But the most interesting part of the paper is what appears once you look gene by gene and trait by trait. The rest of the post looks at the specific findings: which loss-of-function variants point to safer drug targets, which ones raise warning flags, why HBB shows an unexpected cholesterol signal, why LRRK2 matters for Parkinson’s drug development, and what the paper finds when it links inbreeding itself to fertility, education, body size, cardiometabolic traits, tuberculosis, and kidney disease.

There is also a population-genetic twist: Pakistan is not a single genetic block, and one group in the dataset shows a surprising trace of recent African ancestry.

Read more

While we traditionally trace this story through texts, coins, and pottery, ancient DNA (aDNA) introduces a different kind of evidence. It answers a fundamental question written records often obscure: Who was actually there?

Rome was more than a state. It was a demographic process. Conquest, slavery, trade, and migration constantly altered the genetic composition of the people living in Italy.

This post is a follow-up to a paper I published in 2023, where I analyzed Educational Attainment polygenic scores in ancient Rome. That paper was based on a single ancient DNA dataset from modern-day Lazio, the region of which Rome is the capital, so the question was necessarily narrow: what happened to EA PGS in and around the Roman capital across time?

Here I expand the analysis to Italy as a whole. Instead of relying on one dataset from one province, I combine ancient DNA samples from different studies, covering multiple regions of the peninsula and Sicily. This makes it possible to ask a broader question: was the genetic trajectory observed in ancient Rome just a local pattern, or part of a larger Italian cycle linked to the rise, transformation, and collapse of Roman civilization?

The Core Question: Linear Growth vs. Civilizational Cycles

Before getting to Rome, it is worth clarifying what is being measured. A polygenic score, or PGS, is a summary of many genetic variants that are each weakly associated with a trait in genome-wide association studies. No single variant matters very much. The score only becomes informative when thousands of small effects are added together.

Here I use the Educational Attainment polygenic score, or EA PGS, not as a literal measure of schooling in ancient people. Ancient Romans obviously did not take modern exams or attend modern universities. Rather, EA PGS is used as a broad index of genetic human capital: a noisy but useful proxy for the inherited component of cognitive potential, personality traits related to persistence and planning, and other behavioral tendencies that affect learning, achievement, and social complexity.

With that in mind, ancient Italian EA PGS can be used to test two competing hypotheses about how populations changed over time:

• The Linear Trend (Long-Run Selection): Ancient Italy was merely a local example of a broader, steady upward trend in EA PGS seen across Western Eurasia over millennia. In this scenario, Rome requires no special explanation. It’s just a point on a long, continuous slope.

• The Civilizational Cycle: The genetic timeline follows Rome’s historical arc: rising during the Republic, falling during the Imperial/Late Antique influx, and partially rebounding after the collapse of the Western Empire in 476 CE.

If the data reveals a cycle, it raises a fascinating possibility: that the rise and fall of civilizations is tied to fluctuations in the genetic human capital of their populations.

What the Data Shows

The results do not show a simple, flat national average. Instead, the genetic timeline changes by region, perfectly mirroring the broad historical sequence:

The fact that we see a regionally moderated cycle, starting from the pre-Iron Age baseline, peaking in the Republic, dipping in the Empire, and rebounding after, suggests that civilizational dynamics and genetic human capital are deeply intertwined.

The rest of the post shows how that pattern appears in the data, why the north-central Republican peak matters, and what happens when ancestry controls are added.

Read more

But there is another intuition, familiar from the internet’s favorite civilizational cycle: hard times create stronger, more organized people; good times eventually soften them. The meme is crude, but the underlying question is not stupid. Do pressure and adversity sometimes force societies to become more complex?

Ancient history does not always look like a record of progress under comfort. Major transitions in subsistence, technology, settlement, and social organization often occurred under stress: changing seasons, unstable ecologies, shifting food systems, and population movements that forced human groups to adapt. The question, then, is not whether cold weather was pleasant. It is whether long-term climate change, measured over timescales meaningful for human populations, is associated with changes in archaeological stage.

To test this, I linked ancient DNA samples to reconstructed temperature histories and asked a simple question: when a location became warmer or colder than it had been 1,000 years earlier, did the samples from that location tend to belong to a later archaeological stage?

This is not a study of modern global warming. The data come from deep history, long before industrial emissions, fossil-fuel economies, or modern climate politics. The temperature changes examined here are natural, slow-moving fluctuations across thousands of years of human prehistory.

The point is not to draw a political lesson about the present, but to ask a historical question: did long-run ecological change leave a detectable trace in the way ancient societies were organized?

The answer, in the current data, appears to be yes.

What Exactly Are We Measuring?

I am not measuring IQ, wealth, literacy, or state capacity here. The outcome is much simpler: the archaeological stage assigned to each ancient genome, such as Mesolithic, Neolithic, Bronze Age, Iron Age, and related period categories.

That is obviously cruder than a full database of institutions or technologies. But it has one major advantage: it stays close to the metadata attached to the actual ancient DNA samples, rather than importing a separate historical classification system.

