Is Göbekli Tepe's Pillar 43 really a 13,000-year-old star map?

In 2017, physicist Martin Sweatman published a paper arguing that animal carvings on Pillar 43 at Göbekli Tepe represent star constellations encoding a precise date — 10,950 BC — corresponding to the Younger Dryas catastrophe. In 2025 he and a co-author stated a probability of about 1 in 1.4 million. If the astronomical reading is right, it would push systematic astronomy back several millennia.

Does the statistical evidence actually support that interpretation?

Sweatman's headline number — about 1 in 1.4 million — is not "rare date times rare picture." It is the chance of getting animal-to-constellation ranks this good, multiplied by a small factor (2/7) for layout details. There is no Younger Dryas date inside that product. On his own sensitivity check, the same headline can drop to about 1 in 110,000.

The date, tested separately, is not rare: nearly half of the years from 20,000 to 5,000 BC sit within 250 years of some listed catastrophe. The visual matches are genuinely unusual, but three of the six carvings do almost all the work — about 1 in 85,184 if those three were named in advance, or about 1 in 4,484 if you look at all six and then keep the best three. Those ranks were scored by the person whose theory depends on them.

Göbekli Tepe is a monumental stone site in southeastern Turkey, built around 9,500 BC. That's about 6,000 years before Stonehenge and 7,000 years before the Great Pyramid. It consists of massive T-shaped pillars arranged in circles, many of them carved with elaborate animal reliefs — scorpions, foxes, vultures, snakes, boars. Nobody disputes its age or its importance. It's one of the oldest known examples of monumental architecture on Earth.

Pillar 43, known as the "Vulture Stone," is the most densely decorated pillar at the site. Its broadside is covered with animal carvings: a large vulture with a disc balanced on its wing, a scorpion, a wolf-like animal, a tall bird grasping a snake, and several others.

The archaeological team that excavated the site interprets these as funerary art — scenes related to death rituals and excarnation (the practice of exposing the dead to vultures). The disc might be a severed head. The animals might be symbolic representations tied to burial customs.

In 2017, a physicist named Martin Sweatman proposed something radically different. He argued that these animals aren't just animals. They're constellations. And together, they encode a precise date: 10,950 BC.

That date corresponds to the Younger Dryas — a sudden, extreme climate collapse that plunged the Earth back into near-glacial conditions for over a thousand years. Temperatures dropped several degrees within decades. Megafauna went extinct across multiple continents. Human populations were severely disrupted. Some researchers believe it was triggered by a comet or asteroid impact. It's one of the most dramatic environmental events in human prehistory.

If Sweatman is right, the implications are enormous. Enclosure D at the site was built centuries after 10,950 BC — so the carving would be a memorial to something ancestors had lived through. It would also mean people 13,000 years ago tracked the stars, recognized constellations, understood that the sky shifts over millennia, and encoded that knowledge in stone. That would push systematic astronomy back several millennia before the better-known Egyptian and Mesopotamian traditions.

This is why the claim has attracted so much attention beyond academia. It feeds directly into the idea — popularized by Graham Hancock — that an advanced civilization existed before the last ice age and was wiped out by the Younger Dryas catastrophe, with only fragmentary memories surviving in ancient monuments. If the Pillar 43 reading is correct, Göbekli Tepe becomes evidence for that thesis.

His statistical significance claim: 1 in 1.4 million. In other words, the probability that these animal carvings match constellations by pure coincidence is, according to him, so vanishingly small that the astronomical interpretation is essentially proven.

I built an independent statistical pipeline to test that number. 13 experiments, ~1,500 lines of code, every result reproducible from a single script. The code and data are public.

Here's what I found.

How the claim works

The constellations are always in the sky, every year, every era. Scorpius, Sagittarius, Orion — they don't come and go. What changes over thousands of years is where the sun sits on specific calendar days.

Right now, on the summer solstice (the longest day of the year), the sun is in the constellation Gemini. But this slowly shifts. About 2,000 years ago, the summer solstice sun was in Cancer. 6,000 years ago it was in Leo. This drift is called precession, and it completes a full cycle every 25,772 years.

