Sunday, 26 July 2026
Counting R1b: an audit of "The Great Ancient DNA Illusion"
On 25 July 2026, Robert John Langdon published a post arguing that the Bell Beaker migration model rests on a statistical illusion. Its central evidence is a table of 43 pre-Bell Beaker R1b individuals compiled, he says, from the Allen Ancient DNA Resource, together with a Mesolithic R1b individual from Aveline's Hole in Somerset said to move the starting point of British population history back by several millennia.
I have re-run the query against the source file. The table can be checked, and it does not hold. The Aveline's Hole claim has a specific and traceable explanation, and it is not the one offered.
What follows is the audit. The dataset is the AADR 1240k annotation file, release v66.p1, which anyone can download and reproduce.
What the post gets right
It is worth being clear about this first, because the post's opening three sections are sound and the rebuttal does not depend on disputing them.
Ancient DNA is fragmentary. Genomes are partially reconstructed rather than read. Coverage varies by orders of magnitude between individuals who appear side by side in a published table. Radiocarbon results are probability distributions, not dates. Bayesian chronological models are conditional on their priors, and a published range can be more precise than the underlying measurements alone would justify. Databases are samples, not censuses, shaped by preservation, excavation history and research priorities.
All of that is true. None of it is news — these caveats appear routinely in the supplementary material of the papers the post is criticising — but stating them for a general readership is a service rather than a fault.
The post is also correct that R1b existed in Europe before the Bell Beaker period. That has been published, mainstream, uncontroversial data since 2017.
The difficulty is what is built on top.
The category error
"R1b" is used throughout the post as though it named a population. It does not. It names a macro-haplogroup roughly eighteen to twenty thousand years deep, with two primary branches, and the distinction between its subclades is the entire substance of the question at issue.
The individual from Villabruna in northern Italy, dated to about 14,000 years ago, is R1b1a-L754. The Iron Gates Mesolithic individuals from Serbia and Romania are L754-derived. The Balkan Chalcolithic individuals from Pietrele, which supply eight rows of the post's own table, are basal or V88-type lineages. Blätterhöhle in Germany is V88.
None of these are ancestral to R1b-M269 > L51 > P312, the clade that expands across north-western Europe after 2500 BCE and that accounts for the overwhelming majority of modern British paternal ancestry. They are collateral branches. Their presence in Mesolithic and Neolithic Europe is a well-known fact that no researcher has ever disputed, and it bears on the Beaker question in roughly the way that the presence of wolves in Pleistocene Europe bears on the origin of the Labrador retriever.
The post's argument requires the reader not to notice this. Once noticed, the argument does not survive it.
There is a nomenclature trap here that catches honest readers too. AADR haplogroup strings have shifted across ISOGG releases, and "R1b1b" has denoted different nodes at different times. Anyone working from the strings rather than the primary publications will find them genuinely confusing. That is a reason for care, not a defence of the conclusion.
The chronological conflation
The post treats "pre-Bell Beaker" as though it meant "pre-steppe." It does not, and the gap between the two is where a fifth of its evidence sits.
Nine of the 43 are Bohemian: Plotiště nad Labem, Obříství, Vliněves, Stadice, Konobrže, dated between roughly 2919 and 2636 calBCE and carrying P312-derived subclades. These are Corded Ware burials.
Papac and colleagues, publishing in Science Advances in 2021, reported that Corded Ware appeared in Bohemia by about 2900 BCE and that R1b-L151 was already the most common Y-lineage among the earliest Corded Ware males there — six of eleven — with P312 most likely diversifying from within that pool. The same study documents the subsequent volatility with some precision: P312 rises to complete fixation in late Bell Beaker Bohemia and then falls to twenty per cent in preclassical Únětice, implying a minimum eighty per cent influx of new Y-lineages at the onset of the Early Bronze Age.
The Bohemian individuals in the post's table are therefore not counter-evidence to the steppe migration. They are its evidence, dated and labelled, filed under the wrong heading.
Aveline's Hole
Chapter 4 of the post rests on the identification of an R1b lineage from Aveline's Hole dating to the Mesolithic. No such individual appears in the post's own appendix. The only two British entries are Neolithic.
The sample exists. It is I3004, from Aveline's Hole, Burrington Combe, directly dated to 8606–8300 calBCE (OxA-34339), published in Brace et al. 2019. Its Y-haplogroup in AADR is R1b1a1b1a1a2a5a~, terminal SNP R-Y11281.
Its assessment in AADR is CRITICAL.