The climate variable is also straightforward. For each sample, I measured how much the local temperature had changed over the previous millennium:

delta1000 = current temperature − temperature 1,000 years earlier

A positive value means the location had warmed over the previous thousand years. A negative value means it had cooled.

The models then ask whether this temperature change predicts archaeological stage after accounting for the obvious confounders: current temperature, time, sample coverage, ancestry principal components, and educational-attainment (EA) polygenic scores.

The EA PGS control is especially useful because it addresses a possible objection directly. If climate change predicts archaeological stage, is that really a climate signal, or is it just picking up genetic differences correlated with educational attainment? Adding EA PGS lets us test whether the climate-stage association survives after that source of variation is included.

The Main Result: Linear Temperature Change

To get the cleanest possible result, I didn’t just rely on default database labels. I used a manually verified subset of the data, meaning rows where the archaeological period assignment was rigorously cross-checked and verified against the original published supplementary materials to filter out noise.

It’s also vital to clarify the mechanics of the model: this does not estimate a separate, isolated cooling slope. Rather, it evaluates continuous temperature change across the board.

To make the direction intuitive and readable, Table 1 reports the coefficient as the implied effect of 1°C of cooling. Because the underlying model was fit as current - lag1000, the reported cooling effect is simply the sign-flipped coefficient of that linear model.

Table 1. Linear Temperature Change, Displayed as the Implied Cooling Effect

In this linear temperature-change model, the negative-delta direction implies that cooling over the previous 1,000 years predicts a higher ordered civilization stage.

Is this effect driven by genetics? How large is this shift in reality? And does this mean cold weather mechanically caused civilization?

Read more

In one world, it can mean a chain of neighboring communities, regular marriage contacts, shared routes, seasonal movement, trade, and overlapping political horizons. In another, it can mean a hard break: different worlds separated by mountains, forests, water, language, subsistence, and time.

Ancient DNA gives us a way to ask whether that intuition leaves a genetic trace. At the same physical distance, were older ancient populations more genetically differentiated than more recent ones?

After controlling for geographic distance and broad region-pair, older group pairs have higher Hudson FST. Put plainly: the same number of kilometers corresponded to more genetic distance in older periods.

First, distance matters today

Before asking whether distance mattered more in the past, we need a baseline.

Among present-day populations in the AADR dataset, populations that live farther apart tend to be more genetically different. This is hardly surprising, but it is important to establish before moving on to the ancient data.

To show this, I matched the modern AADR Hudson FST table to the geographic coordinates of each population and calculated the distance between population centers. I restricted the analysis to groups with at least five individuals and known coordinates.

The relationship is far from perfect, but it is clear: greater geographic distance generally corresponds to greater genetic distance.

Figure 1. Modern FST by geographic distance

The real question is whether a given distance meant the same thing throughout history. Did 1,000 kilometers separate populations more strongly 8,000 years ago than it did 2,000 years ago? Ancient DNA allows us to test exactly that.

The continuous trend

Even after accounting for how far apart two populations were and which broad regions they belonged to, older population pairs were consistently more genetically different from one another.

The effect is large enough to matter. For every additional 1,000 years into the past, Hudson FST increases by about 0.01.

In practical terms, this means that a journey of 1,000 kilometers connected people much less effectively in the distant past than it does in more recent periods. Populations separated by the same physical distance were genetically more distinct, suggesting that migration, trade, communication, and marriage networks became progressively better at bridging geographic space over time.

Table 1. Core model coefficients

One way to read the result is to fix the distance. At 1,000 km, the predicted FST rises steadily as pair age gets older.

Table 2. Predicted FST at 1,000 km by mean pair age

Because this is a linear model, the smooth spacing between rows is the fitted time slope. It should be read as a compact model summary, not as a claim that genetic distance changed mechanically by the same amount every millennium.

Figure 1. Predicted FST by age at fixed geographic distances

Figure 2 isolates the 1,000 km case. This is the most direct visualization of the question: hold distance fixed and move through time.

In this model, 1,000 km around 1,000 BP corresponds to predicted FST of about 0.064. The same 1,000 km around 9,000 BP corresponds to about 0.142. The exact numbers depend on the model and the AADR groups that survive the sample-size filter, but the direction is robust.

Subscribe now

Some concrete same-country-sized examples

The abstract version of the result is easy to miss. A more intuitive way to read it is to ask what distances inside modern countries would have implied in the Neolithic.

Modern countries did not exist in the Neolithic, of course. London and Edinburgh were not endpoints in a British population system, and Rome and Milan were not Italian cities. The point is only to give the distance scale a familiar shape. These are present-day city pairs separated by roughly 500-660 km.

Table 3. Model-predicted FST for same-country-sized European distances

Those values are large compared with modern within-country or near-within-country population proxies in the same AADR-derived modern FST benchmark.

