The Corded Ware: The Westward Dilution, and the Four Women Who Carried None of It

Between about 2900 and 2300 BC a single burial rite, a single pot shape and a single stone axe spread from the Rhine to the upper Volga, and the people buried under it turn out to have been, in the plainest genetic sense, a new population. The Corded Ware is the largest population turnover in European prehistory and the one that put roughly half of the modern European genome in place. This article measures two things about it that are usually asserted rather than counted: how much the incoming ancestry was diluted as the culture moved west and north, and what happened to the people who were already there. The first has a clean answer, from 96 per cent in the earliest eastern Baltic graves down to 59 per cent in Switzerland. The second has an answer too, and it is carried by four burials in Bohemia rather than by any summary statistic.

A culture with no country

The Corded Ware is defined by its rubbish. Cord-impressed beakers, stone battle axes, a body laid on its side with the knees drawn up, men on their right side facing south and women on their left facing north, under a low round barrow, one person to a grave. That package appears almost simultaneously across an area of something like two million square kilometres, from the Netherlands and Switzerland in the west to Finland and the Moscow region in the east, and it replaces a patchwork of long-established farming cultures that had been building collective tombs and living in villages for two thousand years.

Archaeologists argued about what that meant for a century. The polite reading was diffusion: a fashion in pots and a change in ideology, spreading between neighbours who stayed where they were. The alternative, associated with Gustaf Kossinna and later with Marija Gimbutas, was migration, and it had acquired a bad reputation for good historical reasons by the middle of the twentieth century.

In 2015 two papers settled the demographic half of the question. Wolfgang Haak and colleagues found that four Corded Ware individuals from Germany could trace about three quarters of their ancestry to a population resembling the Yamnaya of the Pontic and Caspian steppe, and Morten Allentoft and colleagues reported the same pattern across a wider set of Bronze Age genomes. That result has survived a decade of much larger datasets. What has changed since is everything about the detail: where the incoming population actually came from, how quickly the proportion dropped as the culture spread, and who exactly was absorbed into it along the way.

What the uniparental markers already said

The Y chromosome data is the reason anybody talks about sex bias here, and it is worth setting out precisely, because it is both stronger and stranger than the summary version.

Before the Corded Ware, the male lineages of central European farmers were dominated by haplogroup G2a, with I2 and a scatter of others. After it, they are dominated by R1a and R1b. The turnover on the paternal side is close to total, and it is much sharper than the turnover in the rest of the genome.

The strangeness is in the detail. Luka Papac and colleagues published 271 genomes from Bohemia in 2021, covering the whole sequence from the Neolithic to the Early Bronze Age at a resolution nobody had managed before, and their Y chromosome results do not fit the simple story. The Yamnaya males sequenced so far are overwhelmingly R1b-Z2103. The earliest Corded Ware males in Bohemia are not: six of eleven carry R1b-L151, a lineage not yet found in Yamnaya, and three carry R1a-M417. Within five centuries the picture changes again, and ten of eleven late Corded Ware males in the same region carry R1a-M417. Then the Bell Beaker expansion brings R1b-P312, derived from that same L151 pool, to complete fixation, and the transition to the Early Bronze Age knocks it back down to twenty per cent, which implies at least an eighty per cent influx of new male lines in a couple of centuries.

So the paternal record shows three violent reorganisations in a thousand years, and it does not permit a direct father-to-son line from the sampled Yamnaya to the sampled Corded Ware. Papac and colleagues were explicit about that, and it is the reason the origin of the Corded Ware is still an open question even though its ancestry proportions are not.

The maternal record does nothing of the kind. Corded Ware mitochondrial pools are diverse and they look substantially local: the standard central European Neolithic set of H, K, J, T and U5, with the addition of a minority of steppe-associated lineages such as U4, U2e and W. Anna Linderholm and colleagues found five different mitochondrial haplogroups among eleven assigned individuals in south-eastern Poland, in a series where eleven of sixteen males carried haplogroup R. Lehti Saag and colleagues found the same asymmetry in Estonia, where the Corded Ware male lineages are uniform and the female ones map inside the modern Estonian range.

One set of chromosomes was replaced almost completely. The other was not. That is the observation the rest of this article is built around, and the awkward part is that the autosomal data, which is what a Global25 calculator sees, cannot measure it directly at all.

Data and method

Everything below uses Global25 scaled coordinates, with population averages merged from Davidski's published files and the Moriopoulos 2026 collection. Ancestry proportions come from non-negative least squares with a sum-to-one constraint imposed by appending a heavily weighted row of ones, with the resulting weights normalised. Distances are scaled Euclidean distances multiplied by one thousand, the convention used throughout this site.