The hapConX X-chromosome contamination estimate is [0.212, 0.364] — between twenty-one and thirty-six per cent. The documented threshold at which AADR marks a sample CRITICAL or FAIL is a lower bound above 0.03. This sample exceeds that by a factor of seven. The second warning on the record is high.popgen.heterozygosity, the other standard signature of contamination. The individual carries 112,612 SNPs on 1240k targets.
The Y-call itself is the tell. R-Y11281 sits deep within P312 — which is what the modern British population overwhelmingly carries. A low-coverage British Mesolithic sample with a fifth to a third of its X-chromosome reads coming from somewhere else, returning a deeply derived P312 subclade, is not a discovery about the Mesolithic. It is a description of laboratory contamination, and the curators recorded it as such.
For completeness, the other Aveline's Hole individuals in AADR are I3006 (Mesolithic, female), I3005 and I3010 (both Early Neolithic, both female). This matches Brace et al. 2019 and the subsequent site-specific reassessment in the Proceedings of the University of Bristol Spelaeological Society: four individuals yielding genomic data, two Mesolithic with the expected Western Hunter-Gatherer signature, two Early Neolithic with Aegean farmer ancestry, separated by nearly five millennia and identified only when the crania were directly dated.
That last detail deserves emphasis, because the post has the moral of the story backwards. Aveline's Hole is the textbook case of undated cave material proving to be thousands of years younger than assumed. It is an argument for stringent chronological control, not against it.
The count
The post states that it examined every published prehistoric male dated before 2500 BC in the AADR, and reports 1,351 individuals of whom 43 are R1b, an observed frequency of 3.18 per cent.
I cannot reproduce those figures. Deduplicating on Individual ID as the AADR README instructs — the same individual can appear under several Genetic IDs, and counting rows inflates everything — and taking ancient males with a mean date at or before 2500 BCE:
|
|---|
The fourth row is the closest approximation I can construct to the post's denominator. Restricting to non-steppe Europe gives 1,452 males, near enough 1,351 that something of this kind was probably done.
That same filter contains 111 R1b individuals, not 43.
Sixty-nine are missing from a table captioned as listing all confirmed pre-Bell Beaker R1b individuals: six Iron Gates Mesolithic individuals from Serbia, eight from Latvia across the Kunda and Narva groups, twelve from Bulgaria including four from Varna, nine from Denmark, three further Czech Corded Ware, and others from Spain, Italy, Germany, Hungary, Slovakia, Sweden, Switzerland, Poland and Estonia.
The observed frequency is therefore not 3.18 per cent under any reading I can construct. It is 7.6 per cent on the post's own apparent geography and 16.3 per cent for Europe as a whole.
This does not support the post's argument. It damages it. The bulk of pre-2500 BCE European R1b sits in Yamnaya burials from Samara, Kalmykia, Orenburg, Rostov and Moldova, in Khvalynsk and Ekaterinovka, in Afanasievo, and in Corded Ware. Making the number larger makes the steppe expansion more visible, not less. The post's central statistic understates its own database by a factor of two and a half, and correcting the error strengthens the model it was assembled to refute.
The arithmetic that follows
Even taking 3.18 per cent at face value, the extrapolation to "approximately 8,000 to 16,000 R1b individuals" does not work.
A Y-haplogroup frequency is a frequency among males. Applying it to a total population of 250,000 to 500,000 counts everybody twice; the internally consistent figure would be roughly 4,000 to 8,000. The denominator spans about 7000 to 2500 BCE, mostly Neolithic and Chalcolithic, while the population estimate is Mesolithic — and European population grew by an order of magnitude across that interval, so combining them is not conservative but incoherent. Those population estimates themselves span more than a factor of two in the published literature and rest on sparse ethnographic analogy and climate-envelope modelling.
More fundamentally, multiplying any non-zero frequency by a large population yields a large number. That establishes nothing about descent. The Beaker model is not a claim that no R1b existed before 2500 BCE. It is a claim about which subclade expanded and left descendants. Since most of the 43 belong to lineages with negligible modern Western European paternal descent, the headcount is irrelevant to the proposition it is deployed against.
The post also invokes sampling bias in one direction only. If the database undercounts R1b males, it equally undercounts I2a, G2a and everyone else. A frequency is a ratio, and bias in the denominator does not preferentially inflate the numerator.
The 43
All 43 identifiers exist in v66.p1. The table is not fabricated, and it is worth saying so.