Table 4. Modern within-country or regional proxy FST values

A 500 km separation inside present-day Europe often corresponds to tiny modern FST values, on the order of a few thousandths in these proxies. In the ancient model, a similar geographic separation around 6000-8000 BP corresponds to predicted FST around 0.11-0.13, similar to the distance between East Asians and Europeans. That is not a small difference in scale.

What this means historically

Older ancient populations were more genetically patchy over space. The same physical distance separated more differentiated populations in the deeper past than in later periods.

That does not require a single mechanism. Several processes can produce the pattern. Small effective population sizes can make local drift stronger. Sparse contact networks can allow neighboring regions to diverge. Later expansions and mixtures can smooth older discontinuities. Farming, pastoralism, metallurgy, trade, ships, horses, roads, empires, and marriage networks can all make distance less absolute, but FST does not tell us which one is responsible in any particular case.

Methods

I use FST as the genetic-distance measure. An FST of zero would mean two groups are genetically indistinguishable at the measured markers; higher values mean greater allele-frequency differentiation. Geographic distance is the great-circle distance between group centroids.

The unit of analysis is AADR Group ID pairs, not individual pairs. I first KING-pruned individuals, retained ancient groups with at least 25 post-pruning individuals, computed Hudson FST between retained groups, and then modeled group-pair FST as a function of geography and time.

The retained dataset contains 141 ancient groups, 7,494 individuals, and 9,870 group-pair comparisons. For the region-controlled continuous model, pairs with unclassified broad-region labels were dropped, leaving 9,045 group pairs.

The core model is simple:

Hudson FST ~ geographic distance + mean pair age + broad region-pair

Mean pair age is the mean of the two groups’ median Years BP. Broad region-pair is a dummy for combinations such as Europe-Europe, Europe-West Asia, West Asia-West Asia, and so on.

References

Bhatia, G., Patterson, N., Sankararaman, S. & Price, A.L. (2013). Estimating and interpreting FST: The impact of rare variants. *Genome Research*, 23(9), 1514-1521. https://doi.org/10.1101/gr.154831.113

Chang, C.C., Chow, C.C., Tellier, L.C.A.M., Vattikuti, S., Purcell, S.M. & Lee, J.J. (2015). Second-generation PLINK: rising to the challenge of larger and richer datasets. *GigaScience*, 4, 7. https://doi.org/10.1186/s13742-015-0047-8

PifferPilfer is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Mallick, S., Micco, A., Mah, M., Ringbauer, H., Lazaridis, I., Olalde, I., Patterson, N. & Reich, D. (2024). The Allen Ancient DNA Resource (AADR) a curated compendium of ancient human genomes. *Scientific Data*, 11, 182. https://doi.org/10.1038/s41597-024-03031-7

Manichaikul, A., Mychaleckyj, J.C., Rich, S.S., Daly, K., Sale, M. & Chen, W.-M. (2010). Robust relationship inference in genome-wide association studies. *Bioinformatics*, 26(22), 2867-2873. https://doi.org/10.1093/bioinformatics/btq559

Educational attainment (EA) was the next obvious trait to examine.

Ancient people did not attend universities, take standardized tests, or spend years in formal schooling. Yet the modern educational attainment polygenic score captures much more than education itself. It reflects thousands of genetic variants associated with cognition, persistence, long-term planning, and the ability to navigate complex social systems.

That raises an intriguing question:

When a region became colder than it had been a millennium earlier, did the genetic profile associated with educational attainment also change?

The answer depends on how the question is asked.

Using my primary climate-history model, the effect is essentially zero. But when cooling and warming are allowed to operate differently, a more interesting pattern appears: cooling predicts higher EA polygenic scores, while warming shows little evidence of producing the reverse effect.

Unlike height, where the signal is straightforward, EA turns out to be a more complicated and perhaps more revealing case.

I will not repeat the full methodological details here because they are identical to those used in the previous height and personality analyses. This post uses the same AADR sample, climate reconstructions, Mundlak within/between-location framework, ancestry controls, and robustness procedures described earlier.

The only thing that changes is the outcome variable: educational attainment polygenic score (EA PGS).

The first question is simple: if a location became colder or warmer than it had been a millennium earlier, did its EA polygenic scores change as well?

To test that, I added 1,000-year temperature change to the baseline model and asked whether it improved prediction.

Table 1. Linear 1,000-Year Temperature Change and EA PGS

The result is straightforward: there is no detectable linear relationship.

Adding 1,000-year temperature change does not improve prediction of EA polygenic scores in either the pooled or Mundlak specifications.

If this were the only model examined, the conclusion would be simple:

There is no evidence that millennium-scale climate change influenced ancient EA polygenic scores.

Goodbye cold-winters theory!

The Split-Slope Diagnostic Changes the Picture

The problem with the linear model is that it assumes cooling and warming are mirror images of one another.

Nature does not necessarily behave that way.

To test this possibility, I replaced the single temperature-change term with separate cooling and warming variables:

• Cooling = max(−Δ1000, 0)

• Warming = max(Δ1000, 0)

This allows cooling and warming to have independent effects.

Read more