The pooled Corded Ware target is a weighted average of ten regional groups totalling 86 individuals: Bohemia early and late, Germany, Poland, Estonia, Lithuania, Latvia, Switzerland, the Battle Axe culture of Sweden and the Single Grave culture of Denmark. The Fatyanovo groups of the upper Volga are kept out of that pool and reported separately, because they are the eastern branch and their inclusion would blur the geographic question this article is asking.

Alongside the group averages, 120 individually dated Corded Ware genomes were pulled from the ancient sample file, of which 47 from Bohemia pass a fifteen per cent coverage threshold. Those individuals do most of the interesting work later on, because a group average cannot show you a distribution and this particular distribution has a tail that matters.

One panel does the work in every decomposition below, so that rows are directly comparable: the Yamnaya of Samara, the Anatolian Neolithic of Barcin, and the western hunter-gatherer genome from Loschbour. The condition number of that source matrix is 2.44 and the three pole separations are 338, 410 and 494 units. Readers of our article on the Quebecois, where the French regional panel had a condition number of 145 and pole separations under ten units, will appreciate the difference. This is about as well conditioned as a three-source model ever gets, and it is the reason the numbers below barely move when the sources are swapped.

A word on what the steppe pole measures. It is chosen for its position, not its ethnonym. Yamnaya of Samara and Afanasievo of the Altai sit 7.9 units apart, which is to say they are the same population for these purposes, and the weight this pole takes should be read as ancestry from the eastern European steppe pastoralist cluster of the third millennium BC rather than from any particular kurgan.

Where the Corded Ware sits

The simplest measurement sets up everything else.

The nearest populations to the pooled Corded Ware average are other Corded Ware populations, which is unremarkable, and then a set of things that are not: the Fatyanovo of Yaroslavl at 16.3 units, Sintashta of the southern Urals at 17.3, and the Bell Beaker of Bohemia at 26.4. The Yamnaya of Samara, the source of nearly three quarters of the ancestry, sit at 82.2 units. The Globular Amphora farmers of Poland, the source of most of the rest, sit at 203.6.

Two things on that chart deserve attention before anything is made of the rest.

The first is the position of Sintashta. A group of people buried in fortified settlements in the southern Urals around 2000 BC, who built the earliest known chariots and whose descendants carried Indo-Iranian languages into Central Asia and India, sit closer to the pooled Corded Ware average than the Corded Ware of Switzerland does. Sintashta is 20.2 units from the late Corded Ware of Bohemia and 17.6 from the Fatyanovo. The eastern trajectory of this culture is not a metaphor.

The second is the internal spread. Bohemia early to Bohemia late is 12.5 units. Estonia to Switzerland is 42.3. The Corded Ware is not one population in the sense that the Yamnaya of Samara are one population; it is a cline, and the shape of that cline is the subject of a later section.

The modern rows are worth reading against the ancient ones. Swedes sit 60.6 units from the Corded Ware average, Norwegians 62.8, Poles 75.8, Czechs 75.1, French 101.8. Every one of those populations is closer to a group of people buried under barrows four and a half thousand years ago than to the Anatolian farmers at 268.6 whose descendants those Corded Ware people had largely replaced.

The deep decomposition

Breaking the same populations into ancient components gives the shape of the gene pool.

The pooled Corded Ware average comes out at 71.7 per cent steppe, 19.9 per cent Anatolian Neolithic and 8.3 per cent western hunter-gatherer, with a fit of 20.5 units. That is the figure that has been quoted since 2015, reproduced here on an independent panel: roughly three quarters of the genome from the steppe, roughly a quarter from the farming populations of Europe.

The rows below it show what the panel can resolve. The Funnel Beaker farmers of Bohemia take zero steppe. The Globular Amphora farmers of Poland take 1.7 per cent. Neither of them is anywhere near the Corded Ware, and the transition between them is not gradual: the early Corded Ware of Bohemia sits 232 units from the Funnel Beaker population that preceded it in the same landscape, and 215 units from the Globular Amphora people who were its immediate neighbours and, on the evidence below, its in-laws.

The rows above it are the interesting ones. Two individuals buried in Latvia around 2900 BC, at the very beginning of the eastern Baltic Corded Ware, come out at 96.1 per cent steppe. They sit 41 units from the Yamnaya of Samara. Whatever else was happening at the western end of this culture, at the eastern end the first generation of arrivals had not yet mixed with anybody.