On AADR's own annotations, however:
- Eight carry
ASSESSMENT = Questionable: I6912, PNL001, I18101, JK2804, ATP3, I3035, I2611, OC. - Thirteen have contextual dates only, with no direct radiocarbon determination. The AADR file distinguishes these explicitly; the post's table preserves the distinction in its "cal BCE" versus "BCE" notation without remarking on it.
- Two carry the outlier group label
England_N-o— both British entries. - VLI011 and VLI015 are father and son, recorded as a first-degree pair. Two of the 43 are one observation. KON003 has two second-degree relatives at the same site; I14176 has a 2.5-degree relative.
- Twenty-five both pass AADR quality control and are directly dated. Twenty-four carry the literal
Passvalue; the twenty-fifth, I1590 from Blätterhöhle Cave, isMERGE_PASS.
"Forty-three confirmed" is not a description the source file supports. The point matters because the post's rhetorical structure depends on accumulation — that independent discoveries across many countries and laboratories cannot all be anomalies. Kinship, QC flags and undated contexts all reduce the number of independent observations.
The two British entries deserve individual attention, since they are the ones bearing on Britain.
I2611, from Summerhill, Blaydon, directly dated 3092–2905 calBCE and called R-L21, is the single most striking item in the table. Its AADR warning field reads: carbon date is unexpected for the archaeological context and genetic profile, with hapConX at [0.016, 0.038]. It is grouped as England_N-o, an outlier, and published in Patterson et al. 2021. The curators identified the anomaly before the post did, and recorded their view of it.
I3035, from Fox Holes Cave, has a contextual date only of 4000–3500 BCE and a U106-derived call. Its warnings record ANGSD contamination at [0.015, 0.028], hapConX at [0.014, 0.022], a library with technical problems, and that the observed and expected predicted dates differ by more than a thousand years. Another cave assemblage, another chronological mismatch.
What the evidence actually shows
The Beaker model does not rest on Y-haplogroups. It rests on genome-wide ancestry proportions, and the post does not mention them once.
Olalde and colleagues, in Nature in 2018, estimated that around ninety per cent of Britain's gene pool was replaced within a few centuries of the Beaker horizon, on the basis of autosomal ancestry. The alternatives the post proposes in its seventh section — gradual admixture, regional survival, cultural diffusion, multiple episodes — are not alternatives to the mainstream picture; they are components of it. Papac's Bohemian study documents Corded Ware males assimilating females of diverse local backgrounds. Work on Bronze Age Orkney has shown Neolithic I2a lineages persisting locally long after Beaker-derived ancestry arrives elsewhere in the genome.
But none of the four alternatives explains the observation the model was built to explain, and one of them cannot in principle. Cultural diffusion does not transmit autosomes. Pottery styles travel without people; steppe ancestry does not.
There is a final irony in the post's framing. It presents the Beaker migration model as an entrenched orthodoxy of more than two decades, self-reinforcing and resistant to revision. The model dates from 2015 and 2018. Before roughly 2010 the dominant view held R1b to be a Palaeolithic or Mesolithic Western European lineage — close to the position the post presents as suppressed heterodoxy. The consensus moved because the evidence moved, which is precisely the process the post claims does not occur.
Method, and where I might be wrong
The audit uses the AADR 1240k annotation file, release v66.p1, dated 8 June 2026. The post cites "AADR v66.1," which is not a version string the project has issued; the sequence runs v66.0 (12 April 2026), subsequently decommissioned, then v66.p1. The difference between those two releases concerns 161 modern Papuan individuals and cannot affect any count of prehistoric European males, so nothing substantive turns on it.
Two filter judgements are mine and are contestable. I deduplicated on Individual ID rather than counting rows, and I used the mean-BP date field rather than range endpoints. Both follow the README's guidance, but neither is the only defensible choice, which is why the sensitivity table above shows several variants rather than a single number. The gap between 43 and 111 survives all of them.
I could not reproduce 1,351 exactly, and it is possible I have misread the filter that produced it. The remedy is straightforward: publish the query. A count presented as a direct read of a public database, in a post whose thesis is that archaeologists hide their assumptions inside models, should be reproducible by anyone who downloads the file.
Sources
- Brace, S. et al. (2019) Ancient genomes indicate population replacement in Early Neolithic Britain. Nature Ecology & Evolution 3, 765–771.
- Schulting, R., Booth, T., Brace, S. et al. (2019) Aveline's Hole: an unexpected twist in the tale. Proceedings of the University of Bristol Spelaeological Society.