This means each constellation gets a turn as the "summer solstice constellation," and each turn corresponds to a specific era. It's like a clock that takes 26,000 years to go around once. If you know which constellation the sun was in on the summer solstice, you know roughly what millennium you're in.

Sweatman says: the vulture carving represents Sagittarius. The disc on its wing represents the sun. Putting the sun inside Sagittarius on the summer solstice — that only happens around 10,950 BC. That's the timestamp.

The other animals — scorpion, a canid (wolf-or-dog; that species call is still tentative), bird-with-snake, ibex, frog — are additional constellations that fill out the scene. He compared each carving to a pool of 44 classical constellations and ranked them by visual resemblance. A rank of 1 means best match out of 44. A rank of 44 would mean the worst.

His results: the scorpion carving most closely resembles Scorpius — rank 1 out of 44. The canid carving most closely resembles Lupus — rank 1 out of 44. The bird-with-snake most closely resembles Ophiuchus (a figure grasping a serpent) — rank 1 out of 44. Three top matches.

The other three are decent but not perfect: the bending bird matches Pisces at rank 4, the ibex matches Gemini at rank 5, the frog/bear matches Virgo at rank 6.

From those six ranks he gets a probability of about 1 in 390,000 for ranks this good by chance. He then multiplies by 2/7 — a layout-symmetry factor and a spatial-correlation allowance from earlier work — and arrives at about 1 in 1.4 million. There is no Younger Dryas date inside that product. The date is a separate claim. On his own sensitivity check, switching stick-figure sets moves the same headline down to about 1 in 110,000.

Problem 1: The date isn't special

The period from 20,000 to 5,000 BC was catastrophically eventful. Ice ages ending. Massive floods. Supervolcanic eruptions. Abrupt climate reversals. Species extinctions. The Younger Dryas is the most famous of these events, but it's far from the only one. There's the Older Dryas (a similar cold snap about a thousand years earlier), the Bonneville megaflood, the Heinrich Event 1 iceberg collapse, the Bølling-Allerød rapid warming, and at least a dozen more.

I compiled a catalog of 14 event clusters from the published paleoclimatic record and asked a simple question: how much of this era is "near" a notable disaster?

The answer: about 42%. Nearly half of the entire 15,000-year window is within 250 years of at least one catalogued event. Pick a random date from this era and you have close to a coin-flip chance of landing near something terrible.

Then I tested all 52 possible constellation-and-cardinal-day combinations (13 ecliptic constellations × 4 cardinal days — spring equinox, summer solstice, autumn equinox, winter solstice). Each combination produces a different precession date. Roughly a quarter to half of them land near a catalogued event, depending on whether you look at the center date or the full window.

Sweatman's Sagittarius at summer solstice is just one of many that work. And here's the twist: removing the Younger Dryas itself from the catalog doesn't change the center-date hits. The Sagittarius center date is actually closer to the Older Dryas than to the Younger Dryas onset. The chosen timestamp doesn't even point cleanly at the event he claims it commemorates.

Landing on a date near a catastrophe during the deglacial era is like landing near a restaurant in Manhattan. It doesn't mean someone was aiming for one.

Problem 2: The visual similarity is real — but three carvings do all the heavy lifting

Now here's where it gets interesting, because the visual matches aren't fake.

Remember, Sweatman ranked each of his 6 animal carvings against 44 constellations. If the carvings had nothing to do with constellations, you'd expect random ranks — something like 20, 30, 15, 38, 7, 42. Add those up and you get a sum around 135. That's what chance looks like.

Sweatman's ranks are [1, 1, 1, 4, 5, 6]. His sum is 18. That's wildly below what chance predicts.

I computed the exact probability of getting a sum this low or lower by pure chance: about 1 in 391,000. That's genuinely unusual.

But then I did something Sweatman never did. I looked at the six matches in two groups.

The three best — scorpion → Scorpius, canid → Lupus, bird-with-snake → Ophiuchus — all ranked 1st. If those three animals were named before anyone saw the ranks, three rank-1 draws are about 1 in 85,184. If you look at all six first and then keep the best three, the chance of at least three rank-1s drops to about 1 in 4,484. That second number is the fairer one for a "best of six" reading.

The three weakest — bending bird → Pisces (rank 4), ibex → Gemini (rank 5), frog/bear → Virgo (rank 6) — together produce a probability of about 1 in 187. Still notable, but far weaker.