Three quarters and one quarter

The three-source panel is deliberately deep, and deep panels are good at proportions and bad at identifying which particular population supplied them. A proximal model can do better, and there is a specific published claim to test against.

In 2024 Harald Ringbauer and colleagues used long shared haplotype segments, which is a far more direct instrument than anything a coordinate calculator can offer, and reported two things. Corded Ware and Yamnaya groups share long segments of identity by descent, meaning their common ancestry is a matter of a few hundred years rather than a few thousand. And the non-steppe quarter of Corded Ware ancestry can be pinned specifically to the Globular Amphora culture of eastern Europe rather than to farmers in general.

Model the pooled Corded Ware average as a two-source mixture of the Yamnaya of Samara and the Globular Amphora of Poland, and it returns 71.5 per cent and 28.5 per cent, with a fit of 14.4 units. That is a better fit than the three-source deep model achieves, from a two-source model whose poles are 284 units apart and whose condition number is 1.42. Bohemia early returns 74.4 and 25.6, Bohemia late 71.7 and 28.3, Estonia 72.3 and 27.7.

Three quarters and one quarter, arrived at from coordinates, matching a figure arrived at from haplotype sharing. This is the one place in this article where a Global25 model reproduces a much more powerful method closely enough to be worth pointing at.

The dilution west and north

Now the geography.

The earliest eastern Baltic burials return 96.1 per cent. The core central European belt, from Lithuania through Germany and Poland to Bohemia, returns between 72 and 78, and it does so with remarkable consistency across four countries and four separate research programmes. The northern and western margins return between 59 and 62: Sweden at 62.0, Denmark at 60.8, Switzerland at 59.1.

That is a drop of thirty-seven points from the earliest arrivals to the western edge, and it is not a chronological effect. Bohemia early is 74.5 and Bohemia late is 72.2, a difference of two points across five hundred years in the same region. The dilution happens in space, not in time. A population that entered the eastern Baltic essentially unmixed had, by the time its descendants reached the Rhone and the Danish islands, taken up something like forty per cent of its ancestry from the farming populations it passed through.

The individual-level data confirms the group averages rather than complicating them. Bohemia returns a mean of 74.8 per cent across 43 mainstream individuals, with a standard error of 1.6. Germany returns 74.1 across 8, Poland 74.0 across 13, Estonia 71.1 across 5. Four regional research programmes, four sets of laboratory protocols, one number.

Two honest caveats. The Danish Single Grave average rests on a single individual, and the Latvian figure on two, so neither end of the range is as firm as the middle. And the Swedish Battle Axe average excludes the Gotland Pitted Ware burials, which are sometimes filed under the same heading in older datasets and which are a different population entirely, sitting between 29 and 39 per cent steppe. Anyone who pools them will get a Scandinavian figure in the forties and conclude that the Battle Axe culture was half local, which it was not.

Four burials at Vlineves and Stadice

Group averages hide people. This is the section where they stop hiding them.

Forty-seven Bohemian Corded Ware individuals pass the coverage threshold. Forty-three of them form a single tight distribution: mean 74.8 per cent steppe, standard deviation 10.6, running from 40.8 to 97.4 per cent. Four do not. They return 15.5, 11.8, 2.6 and 0.0 per cent.

Those four are not a tail of the distribution. STD003, from Stadice, sits 219.6 units from the mean of the community she was buried among. VLI008, from Vlineves, sits 211.1 units from it. Their distances to the Globular Amphora farmers of Poland are 37.3 and 25.9 units. Pooled and pushed through the same three-source model, the pair returns 0.5 per cent steppe, 74.6 per cent Anatolian Neolithic and 24.9 per cent western hunter-gatherer, against 1.7, 74.6 and 23.7 for the Globular Amphora average itself. To the resolution of this method they are Globular Amphora farmers.

They were buried with Corded Ware rites, in Corded Ware cemeteries, among Corded Ware people, and they carried none of the ancestry that defines the culture.

Papac and colleagues, who excavated the genomic side of this, reported four such individuals among the early Corded Ware of Bohemia, and reported one further fact about them. All four were female. They wrote that observing only females among early Corded Ware individuals without steppe ancestry suggests that the process of assimilating the earlier population into Corded Ware society was female biased, and they noted that STD003 and VLI008 plot in the same part of the principal components space as Globular Amphora individuals from Bohemia and Poland. Our coordinates say the same thing in different units.

One more individual is worth a sentence. VLI092, buried at Vlineves around 2769 BC, returns 40.8 per cent, roughly midway between the community mean and the incoming women. That is where a person with one mainstream parent and one local parent would land. It is a single genome and the inference is loose, but the whole process the previous paragraph describes has to produce people like that, and here is one.

