Göbekli Tepe Pillar 43: does the 10,950 BC date beat chance?

The question

In 2017 Martin Sweatman and Dimitrios Tsikritsis read the animals on Pillar 43 at Göbekli Tepe, the "Vulture Stone", as star groups. The disc above the vulture's wing is the Sun, "slightly above the spout of the teapot" in Sagittarius on midsummer's day. Three "handbag" shapes, each with an animal, mark the other turning points: a bending bird for Pisces, an ibex for Gemini and a frog for Virgo.

That gives 10,950 BC, give or take 250 years, near the start of the Younger Dryas cold spell, with the odds against chance put at between one in 5 million and one in 100 million. Those odds count how the carved animals could be ordered, given the readings. They ignore the freedom to choose the readings and where the Sun sits, and how many events a date might land near. This test covers those. It tests one dating method, not whether the carvers watched the sky.

The test

Paul asked an AI model (Claude) to rebuild the method in code, on public data, and compare it with chance. The sky came from the Yale Bright Star Catalogue, the Vondrák precession model (the one the authors' Stellarium 0.15 uses) and IAU constellation figures.

The plan was written in a dated file before any chance model was scored. It fixed 14 notable events and what would count as "special". Two amendments were logged before scoring. The one that matters loosened the tie rule for choosing between dates, because the original sent the authors' own reading to 24,090 BC. Both versions were run, and the fit barely moved. Sampled models used 100,000 draws.

What came out

The fit is good if the readings are fixed first. The authors' reading, as rebuilt, scores 2.19 degrees of total misfit.

By the plan's rule the fit is special under the first two. But Sweatman's own papers read a frog as Virgo (later a bear, still Virgo), a bird as Pisces and a duck as Libra. That is freedom at the level of "some bird" and "some four-legged animal", and with it 51% of random pillars can be read into a fit as good.

The pillar cannot choose its own date. The four seasonal readings sit a quarter circle apart, so the fit is almost identical (1.93 to 2.19 degrees) at 24,090 BC, 17,780 BC, 11,256 BC, 4,630 BC and AD 1910. The age of the site does the choosing. Taken literally, the midsummer Sun sits at the spout tip in 11,256 BC, 306 years before the published date. At 10,950 BC it is 4.15 degrees away.

A date near an event is easy to find. Of the 217 random readings that fitted as well as the published one, 50% gave a date within 250 years of one of the 14 events. A date thrown at random into the same stretch of time does so 53% of the time. None of the 217 landed near the Younger Dryas dates on the list: 11,256 BC is the youngest the rule allows, 316 years from the nearest of them. Only an unplanned check, letting the Sun sit up to 4.15 degrees from the spout tip, makes the Younger Dryas reachable (39% within 250 years).

So the claim that Pillar 43 encodes a date near a notable event, beyond what the method's freedom produces, is not supported: the plan needed free-reading pillars to fit under 1% of the time, and they fit 51.3%. Eleven further runs, varying the model and settings, changed no conclusion.

What this does not show

It does not show that the carvers ignored the sky. The fixed-reading fit, about 1 in 290 to 460, is the one result here that points towards intention, and it stands. A test with readings fixed first, on a second pillar, could still find something.

The limits:

• The carvings were taken from the authors' descriptions, not excavators' photographs. No other pillar, Karahan Tepe or later paper (such as Sweatman's 2024 calendar of V marks) was tested.
• "Contains the Sun" was decided by star positions or official boundaries. The authors judged by eye.
• The event list was chosen by the study and is crowded between 12,900 and 9,300 BC, which raises hit rates.
• Quotations came through a web summariser and are still to be checked against the papers.

The strongest objection is that the free-reading model is too free: under Sweatman's exact readings, random pillars fit about 1 time in 1,850. The excavators' separate objection, that 10,950 BC is 700 to 1,000 years older than the oldest radiocarbon date for Enclosure D, is not part of this test.

Where it goes in the books

This is Book 6, The Sky Remembers. Chapter 2, "The Vulture Stone", gives the short version and the Interlude, "The Machine Checks the Sky", the full account.

Check the first result yourself. Set a planetarium program to midsummer, step back towards 10,950 BC and measure how far the Sun is from the spout of the Teapot. The code says 4.15 degrees; a different precession model will move it. The author holds the pre-registered plan, the notebook and the code, and you can ask for them through the contact form.