ENCS
Wednesday, 19 August 2026
Your paper, your pace.
Wonder

The Day Is Getting Longer by 1.78 Milliseconds a Century. Broken Clay Tablets Are How We Know.

Scribes in Babylon timed eclipses against a water clock and pressed the results into clay. Twenty-seven centuries of such records show the Earth slowing down - and slowing rather less than the tides say it should.

8 min 8 sources Confidence 92/100

In short

What happened. An analysis published in Proceedings of the Royal Society A in 2016 used eclipse records from 720 BC onwards to measure how fast the Earth turns. The length of the mean solar day is growing by 1.78 milliseconds a century.

What it means. That is less than it should be. Tidal friction on its own predicts 2.3 milliseconds a century, so something is quietly handing spin back to the planet.

Risks and impact. Nothing you can feel. It is the reason atomic clocks and clocks based on the sky drift apart, and the reason leap seconds exist.

What can be done. The International Earth Rotation and Reference Systems Service publishes the measured length of each day. Anyone can look it up.

What to watch. Whether the missing half-millisecond is ever fully explained, and whether the long, slow swing in day length that the records only hint at turns out to be real.

Shown as a summary because of your reading settings.

What happened

A flat roof in Babylon, well after dark. A bowl of water empties through a small hole while a man watches the moon go the wrong colour. When it is over he presses wedge marks into damp clay: the night, the hour, how much of the disc went dark.

📖 What the sources say. Nearly all the known Babylonian astronomical texts were acquired by the British Museum in the late nineteenth century, and about 2,000 devoted to astronomy are kept there. Fewer than one in ten of the original records is thought to survive.

🏺 What the evidence supports. Those tablets carry 180 timings of lunar and solar eclipses from Babylon, spread between 720 BC and 9 BC. Only two of them describe a total eclipse of the sun: one in 136 BC, one in 10 BC.

🎬 Reconstruction (we are imagining). The roof, the bowl, the man. Babylonian timings were taken with water clocks, but this particular night is invented.

In December 2016, Richard Stephenson, Leslie Morrison and Catherine Hohenkerk published in Proceedings of the Royal Society A a new compilation of those eclipse records, together with Chinese, Greek, Arab and medieval European ones down to AD 1600, and 478,843 recorded observations of the moon passing in front of stars, beginning in 1623.

Set end to end, that chain measures one thing: whether the Earth turns at a constant rate. It does not. In their words, “the change in the length of the mean solar day increases at an average rate of +1.8 ms per century” - more exactly +1.78, give or take 0.03.

Then comes the awkward part, in the next sentence of the same abstract: “This is significantly less than the rate predicted on the basis of tidal friction, which is +2.3 ms per century.”

What the evidence supports

Three things are firm. The Earth’s rotation is not uniform, and the departure is large enough to be read off records made with a bowl of water. The modern half of the chain is enormous and precise. And the tidal prediction is not guesswork: bouncing lasers off the moon gives its tidal acceleration as -25.82 arcseconds per century squared, to within 0.03.

One thing is thin, and the authors say so. The ancient half rests on very few points: 180 Babylonian timings, 111 Chinese ones between AD 434 and 1280, more than 50 Arab measurements from about AD 830 to 1020, and 11 Greek. Every one of them depends on a translator’s judgement about what a fragmentary text is describing.

And one thing is genuinely open. The gap between 2.3 and 1.78 milliseconds means some mechanism is handing spin back to the planet - an acceleration the paper puts at 1.5 times ten to the minus twenty-two radians per second squared, plus or minus 0.4. Their explanation is careful rather than confident: it is “probably in part” the solid Earth springing back after the last ice sheets melted, redistributing mass towards the poles, but “by itself, this mechanism cannot account completely for the non-tidal acceleration, and some additional correction for core-mantle coupling is required.” They also report “some indication of an oscillation in the lod with a period of roughly 1500 years”, and extra wobble at around six years. Nobody has nailed either down.

The background

Start with why milliseconds turn into hours. If every day is a shade longer than the one before, the error does not simply add up - it compounds, because you are accumulating a lag that itself grows. The team fits the whole 2,700-year record with a single parabola: the accumulated difference in seconds equals 32.5 times the square of the number of centuries since 1825, minus 320.

Put 720 BC into that and it returns about 20,700 seconds. Just under six hours. The Earth turns fifteen degrees of longitude in an hour, so anyone calculating that eclipse on the assumption of a steady spin would have the planet turned roughly eighty-six degrees out of position. At the moment of the event, their arithmetic would put Babylon somewhere over the Atlantic. That is the measurement. The tablets say where the shadow was; the arithmetic says where it should have been; the difference is the slowdown.