- Olalde, I. et al. (2018) The Beaker phenomenon and the genomic transformation of northwest Europe. Nature 555, 190–196.
- Papac, L. et al. (2021) Dynamic changes in genomic and social structures in third millennium BCE central Europe. Science Advances 7, eabi6941.
- Patterson, N. et al. (2022) Large-scale migration into Britain during the Middle to Late Bronze Age. Nature 601, 588–594.
- Mathieson, I. et al. (2018) The genomic history of southeastern Europe. Nature 555, 197–203.
- González-Fortes, G. et al. (2017) Paleogenomic evidence for multi-generational mixing between Neolithic farmers and Mesolithic hunter-gatherers in the Lower Danube basin. Current Biology 27, 1801–1810.
- Fu, Q. et al. (2016) The genetic history of Ice Age Europe. Nature 534, 200–205.
- Haak, W. et al. (2015) Massive migration from the steppe was a source for Indo-European languages in Europe. Nature 522, 207–211.
- Dulias, K. et al. (2022) Ancient DNA at the edge of the world: continental immigration and the persistence of Neolithic male lineages in Bronze Age Orkney. PNAS 119, e2108001119.
- Allen Ancient DNA Resource, v66.p1 (8 June 2026), 1240k annotation file. Harvard Dataverse.
Playing with the Stones: The New Locative Games at Avebury
Avebury has always been a place that rewards slow looking. The scale of the bank and ditch, the awkward tilt of individual sarsens, the way the landscape folds around the circle — these things reveal themselves gradually. Yet for many visitors the experience can still feel static: walk, look, read a panel, move on. In 2025 a group of researchers and designers tried a different approach. They turned the monument into a series of short, location-aware smartphone games.
The result is an anthology of roughly twenty free apps collectively known as the Avebury Adventures (or the Avebury Anthology). They form part of the larger European research project LoGaCulture — Locative Games for Cultural Heritage — run by Bournemouth University, the University of Southampton and the National Trust, with funding from the EU Horizon programme and UK Research and Innovation.
What the games actually do
Most of the apps use the phone’s GPS so that content appears only when you are standing in a particular place. Some add augmented reality, so virtual objects or characters sit on top of the real stones through the camera. Others are simpler narrative or puzzle experiences that unfold as you walk.
Examples include:
- Ages of Avebury (University of Southampton) — you take the role of a surveyor searching for missing stones and uncover the story of a disappearing archaeologist, with an AR reconstruction of a complete circle.
- Henge Hunts — follow trails and “hunt” the giant cattle, bears and wolves that once lived in the Neolithic landscape.
- Replicas, Uncovering Avebury, Avebury Research Challenge — more archaeological in flavour, involving collecting fragments, helping a fictional Professor Stone, or digging up knowledge around the stones.
- Lighter or more playful titles such as Don’t Harm the Sheep!, Journey Home (aimed at children), and even a cyberpunk-themed cyber@avebury.
There are also creative experiences where you collect stickers or visual elements to design your own Avebury postcard, and a couple of titles that can be partly experienced away from the site.
The games launched with QR-code boards in the visitor barn in late July 2025. Many remain available on both major app stores.
The thinking behind it
Locative games sit at the intersection of several research strands. First is the long-standing question of how people develop a sense of place and presence at heritage sites. Traditional interpretation often treats the visitor as a passive recipient of information. Locative and mixed-reality experiences try to make the visitor an active participant whose physical movement through the landscape drives the narrative.
Researchers in this field talk about “entanglement” — the idea that the digital layer, the physical stones, the visitor’s body and the stories become interwoven rather than sitting side by side. One of the 2026 papers from the project, The Relational Entanglement Diary, even proposes a method for evaluating these multi-layered experiences.
There is also a practical research question: can playful interaction increase engagement, especially among younger visitors or those who find conventional interpretation dry, without trivialising the archaeology? Early results from the Avebury trials suggest many people stay longer, notice details they would otherwise have missed, and report a stronger emotional connection to the site. Whether that translates into deeper historical understanding is harder to measure and remains an open question the academic papers are still exploring.
The technical side draws on earlier work in hypertext narrative, location-based storytelling systems, and the design of authoring tools that let writers and artists create these experiences without needing to be programmers. The Bournemouth team in particular developed frameworks (including the LUTE toolkit) to make locative game creation more accessible.
Where to find them
The simplest way to browse the Android versions is the Google Play search the project itself recommends:
https://play.google.com/store/search?q=avebury&c=apps
On iOS, search the App Store for “Avebury” together with titles such as Replicas, Henge Hunts, Ages of Avebury or cyber@avebury. Most of the apps are credited to LoGaCulture / Bournemouth University or the University of Southampton.