So the impressive overall number is mostly coming from just three carvings. Remove the scorpion and the significance drops by a factor of about 15. Same for the canid or the bird-with-snake. Remove the frog/bear instead and it barely moves.

What this means: Sweatman presents six constellation matches as though they're all contributing equally. In reality, three of them are doing almost all the work. His 1.4 million is the full rank-sum times 2/7 — not the top-three figure, and not a rare date multiplied by a rare picture.

Problem 3: The rankings are self-scored

Sweatman ranked how well each animal matches each constellation himself. All 264 pairs (6 animals × 44 constellations). He proposed the hypothesis, and then he generated the data that his own statistical test evaluates.

This isn't an accusation of dishonesty. Confirmation bias in visual pattern matching is well documented. When you know what you're looking for, you tend to find it. And here, the person deciding "how much does this scorpion look like Scorpius?" is the same person whose theory depends on the answer being "a lot." That's a setup for inflated results, even with the best intentions.

How sensitive is the result to those specific rankings? Imagine a slightly more skeptical scorer who looks at the same carvings and says: "These are good matches, but maybe not quite rank 1 — more like rank 2." Just one notch less generous across the board.

That single-notch shift weakens the six-animal result by roughly a factor of seven. Push the top three down to rank 3 and the top-three piece moves from about 1 in 85,000 to about 1 in 1,000. The signal doesn't disappear — but its strength is highly sensitive to subjective scoring decisions made by the person whose theory depends on them.

The entire statistical case balances on the knife's edge of three subjective judgments, and nobody outside Sweatman's group has ever independently evaluated them.

Problem 4: You can't separate the match from the name

This one is subtle but possibly the deepest issue.

Why does scorpion → Scorpius work so well as a visual match? Think about it. The constellation is called Scorpius because the ancient Greeks thought those stars looked like a scorpion. Wolf → Lupus works because Lupus is literally the Latin word for wolf. Ophiuchus means "serpent-bearer" — a figure grasping a snake.

The constellation names carry the animal identity built in.

So when Sweatman holds up a scorpion carving and says "look, it matches Scorpius!" — is that because Neolithic people independently saw a scorpion in that patch of sky 13,000 years ago? Or is it because the Greeks later named those same stars "the scorpion," and now any scorpion carving from any culture in any era will automatically match?

There are scorpion carvings all over the ancient world — Egypt, Mesopotamia, Mesoamerica. None of them are assumed to represent the constellation Scorpius.

Any ancient culture that carved common animals — scorpions, wolves, birds, bears — will produce matches against a constellation set that was named after the same common animals. The match might be real evidence of a shared astronomical tradition stretching back into deep prehistory. Or it might be an inevitable collision between two naming systems that both drew from the same pool of familiar creatures.

It's possible in principle that the Greek constellation names preserve a much older tradition — that Scorpius was already "the scorpion" long before Greece. But that would need independent evidence: recurring animal-star associations across unrelated cultures, or matches that aren't trivially explained by the animal's name.

This isn't hypothetical. At least one other independent analyst has applied the same approach to the same pillar and arrived at a completely different mapping — different constellation assignments, different animal names, and a date of 14,800 BCE instead of Sweatman's 10,950 BC. Both presentations look compelling when laid out visually. Both produce dates near major catastrophes. They can't both be right — but the fact that the same carvings yield multiple convincing-looking readings, separated by nearly 4,000 years, tells you how much flexibility the method allows.

The current statistical framework cannot tell the difference. And that's the problem.

How researcher choices change the result

Every statistical analysis involves choices. What counts as a "match"? Which events do you include in your catalog? What margin of error do you allow? Do you test only the summer solstice, or all four cardinal days? Do you use all six animal carvings, or just the best ones?

None of these choices are obviously wrong. But each one nudges the final number.

I tested many combinations of these choices. Same data, same carvings, same constellations — just different reasonable methodological decisions. The joint result moves a lot depending on the path. On the paths where Sweatman's published Sagittarius reading is actually counted as a hit, about one in five are at least as "significant" as a simple exploratory reference. He isn't reporting the significance of the evidence in general. He's reporting the significance of one particular path through a garden of forking choices.