The cause is the moon, and it works like a hand on a flywheel. The moon’s gravity raises a bulge of water and rock; the Earth spins faster than the moon orbits, so friction drags that bulge slightly ahead of the moon rather than letting it sit under it. The bulge then pulls the moon forwards, lifting it into a wider orbit, and the same tug in reverse drags on the Earth’s spin. The planet’s rotation is paying for the moon’s altitude.

You can watch the payment. Apollo 11, 14 and 15 left arrays of mirrors on the lunar surface, and observatories have been timing laser pulses off them ever since, measuring the distance to within a few millimetres. The gap widens by 3.8 centimetres a year.

Life kept its own records. A single fossil clam of the species Torreites sanchezi, collected in the Al-Hajar Mountains of Oman and about 70 million years old, was sampled by laser at ten-micrometre steps to count its daily growth layers. There were 372 in a year. The year itself has barely changed in length, so the day was about 23.5 hours.

One caution, from the same work: today’s recession rate cannot be run backwards in a straight line. Doing so would put the moon inside the Earth 1.4 billion years ago, and the moon is far older than that.

All of this reaches your pocket in one place. Coordinated Universal Time is held within 0.9 seconds of time told by the Earth’s actual rotation, and when the gap threatens to open, the International Earth Rotation and Reference Systems Service, which measures the planet’s rotation, announces a leap second. Those corrections began in 1972. A time and frequency bulletin published by the US National Institute of Standards and Technology in 2024 puts atomic time 37 seconds ahead of civil time, and records that no leap second was added at the end of that June.

Who it touches

The scribes were not doing astronomy for us, and they were not doing physics. A study of the same tablet series describes the diaries as written and compiled by “the families of astronomer-astrologers” and sponsored by the assembly of Esangil, the temple of Marduk in Babylon. The earliest known tablet in that series is dated 652 BC, the latest 61 BC. These were people whose job was to read the sky for meaning. Their theory of what they were seeing has not aged well. Their timekeeping has.

Almost none of them left a name. What they left was a habit: write down the night, write down the hour, do not round it off. Most of what they wrote is gone. The tablets were dug up at Babylon in the late nineteenth century and taken to the British Museum, and the same study puts what survives at five to ten per cent of the original run.

On the other end of the chain, Stephenson, Morrison and Hohenkerk spent much of their working lives assembling those fragments alongside Chinese court records, Arab measurements and four centuries of telescopic occultation timings, so that the whole run could be treated as one instrument.

The deeper story

There is something strange in the fact that this measurement exists at all.

No instrument built on purpose could have made it. To detect a drift of 1.78 milliseconds per century you need a baseline of thousands of years, and nothing designed for the job has been running that long. The only device with a long enough arm turned out to be a pile of administrative notes, kept by people who were watching for omens and had no concept of what they were storing.

What made those notes usable was not insight. It was discipline about a detail: the time. A scribe who wrote “there was an eclipse” would have left nothing. A scribe who wrote how long after sunset it began left a data point that would still be working twenty-seven centuries later, long after the question it now answers had been invented.

That is worth sitting with, and so is its shadow. Fewer than one record in ten survived. The chain is thin exactly where it is oldest, and every fragment that broke took a measurement with it. What we have is partly the product of care and partly the product of luck, and the two are no longer separable.

And the answer is still not clean. The tides say the day should be growing by 2.3 milliseconds a century; the sky says 1.78. Half a millisecond a century is unaccounted for, parcelled out between rebounding continents and something happening between the core and the mantle that nobody can watch directly. A result that arrives with its own unfinished business is usually a sign that the measuring was honest.

The planet under your chair is running down, imperceptibly, and the record of it was kept by people who thought they were reading messages.

Something to sit with

What are you in the habit of writing down precisely, without knowing what it is for?

If most of what you recorded were lost, would the tenth that survived still be worth having?

Sources

We report facts from the sources above in our own words and link to the originals. Interpretation is ours, not theirs.

Keep reading

♻︎ Free to republish

Copy this HTML into your CMS. Credit line and licence are included. Republish our work — free

Every headline has a deeper story. This is ours.

What we are doing here

My Paper in your inbox

Every morning, or once on Saturday — you choose. The story worth your time, and the question underneath it. One click to leave.

One click to leave. We never sell or share your address.