The central project site is https://logaculture.eu/. It contains background on the wider research, design patterns, and links to the games and related tools.
Key people involved include Dr Charlie Hargood (Bournemouth University, games technology), Professor David Millard (University of Southampton), and Dr Ros Cleal (National Trust curator at Avebury).
The academic papers
Two papers presented at the 2026 ACM Creativity & Cognition conference discuss the Avebury work in more formal terms:
- In Search of Lost Times: Reimagining Mixed Reality for an Ancient Site (Rimington, Baker, Blount, Jones, Jordan, Malinov & Millard) — DOI: 10.1145/3803784.3809283
- The Relational Entanglement Diary: A Novel Method for Evaluating Entangled Transmedia Engagement (Ferreira et al.) — DOI: 10.1145/3803784.3809275
They are written for a specialist audience and use the dense theoretical language common in interaction design research. The practical outcome — the apps themselves — is more approachable.
Thursday, 23 July 2026
Dropping the Central Trilithon: a physics toy
Dropping the Central Trilithon: a physics toy for Stonehenge's oldest cold case
Update: the model now shows stone 56 leaning, with a slider for its inclination at the moment of the fall. Gowland's 1901–02 record has the "leaning-stone" declining from 77° above horizontal in 1650 to a dangerous 61° by 1901 — that is, from 13° to 29° off vertical — while the collapse itself predates 1574, so a small lean at the fall (the default is 5°) is the sensible back-extrapolation. Two honesty notes. First, the lean is entered as scenery — 56 settling in its own deep socket — not produced by the fall: experiments with fully physical tenon joints showed the collapse transmits almost no overturning force to its partner. Second, the model quietly endorses the back-extrapolation: 56's lean moves the lintel's perch, and if you set the full 1901 angle of 29° the simulation hurls 156 about three metres past where it actually lies. The lintel's resting place prefers a nearly-upright 56 at the moment of collapse — which is just what the documentary decline implies. The stones and the archive agree, which is always worth a small celebration.
Update the second — the plan view, and the lintel's pirouette: looking at where the stones actually lie in plan, lintel 156 rests in front of stone 56, not out along 55's line of fall — which suggested it twisted as it went. Measuring Gowland's 1902 General Plan confirms it, precisely: 156's centre lies 4.5 m out along the fall direction but only 0.4 m west of 56's own centreline — dead in front of it — and its long axis bears about N27°W, roughly 72° (or, equivalently for a symmetrical slab, 108°) twisted from how it was bedded across the trilithon. Even 55b lies twisted the same sense, at about N21°W.
The mechanics of that twist fall straight out of the joint geometry. The lintel's two mortises sat about 2.4 m apart. When 55's top departed along the axis, the lintel could not keep both engagements: it had to swing like a gate, pivoting in plan on 56's tenon while 55 hauled its far end forward. That gate-swing has a hard geometric ceiling — once 55's top has travelled the full 2.4 m, the far mortise must have slipped off — which caps how long the tenons can possibly have dragged the lintel, at about 31° of 55's rotation. The model now enforces that ceiling, shows the whole thing in a small plan-view inset (drawn after Gowland's plan, with his measured position as a dashed ghost), and scores every run against the plan as well as the section: final twist against the measured 72–108°, and lateral position against "dead in front of 56."
Two consequences worth savouring. The tenon-grip knob is now over-constrained — it must satisfy the landing distance in section and the twist in plan simultaneously — and a sweet spot duly emerges: joints that slip too easily deliver the lintel untwisted; joints that grip to the geometric ceiling overspin it; only the middle of the range does both, which is the difference between a knob and a measurement. And one small prediction came free: for a gate-swing, the lintel centre's sideways offset is simply half the mortise spacing times the cosine of the twist — at the measured twist that gives 0.38 m west of 56's line. Gowland drew 0.4 m. The hypothesis didn't just accommodate the plan observation; it predicted its second coordinate.
Nobody recorded the tallest stones at Stonehenge coming down. Stone 55 of the Central Trilithon — the pair that once framed the winter-solstice sunset, and which most books will insist on calling the "Great" Trilithon (a friend who knows these stones far better than I do assures me the name properly belongs elsewhere, and I've learned not to argue with him) — fell and snapped in two at some unrecorded moment in antiquity, taking lintel 156 with it. By the time anyone drew the monument reliably, the wreckage was already lying where it lies today: 55a, 55b and the lintel sprawled across the recumbent Altar Stone, pinning it to the ground. Its partner, stone 56, leaned on alone until it was hauled upright and set in concrete in 1901.