What the autosomes cannot see

Here the article has to stop and be honest about its instrument, because the temptation at this point is to convert a striking observation into a quantity, and the quantity is not available.

Autosomes are inherited from both parents in equal measure and they carry no label saying which side they came from. A Global25 coordinate is an autosomal summary. It can tell you that a Bohemian Corded Ware individual is roughly three quarters steppe and one quarter local farmer. It cannot tell you whether the local quarter arrived through grandmothers or grandfathers, and no amount of source selection will make it tell you.

What can, in principle, is the X chromosome, which spends two thirds of its time in females and therefore over-represents maternal ancestry. In 2017 Amy Goldberg and colleagues did exactly that comparison for Bronze Age central Europeans and found 36.6 per cent steppe ancestry on the X against 61.8 per cent on the autosomes, which under their model implies something between five and fourteen migrating males for every migrating female.

That estimate was immediately contested. Iosif Lazaridis and David Reich repeated the analysis with a different method and found no deficit of steppe ancestry on the X at all, and showed by simulation that the clustering approach used in the original paper carries an estimation bias of about nineteen and a half percentage points at the relevant ancestry level, which would account for most of the gap. Goldberg and colleagues replied that their inference held across several independent procedures. The exchange has not been resolved, and the honest summary is that the direction of the sex bias is not in doubt while its magnitude very much is.

What this dataset can contribute is indirect, and it comes from variance rather than from means. Split the Bohemian individuals at 2600 BC. The twenty earlier individuals have a standard deviation of 14.6 percentage points in their steppe ancestry. The twenty-three later ones have a standard deviation of 5.4. The means are identical, at 76.0 and 73.8, and a test of the means returns nothing at all. The difference in spread is a factor of more than seven in variance and is significant at any threshold you like.

An early population that is still taking people in from outside is heterogeneous. A closed population that has stopped is homogeneous. The Bohemian Corded Ware went from one state to the other in about three hundred years, and it did so over exactly the period in which its Y chromosome diversity collapsed from five lineages to one.

The two series on that chart move independently. Across the Corded Ware period the autosomal profile shifts by two points, from 74.5 to 72.2, while the commonest paternal lineage goes from 55 per cent to 91 and changes identity. Across the Bell Beaker period the autosomal profile falls by more than twenty points while the paternal pool reaches complete fixation. At the Early Bronze Age transition the paternal pool collapses again, from 100 per cent to 20, while the autosomal profile moves back up by ten points.

Whatever was governing the male lineages was not the same thing that was governing the rest of the genome. A small number of related men can take over a paternal pool without changing the autosomal composition of the population very much, because their wives and their wives' mothers keep contributing everything else. That is the mechanism, and it is visible here as two lines that refuse to track each other.

Does the choice of steppe source matter?

This is the objection that has to be answered with numbers, because the steppe pole is the load-bearing part of the argument.

It does not matter much, and where it does the direction is predictable. Yamnaya from Samara gives Bohemia early 74.5 per cent. Yamnaya from Krasnodar gives 77.4, from Hungary 77.3, from Bulgaria 80.6. Afanasievo of the Altai, four thousand kilometres away and belonging to a different archaeological culture, gives 73.3. The Catacomb culture gives 75.7. The ranking is identical under every one of them: Bohemia above Estonia, Estonia above Sweden, Sweden above Switzerland, and the Globular Amphora control never above 2.0 per cent.

Two rows are deliberate stress tests. Khvalynsk, the Eneolithic predecessor on the middle Volga, gives 66.7, and eastern hunter-gatherers give 60.6, because a source further from the target has to be admitted in smaller quantities to explain the same displacement. Their fits are correspondingly worse, 43 and 78 units against 20 for Yamnaya, which is the model saying plainly that it does not like them.

The last row is the one to be careful with. Sintashta gives 100.0 per cent, with a fit of 22.6 units that looks perfectly respectable. It is not respectable at all, because Sintashta is itself descended from the Corded Ware. Using it as a source for the Corded Ware is like modelling a father as a mixture of his son and a stranger: the arithmetic works, the fit is fine, and the answer is meaningless. Circular sources do not announce themselves in the fit statistic, and this is the commonest way that a well-formed Global25 model produces a result that cannot be true.

The trap that catches most home modelling

There is a second and simpler way to get this wrong, and it is worth showing because it is one keystroke away from the correct model.