What I'm NOT saying

I am not saying Sweatman is wrong. Three of his six animal-constellation matches are strikingly good, and I haven't found a satisfying explanation for why scorpion, canid, and bird-with-snake all rank 1st against 44 options.

I am not saying the carvings can't be constellations. They might be.

I am not saying Göbekli Tepe isn't remarkable. It's one of the most important archaeological sites on Earth.

What I am saying: the 1 in 1.4 million should be read as what it is — a rank-sum times 2/7, with a published sensitivity down to about 1 in 110,000. It should not be read as a rare date multiplied by a rare visual match. The date, tested separately, is common on this catalog. The visual weight sits in three self-scored ranks, and it is about 1 in 4,484 once those three are recognized as the best of six.

The visual signal remains statistically surprising even after conservative adjustments. But its meaning depends entirely on accepting that these carvings are legitimately being compared to the adopted constellation set. If that interpretive step is wrong — if the carvings are just animals in a funerary scene — then the probability loses its force, no matter how small it is.

What would help settle this

The most obvious next step would be a blind ranking experiment: show the animal carvings to people who know nothing about Göbekli Tepe or archaeoastronomy, ask them to match each animal to its most similar constellation figure, and see if scorpion, canid, and bird-with-snake still come out on top without any prompting.

That would test whether Sweatman's rankings are reproducible by independent observers.

But even a perfect blind replication wouldn't fully settle the question. If everyone agrees the scorpion carving looks like Scorpius, that might just confirm what we already know: scorpions look like Scorpius because the Greeks named it after a scorpion. It wouldn't prove that Neolithic people were mapping the sky.

The carvings might be constellations. They might be animals in a funerary scene, exactly as the excavation team believes. The statistical evidence, as it stands, cannot distinguish between these possibilities as cleanly as Sweatman claims. A 1 in 1.4 million headline makes it sound like the debate is settled. It isn't.

All code, data, and figures are open source:

→ GitHub: github.com/simossss/gobekli-pillar43-statistics

→ Corrected paper (23 October 2026): https://doi.org/10.5281/zenodo.23593750

→ All versions: https://doi.org/10.5281/zenodo.20364946

This project was built entirely with AI systems in distinct roles. Claude handled technical reasoning, statistical design, code development, and manuscript drafting. ChatGPT served as an adversarial reviewer across roughly 10 review cycles, catching real errors each time — incorrect significance framing, a flawed negative control experiment, an invalid combined p-value, and inconsistent multiverse reporting. The analysis ran on an NVIDIA DGX Spark.

The script grew from 4 experiments to 13 through iterative conversation, not a predefined plan. Each result raised a new question, and each review cycle tightened the methodology. The decision not to run a blind visual matching study — and the three-layer argument that replaced it — came out of a conversation about why such a test would be inherently ambiguous.

My role was scoping, sequencing, and final judgment. I decided which questions to ask, which reviewer suggestions to accept, which framing to use, and when to stop. AI provided the throughput. I provided the direction.

Full process writeup →

• Sweatman & Tsikritsis (2017). "Decoding Göbekli Tepe with Archaeoastronomy." MAA 17(1), 233–250. https://doi.org/10.5281/zenodo.400780
• Sweatman & Gerogiorgis (2025). "Origin of some ancient Greek constellations via Pillar 43." Submitted manuscript posted by the authors; not peer-reviewed. https://martinsweatman.blogspot.com/2025/05/new-paper-submitted-origin-of-some-of.html
• Notroff et al. (2017). "More than a vulture: A response to Sweatman and Tsikritsis." MAA 17(2), 57–63. https://www.dainst.blog/the-tepe-telegrams/2017/07/03/more-than-a-vulture-a-response-to-sweatman-and-tsikritsis/
• Peters & Schmidt (2004). "Animals in the symbolic world of PPN Göbekli Tepe." Anthropozoologica 39(1). https://catalogue.frantiq.fr/bib/229734
• Silva (2020). "A Probabilistic Framework for Skyscape Archaeology." JAS 118. https://doi.org/10.1016/j.jas.2020.105138

• Corrected paper on Zenodo (23 October 2026). All versions: 10.5281/zenodo.20364946.
• Code and data on GitHub