That wreckage poses a question archaeologists still argue about: was the Altar Stone standing when the trilithon fell on it — or was it already lying flat? The Altar Stone has never been fully excavated, partly because the fallen sarsen pieces sit on top of it. So the resting positions of four stones are nearly all the evidence there is.
Which makes it a perfect problem for a physics sandbox. If every hypothesis has to end with the stones where they actually are, you can throw hypotheses at a rigid-body simulator all day and see which ones survive.
The toy
The simulation below (or [here], if the embed misbehaves) is a 2D cross-section along the solstice axis, built on the Matter.js physics engine at true scale — the uprights are 6.7 m proud of the ground, the lintel a metre-thick slab perched at over 7 m. Stone 56 doesn't fall but leans, how much is a variable to play with, as the evidence demands. Everything about stone 55 is a slider:
- how far it leaned before letting go, and how sharp the subsidence "kick" was;
- how deep its socket was — 56's was measured at a famous 2.4 m in 1901; 55's is thought to have been embarrassingly shallow, which is presumably why it's the one that fell;
- how strong the sarsen was (how hard an impact snaps it in two);
- how tenaciously the mortise-and-tenon joints dragged the lintel along before letting go;
- and, in the standing scenario, how firmly the Altar Stone was bedded.
Dashed outlines mark where the pieces lie today. After every run, a verdict card checks the outcome against them, and a "trench notebook" narrates the collapse: the base kicking out of its socket, the tenons shearing, the fracture, the strike. Three preset buttons reproduce the runs discussed below, so you don't have to take my word for any of it.
What the model says
The lazy expectation is that one scenario matches and the other doesn't. That's not what happens, and the way it fails is the interesting part.
Fall onto an already-recumbent Altar Stone, and everything works on nearly the first try. With middle-of-the-road parameters, stone 55 tips, its upper end strikes the raised edge of the flat slab at around 4 m/s, and it snaps over that edge like a bar over a fulcrum — the butt end flat at about 2 m, the upper half flat at about 5 m across the altar, the lintel coming to rest on top of the pile. That is, uncannily, the arrangement in the ground.
Fall onto a standing Altar Stone, and the knock-over itself is easy — almost too easy. A ~30-tonne slab sweeping through its arc fells a half-metre-thick pillar without much argument. But here's the result I didn't see coming: the collision absorbs so much of the fall's energy that stone 55 tends to land in one piece. The standing altar acts as a crumple zone. And an unbroken 55 contradicts the most solid fact we have — the real stone broke. To get the full observed sequence in the standing scenario (altar felled flat and 55 snapped), I have to dial the sarsen strength down to a conveniently flawed stone. Push the other way — bed the altar firmly — and you get a third outcome the ground flatly rules out: a permanent stalemate, with 55 propped against a still-standing Altar Stone like a failed game of dominoes.
So the model can't prove which history happened; with enough slider-turning, both scenarios reproduce today's arrangement, and that underdetermination is honestly the deepest lesson in it. But the two scenarios are not on equal footing:
Falling onto a recumbent Altar Stone breaks stone 55 for free — the slab's edge is exactly the anvil the fracture needs. Felling a standing Altar Stone instead cushions the fall, and the break has to be bought with an extra assumption about a weak stone.
Call it an argument from parsimony, delivered by a physics engine. It lines up with where much of the archaeology has been drifting anyway: many researchers suspect the Altar Stone was placed recumbent by design, a threshold rather than a pillar. The simulation adds a small, mechanical voice to that side of the debate: the flat-altar story needs nothing special to be true.
The Tenon Problem and an Insight
A real difficulty is that the mortise-and-tenon joints at Stonehenge were remarkably tenacious. Elsewhere on the monument we can see the evidence: stones 6 and 7 of the outer circle, one leaning out and the other in, still carried their twisted lintel until a scaffold was put up in 1881; observers at the time noted that without the joints that section would have collapsed long before. Stone 56 of the Central Trilithon itself stood for centuries at a lean of something like 12–15° after its partner and lintel had already fallen. A simple model that treats the joints as springs, or even as a clean kinematic coupling that releases after a few degrees of lean, cannot reproduce this. The joints were “sticky”; they held under conditions that look precarious to us.