Drop the hunter-gatherer source. Offer only the Yamnaya and the Anatolian Neolithic, which after all are the two populations everybody talks about, and ask for the Corded Ware. Bohemia early comes back at 81.0 per cent steppe instead of 74.5. Switzerland comes back at 69.1 instead of 59.1. Sweden comes back at 70.1 instead of 62.0.

The inflation is systematic and it is largest exactly where the local farmer contribution is largest, because European Neolithic farmers were never pure Anatolian: they carried between fifteen and twenty-five per cent hunter-gatherer ancestry by the fourth millennium. Remove that source from the panel and the model has to find the hunter-gatherer signal somewhere, and the only pole left that leans in that direction is the steppe.

The diagnostic is in the fit distances. Bohemia early goes from a fit of 19.7 units with three sources to 37.2 with two. The Globular Amphora control goes from 32.1 to 101.4, which is the model screaming. Any two-source model of a Corded Ware population that reports more than eighty per cent steppe is reporting the missing third source.

What is left of it today

The last question is what survived, and the answer is more of it than anybody expected before 2015.

Present-day Europeans run from 57.5 per cent steppe ancestry in Estonia down to 5.6 in Sardinia on the same panel, with a northeast to southwest gradient that has been reported many times and is reproduced here for scale rather than for novelty. Estonians 57.5, Latvians 57.3, Lithuanians 54.1, Poles 51.9, Swedes 52.0, Irish 50.8, Czechs 49.3, Germans 47.2, English 47.5, French 40.0, Portuguese 31.9, Basques 26.7, Sardinians 5.6.

Those absolute numbers should be handled with more caution than the ancient ones. The fits for the northeastern rows run between 60 and 83 units, three to four times worse than any Corded Ware row in this article, because modern Baltic, Finnic and eastern Slavic populations carry components this three and four source panel does not include. The gradient is real and it has been confirmed by far better methods. The second decimal place is not.

It is also worth showing why the obvious fix does not work. Add eastern hunter-gatherers as a fifth source to catch the missing northeastern signal, and the Latvian steppe figure drops from 57.3 to 48.7 while the fit improves from 82.7 to 82.1. The Polish figure drops from 51.9 to 48.9 and the fit improves by one tenth of a unit. A source that moves the composition by eight points and the fit by half a point is not being chosen on the evidence; it is collinear with the steppe pole, which is unsurprising given that the two sit 165 units apart and that the Yamnaya are themselves roughly half eastern hunter-gatherer. This is the same diagnostic that appears in every article on this site, applied here to a case where it kills an attractive-looking improvement.

What the numbers say

Three findings, in descending order of confidence.

The Corded Ware carried about seventy-two per cent steppe ancestry overall, and that ancestry was progressively diluted as the culture spread. It arrived in the eastern Baltic essentially unmixed at 96 per cent, it stabilised across the central European belt at 72 to 78, and it reached the western and northern margins at 59 to 62. The non-steppe remainder is specifically Globular Amphora rather than farmer-in-general, and a two-source Yamnaya and Globular Amphora model returns 71.5 and 28.5 per cent, matching what haplotype-sharing methods report from an entirely different direction.

The absorbed population was absorbed through its women. This is not measurable on the autosomes and this article does not claim to have measured it. What it can show is that the four Bohemian individuals buried with Corded Ware rites who carry no steppe ancestry at all are genetically indistinguishable from Globular Amphora farmers, that they sit more than two hundred units from the community around them, and that every one of them was female. The Y chromosome record says the same thing from the opposite side, and the quantitative estimates of the ratio remain contested.

The population closed. The spread of steppe ancestry among Bohemian individuals falls from a standard deviation of 14.6 percentage points before 2600 BC to 5.4 after it, with no change in the mean, over the same three centuries in which five male lineages became one. A society that had been taking in outsiders stopped taking in outsiders. That is a statement about social organisation, extracted from a coordinate file, and it is the closest this method gets to the thing the archaeology has been arguing about since the 1920s.

Note on sources and the coordinates used, for anyone who wants to check the arithmetic.

Global25 coordinates by Davidski (Eurogenes). Population averages from the Moriopoulos 2026 collection. Modelling carried out with non-negative least squares; readers who prefer a graphical tool can paste the block above into Vahaduo, written by Piotr Kapuscinski. Ancient DNA from Haak and colleagues 2015, Allentoft and colleagues 2015 and 2024, Saag and colleagues 2017, Linderholm and colleagues 2020, Papac and colleagues 2021, Ringbauer and colleagues 2024, and Lazaridis and colleagues 2025.