The reason we have it is that it sank.
Around 87 BC, a ship went down somewhere off the Greek island of Antikythera, in about 45 meters of water. It was carrying bronze statues, glassware, fine pottery, and at least one object that did not belong in the same category as the rest of the cargo — a wooden box roughly the size of a shoebox, packed with interlocking bronze gears. In 1900, sponge divers found the wreck. The corroded lump sat in Athens for fifty years before anyone realized what they had. If the ship had arrived at its destination, the bronze would almost certainly have been melted down. Bronze was too valuable to leave in decorative form. What would have survived is exactly what we have from every other similar device from antiquity: nothing. We know the Antikythera mechanism because the sea hid it.
There are 30-odd surviving gears. Scholars now believe there were once 70 or more. The device is commonly described as an analog computer, which is technically accurate and also somewhat undersells it. Let me be more specific about what it could do.
The front face showed the positions of the Sun and Moon against the zodiac, the phase of the Moon, and — per a 2021 reconstruction by a team at UCL — the positions of all five planets visible to the naked eye: Mercury, Venus, Mars, Jupiter, Saturn. Turn the hand crank and the Greek cosmos moves. The planets track their orbits. The Moon waxes and wanes. The Sun advances through the year. It is a working orrery, the first we know of, small enough to carry.
The rear face carried spiral dials. The Metonic cycle: 235 lunar months equals 19 solar years, a period after which the Moon and Sun return to nearly the same position in the sky. The Saros cycle: 223 months, after which eclipses repeat in a predictable sequence. The Callippic cycle: 76 years. The Exeligmos: triple the Saros, 54 years. A Games dial tracking the schedule of pan-Hellenic athletic festivals. The device held, compressed into bronze, a complete calendar of the Greek world — astronomical, civic, sacred.
But none of that is what stops me.
What stops me is the eclipse predictions. Not just when an eclipse would occur — but its type, its time of day, its magnitude, the angular diameter of the Moon, the direction of obscuration, and its color. Eclipse color. This was inscribed on the Saros spiral as glyphs, each one a compact summary of a future event, computed from first principles and encoded in bronze. The device was not simply keeping track of when eclipses recur. It was characterizing them. That is not a calendar. That is something closer to a model of the sky — one that extracts qualitative predictions, not just dates.
The most mechanically astonishing part of the device is something called the pin-and-slot mechanism, and I want to describe it carefully, because it is easy to read past and miss what it actually is.
The Moon does not move at a constant speed. Its orbit is elliptical, so it moves faster when it is closer to Earth and slower when it is farther away. The Greek astronomer Hipparchus had worked out a mathematical model of this variation in the second century BC — he called it the lunar anomaly, and it was the first quantitatively accurate model of the Moon's irregular motion. The Antikythera mechanism implements this model in bronze.
Here is how. A gear called k1 has an off-center pin — not at the gear's center, but slightly displaced from it. That pin fits into a slot cut into a second gear, k2, which rotates around a different center. As k1 turns at a uniform rate, the pin traces a circular path, but because it is engaging k2 at a displaced center, it drives k2 with a non-uniform rotation. Faster at some points in the cycle, slower at others. The variation follows, approximately, the curve that Hipparchus computed mathematically.
This is not counting. It is not just tracking cycles or marking off intervals. The pin-and-slot is a mechanism that converts uniform rotation into non-uniform rotation in a mathematically prescribed way. It takes an abstract model of how the Moon accelerates and decelerates through its orbit and instantiates that model as a physical object. Hipparchus's equations became gears. This is the earliest known example of what we would now call differential gearing, and it was doing real computation — not just recording astronomical facts, but generating predictions that could not simply be looked up in a table.
The Exeligmos dial is subtler in a different way. The Saros cycle — 223 months — is extremely accurate, but it is not perfectly round. Each time a Saros cycle completes, the eclipse sequence repeats, but shifted by approximately eight hours. After three Saros cycles (the Exeligmos, 54 years), the shift has accumulated to a full day and the cycle closes. The Exeligmos sub-dial on the mechanism corrects for this: it tells you how many hours to add to your prediction based on where you are in the three-cycle sequence. In modern terms, this is carry arithmetic — the Saros cycle produces a fractional remainder that must be tracked and added in. The Exeligmos dial is the mechanism's carry bit, implemented as an analog glyph: 0, +8, or +16 hours. The device is not just counting astronomical periods. It is performing arithmetic on their remainders.
The Games dial is worth a brief stop. It tracked the schedule of pan-Hellenic athletic festivals — the Olympics and others — but the specific festival it includes is the Halieiad, a festival celebrated at Rhodes. This is a fingerprint. The Halieiad was local to Rhodes; you would include it only if the device was made there or for someone there. The pottery in the wreck and other physical evidence points the same direction. Rhodes was, in the first century BC, a major center of astronomical and mechanical science. The Stoic philosopher Posidonius worked there, and Cicero — writing just a few decades after the mechanism was made — describes having seen "sphaera" devices, mechanical models of the heavens, attributed to Archimedes. We have zero other surviving examples. Cicero saw one. We have one. The gap in the record is not the gap in the tradition.
This is what the maturity problem means. The Antikythera mechanism is not a first attempt. The engineering is too refined, the design too integrated, the calibration too precise for that. The pin-and-slot mechanism for the lunar anomaly, the Exeligmos carry correction, the interlocking of Metonic and Saros and Callippic cycles — these are the work of a tradition that had been developing for some time. We simply cannot find the earlier stages. No device anything like it appears in the archaeological record before it, and nothing like it appears again for more than 1,400 years afterward, until the mechanical astronomical clocks of 14th-century Europe.
Why did the tradition die? The honest answer is that we do not know. The knowledge did not entirely vanish — there are Byzantine calendar mechanisms, Islamic mechanical astronomy, threads that continue and eventually re-emerge in European horology. But the specific level of sophistication, the integration of an epicyclic gearing system with a mathematical model of lunar motion, the compression of a working cosmos into a shoebox — that disappears. Bronze was too valuable. Libraries burned. Institutions collapsed. The sea of lost things is larger than the sea that preserved this one object.
What I keep returning to is the color. Not the gears, not the cycles, not the elegant carry arithmetic of the Exeligmos — the fact that someone in 87 BC cared enough about eclipse color to encode it into a predictive system. That implies a theory. It implies a framework in which eclipse color is not just an observed phenomenon but a computable one, something that follows from the geometry of the Sun and Moon and the fraction of the disk obscured. It implies a level of understanding that makes the mechanism feel less like an instrument and more like a philosophy made physical — a claim that the sky is regular enough, and understood well enough, to be run forward in time by turning a crank.
The ship sank. The bronze stayed. And we are left with the question that every piece of wreckage leaves us: not just what was on this ship, but what was on all the others.