What made the Central Trilithon different was almost certainly not a sudden failure of the tenons at the top, but the progressive failure of the shallow socket at the base of stone 55 — the heel being pushed or washed out. Once that footing gave way, the upright was free to rotate past the point of no return while the still-engaged joints dragged the lintel with it. Our 2-D simulations are limited because they cannot properly separate those two mechanisms: a base that can soften and release, and an upper joint that remains tenacious until the geometry finally forces it apart.
Even so, the simplified model yields two clear archaeological implications. First, trilithons fail from the base, not from the top; it is the heel coming out that allows the structure to fall. Second, the observed breakage of stone 55 is far more consistent with an already-recumbent Altar Stone acting as an impact surface than with a still-standing one. More elaborate modelling of the joints can wait; these two results are already visible.
Caveats, cheerfully admitted
This is a toy, and it wears its assumptions on its sleeve. It's 2D, so both uprights live in the same cross-section and the lintel's three-dimensional tumble is collapsed onto a plane. Fracture is a threshold on impact velocity at the contact point, not real crack mechanics; the tenon coupling — genuinely unknowable — is a parametrized "how long did the joints hang on" knob rather than simulated joinery. The stones are rigid rectangles on level ground. None of the slider ranges pretend to precision; they pretend to span the plausible, which is all a hypothesis machine needs.
The right way to read it: not "this is how it happened," but "here is the space of ways it could have happened, and notice which corner of that space doesn't need a coincidence."
Go push the stones over yourself. Start with the "already recumbent" preset, then try to make the standing scenario work without weakening the stone. That failure is the most informative thing in the whole toy.
Built with Matter.js. The trilithon geometry follows the standard published dimensions; socket depths per Gowland's 1901 excavation of stone 56. Nomenclature corrected under expert duress. All remaining errors of physics and prehistory are mine.
Friday, 17 July 2026
The A303 Drive-By Guide to Stonehenge
For the motorist doing 4 mph past a World Heritage Site
You are stuck in traffic on the A303. You know this because you are reading a blog post on your phone, which I trust means you are the passenger. Somewhere ahead, beyond a Eurobox towing a caravan, is Cornwall. Somewhere to your right, rather sooner, is Stonehenge.
The official advice is to visit the visitor centre, park a mile and a half from the stones, and take a shuttle bus. This is excellent advice for people with a spare half-day and a healthy tolerance for other people's children. You, however, have a holiday cottage to get to, and the geometry of the situation is entirely in your favour: the A303 passes within about 165 metres of the monument, considerably closer than the visitor centre does, and your current speed is ideal for detailed observation. What English Heritage sells as a premium experience, the Highways Agency provides free at the point of delivery.
So here is what you are actually looking at, in order, westbound from Amesbury. Eastbound readers should hold the guide upside down.
Countess Roundabout
The jam traditionally begins here, so consider it the ticket barrier. As you crawl off the roundabout and up the hill, the wooded lump immediately on your left is Vespasian's Camp — an Iron Age hillfort that has nothing whatever to do with Vespasian, who never went near it. The name is antiquarian enthusiasm, not history. At the foot of the hill, out of sight by the springs, is Blick Mead, a Mesolithic camp where people were feasting on aurochs thousands of years before anyone at Stonehenge lifted anything heavier than an opinion. You cannot see it. Very few people can. But you have now driven past the oldest bit of the story, which is more than most visitors manage.
The King Barrows
As the road climbs, look right at the clumps of beech trees along the ridge. Those trees are planted on the New King Barrows — a cemetery of Bronze Age round barrows, each one the burial mound of somebody who mattered around 4,000 years ago. The beeches are an eighteenth-century landscaping decision, which tells you the barrows have been furniture in a gentleman's view for longer than the United States has existed.
Stonehenge Bottom
The road then drops into a dry valley. Somewhere in the grass here, invisible from the car, the Stonehenge Avenue crosses it— the earthwork processional route that once linked the monument to the River Avon. Prehistoric people approached Stonehenge along it on foot, in ceremony. You crossed as you climbed the hill in an MPV, which is at least a novel form of procession.
The Main Event
And there they are, on your right, on the skyline and then broadside on. A few points the shuttle bus commentary won't make:
You are looking at the back. Stonehenge has a front — the north-east façade, with its neat run of lintelled sarsens, faces the midsummer sunrise and, conveniently for English Heritage, the paying customers. The A303 shows you the south and south-west side, which is the ruined side. Roughly half the monument is missing from this arc: stones fallen, broken up, carted off over the centuries for bridges, buildings and ballast. Whether that sector was ever properly finished at all was a live question until a dry summer in 2013, when parchmarks revealed the buried holes of missing circle stones — spotted, as it happens, because a hosepipe didn't reach far enough. The discoverers may be known to the management of this blog.
