theHermeticum
PLES Subscribe
Front dial: the machine's Sun and Moon in brass, the real sky in red

The Mechanism

A model you can see the error of

Two thousand years ago someone cut the sky into bronze teeth. Turn the handle and watch where it goes wrong.

…
Egyptian calendar…
Next new moon, sky…
The same new moon, machine…
The machine is off by…

What if nobody had reset it since the day it was set, in 205 BCE?

Drag the handle: one turn is one year, as on the original. Keys: arrows for days, Shift for months, Page Up and Down for years.

Set the machine on
Back dials: the 19-year Metonic spiral above, the 223-month Saros eclipse spiral below

Machine against the sky

Every eclipse it marks, checked

On the back, a spiral of 223 cells counts the months from one eclipse season to its twin. Where the machine expects an eclipse it carries a glyph: Σ for the Moon, Η for the Sun, as on the original. Next to each, what the sky actually does.

SkyMachineError

Why it drifts

Twenty-three seconds a month

The Moon's train is four gears: 64 to 38, 48 to 24, 127 to 32. That is exactly 254 turns of the Moon in 19 years, and 235 months in the same 19 years. Bronze makes the month 29.530851 days. The sky's month is 29.530589.

The Moon gear train turning with the handle
    b2 64 → c1 38 · c2 48 → d1 24 · d2 127 → e2 32 = 254/19 = 13.3684 turns of the Moon for one of the handle
    How the error grows with distance from the day the machine was set
    Gold: average timing error of the eclipses it still catches. Red: share of real eclipses it catches at all. Each point is a century.

    The point

    The honest kind of model

    The oldest computer we know of was pulled from a shipwreck off Antikythera in 1901: bronze gears, a hand crank, dials for the Sun, the Moon and the eclipses. Its makers did not need to know why the sky moves. They knew how often it repeats, counted by Babylonian astronomers over centuries, and they cut those counts into teeth.

    It worked, and it drifted. Over a lifetime the drift is nothing. Over five centuries it costs a day and a half and about half of the eclipses. Nothing in the bronze is broken. A cycle is an approximation, and an approximation has a speed at which it goes stale.

    That is the question worth asking of any model you lean on, including the one that writes for you: where does it drift, how fast, and when do you check it against the sky?

    Gear counts after Freeth et al., Nature 444 (2006), and Freeth et al., Scientific Reports 11 (2021). Sky after J. Meeus, Astronomical Algorithms, 2nd ed. (1998), chapters 25, 47, 49 and 54; times in TT. The spring 205 BCE setting follows one proposed start of the eclipse dial (Carman and Evans, 2014). The machine's eclipse windows (16° from the node for the Sun, 11.5° for the Moon) are this model's choice, not a reading of the original. Original code.