The tall one. The single upright towering above the rest is Stone 56, the surviving half of the Great Trilithon — the largest stones on the site. Its partner fell over long ago and lies broken. Stone 56 itself was leaning at an alarming angle until 1901, when Professor William Gowland winched it upright and set it in concrete. So the most imposing thing you can see from the road is, strictly speaking, an Edwardian restoration of a prehistoric monument. It is not the only one; a fair amount of what stands today was straightened and concreted in the twentieth century. This information is best deployed loudly at the summer solstice.
What you can't see. The Heel Stone, the famous outlier on the solstice axis, is on the far side of the monument by the line of the old A344 — closed in 2013 so that visitors could enjoy the stones without the sound of traffic. The sound of traffic was then supplied entirely by you.
The Left-Hand Side
While the driver gawps right, passengers should look left at the ridge to the south. That skyline is Normanton Down, one of the richest Bronze Age cemeteries in Europe. Among those mounds is Bush Barrow, whose occupant was buried with a sheet-gold lozenge of astonishing craftsmanship, now in the Wiltshire Museum in Devizes. Devizes has parking. Just saying.
Longbarrow Crossroads and Release
At the roundabout by Winterbourne Stoke you pass another barrow cemetery, anchored by a genuine Neolithic long barrow — several centuries older than anything standing at Stonehenge. Nobody photographs it. It has borne this with dignity for five and a half thousand years, and shortly afterwards the dual carriageway resumes and so, briefly, does your holiday.
A Note on Why You Are Stationary
You may reasonably ask why a single-carriageway trunk road still runs 165 metres from Britain's most famous monument. The answer is that for thirty years governments proposed to put the road in a tunnel, and in 2024 a government decided not to, after spending some £179 million on the scheme. In March 2026 the planning consent was formally revoked, so the not-building of the tunnel is now official and complete. You are, in a sense, driving through the result of the most expensive decision to change nothing in the history of British heritage. Do slow down and take it in. You will anyway.
Drivers: eyes forward, hands at ten and two. The stones have waited four and a half thousand years; they will still be there next jam.
Wednesday, 15 July 2026
Enhancing LIDAR with AI
A simple test to see how effective copying Lidar images into AI engines could enhance them. Click any to enlarge them.
The prompt used was:
"Enhance this lidar view to bring out every detail you can, recolour in the yellow- brown spectrum"
Tuesday, 14 July 2026
The Altar Stone Source - The Algorithm Result
"A transparent, multi-proxy desk screen—barium–rubidium
stream-sediment geochemistry (a proxy for the Altar Stone’s diagnostic baryte
cement and K-feldspar deficit), thermal-maturity mapping, clay mineralogy,
sedimentary facies, and detrital-zircon provenance—has been applied iteratively
across the UK Old Red Sandstone (ORS). It has narrowed a national-scale
problem, step by step and at the appropriate level of evidence, to a single
field-accessible target. The screen and Clarke et al.’s (2024, 2026) detrital-zircon
data agree that the source lies on the East Caithness coast; but the prime
barium cluster (Sarclet–Lybster–Clyth) sits in the high-maturity zone mapped by
Hillier & Marshall (1992), where vitrinite reflectance of 3–6% R₀
has driven the delicate expandable and aluminous clays of the Altar Stone past
preservation.
Two
independent lines then refine, rather than defeat, the result. The clay
evidence (Hillier & Clayton 1989; Hillier et al. 2006) shows the Altar
Stone’s tosudite–kaolinite–dioctahedral-chlorite assemblage to be an aluminous
sandstone diagenetic pathway. While classically expressed in the UK Lower ORS,
work on offshore Middle Devonian blocks (e.g., the Clair Group) demonstrates
that this signature is strictly facies- and fluid-controlled rather than
stratigraphically restricted, meaning it can be seamlessly accommodated within
porous sandstone bodies encased in lower-maturity segments of the Caithness
Flagstone Group.
One place is identified where such a rock could occur:"
I've included a photo of a megalithic monument that is there, but I need to just check it again before revealing the site.
The main problem is that the lack of data means that unsampled gaps in the record don't get filtered out, and that small areas are filtered out by being smeared in with the surrounding geology. Other data may highlight such areas, and suggest better screening. So such a desktop screening exercise can only suggest places worth further investigation, with a rock hammer.