The Sky Was Not Regular Enough

One 86,400th of it. Twenty-four hours, sixty minutes, sixty seconds — multiply and that is the whole of it.

The definition pointed at something. Not at a prototype in a vault in Paris, the way the kilogram did, but at the sky. If you wanted to check a second you watched a star cross a wire, waited for it to come round again, and divided.

Everyone was inside the standard. The Earth turned under all of us at once, and the unit was a share of that turning. Nobody owned it and nobody had to be trusted about it.

The rest of this zine is the story of that tie being cut — carefully, deliberately, by people with good reasons, in three documented moves. None of them was a mistake. That is what makes it worth following.

documented  the mean solar day → the unit

Quartz clocks arrived, and for the first time in the history of measurement the clock was more regular than the Earth.

In 1936, A. Scheibe and U. Adelsberger, working with quartz-crystal clocks at the Physikalisch-Technische Reichsanstalt in Berlin, reported that the Earth's rotation varies with the season. In 1937, N. Stoyko at the Bureau International de l'Heure in Paris put a number on it: the day in January ran about two milliseconds longer than the day in July.

The causes are all real and all still with us. Tidal friction from the Moon, bleeding rotation away over geological time. Seasonal exchange of angular momentum with the atmosphere — winds and weather, pushing back. Coupling between the mantle and the liquid core. The slow rebound of continents still rising after the last ice sheets came off them.

This is the reversal that makes everything after it possible. As long as the sky was the most regular thing available, the sky was the standard by definition and could not be wrong. Build a better clock and the sky acquires an error bar.

documented  quartz clocks → the day has an error

If the Earth's spin is unreliable, use its orbit. The orbit is far steadier. So in 1956 the CIPM adopted, and the 11th CGPM ratified in 1960, the ephemeris second.

The wording is worth reading slowly: the fraction 1/31,556,925.9747 of the tropical year for 1900 January 0 at 12 hours ephemeris time.

1900. The reference year had already ended. It could not be repeated, and it could not be gone back to and re-measured. The definition named a stretch of time that existed once, in the past, and was over.

Nobody was hiding this. It was the honest engineering answer to a real problem, and it bought roughly a hundredfold improvement. But notice what had happened to the relationship: you could no longer check the standard yourself. Checking it now required an observatory, an ephemeris, and years.

documented  rotation unreliable → the orbit substituted

"The second is the duration of 9 192 631 770 periods of the radiation corresponding to the transition between the two hyperfine levels of the ground state of the caesium 133 atom." — Resolution 1, 13th CGPM, 1967

Amended in 1997 to specify a caesium atom at rest, at a temperature of 0 K, because at the new precision the motion and the warmth of the atom had become visible.

Read it again and notice what is not in it. No day. No year. No sky. The definition is now a property of a kind of atom, and every caesium-133 atom in the universe is the same as every other one.

The number itself is not arbitrary: it was chosen to make the new second match the ephemeris second as closely as anyone could then measure. The continuity was deliberate. Nothing jumped.

And it means the second no longer measures the day. It never did again.

documented  the orbit → an atom

It is tempting to read the chain in the other direction — physics gets precise, and society is dragged along. The dates say the opposite.

Before 1883 every town in North America kept its own noon, set by the Sun where it stood. That is not disorder; it is the most literal accuracy there is. Noon means the Sun is overhead here, and it was.

Then railways. A schedule spanning two towns with different noons is not an inconvenience, it is a collision. On 18 November 1883 the railroads of the United States and Canada adopted Standard Railway Time — five zones, one time each. Towns reset their clocks as the new noon arrived, and many of them had two noons that day.

What follows is a ratchet: telegraph time signals, then radio time signals, then navigation, then telecommunications, then satellite positioning — each one wanting time held tighter than the last, and eventually tighter than the Earth could hold it.

documented  coordination → demand for a steadier standard

Standard time is a shared reference. It becomes something else the moment somebody writes it down next to your name.

On 20 November 1888, Willard Le Grand Bundy, a jeweller in Auburn, New York, patented a workman's time recorder. Each worker had a numbered key; turning it in the machine stamped the time against that number. The following year he and his brother Harlow incorporated the Bundy Manufacturing Company — the first time-recording company in the world.

Through a chain of mergers it became the International Time Recording Company, then part of the Computing-Tabulating-Recording Company in 1911, which was renamed International Business Machines in 1924. The company that started by stamping the hour on a card became the company that counted people.

Taylor's stopwatch is a different machine and belongs to a different chain — the time recorder logs arrival, the stopwatch takes a task apart. That one is No. 43, and it is not repeated here.

documented  standard time → the time ledger

Once the second is an atomic quantity and the Earth is not, the two disagree — and the disagreement grows. From 1972 there is a rule for what to do about it.

UTC ticks exact SI seconds. When the difference between it and the turning Earth approaches 0.9 seconds, a whole second is inserted. The clock reads 23:59:60 — a minute of sixty-one seconds, which is a real minute that really happens — and the two are back in step.

TAI − UTC now stands at 37 seconds. Ten of those were the offset it started with in 1972; the other 27 are the announcements.

It is a nuisance. It is also a receipt.

documented  the divergence → a public, dated correction

There has not been a leap second since the last day of 2016. Not because the problem was solved. Because the Earth started spinning faster.

Every leap second so far has been an insertion: the Earth running slow, the clock waiting for it. Since around 2016 the trend has reversed, and recent years have set records for the shortest day since precise measurement began. If it holds, the correction that comes due is a negative leap second — a minute with fifty-nine seconds, 23:59:59 deleted.

That has never been done. Resolution 4 of the 27th CGPM names it directly: "the first negative leap second whose insertion has never been foreseen or tested."

And the same resolution notes that even positive leap seconds are handled by operators using "different methods… which do not follow any agreed standards," and that these uncoordinated methods threaten the resilience of the infrastructure they were meant to protect. The announcement was real; the response to it never got standardised.

documented  faster rotation → an untested correction

Then a geophysicist looked at why the deadline kept moving, and found the answer at the poles.

Duncan Agnew, of the Scripps Institution of Oceanography, published "A global timekeeping problem postponed by global warming" in Nature 628, 333–336 (2024). Melting in Greenland and Antarctica — measured by satellite gravity — moves mass from the poles toward the equatorial ocean. Spread the mass outward and the spin slows. It is the skater putting their arms out, at planetary scale.

The finding: without that melt, the first negative leap second would have been needed in 2026. With it, 2029.

Take the measure of that. You can no longer read a clock without reading the ice. The standard we cut loose from the sky in order to make it independent turns out to be coupled, through the length of the day, to how much of Greenland is still frozen.

documented  polar melt → the deadline moves

In November 2022 the General Conference on Weights and Measures adopted Resolution 4, "On the use and future development of UTC." It decides that the maximum value for the difference UT1 − UTC "will be increased in, or before, 2035."

It asks the CIPM to consult the ITU and others, and to propose a new maximum that ensures the continuity of UTC "for at least a century"; a plan to implement it; a review period; and a draft resolution for the 28th CGPM in 2026.

That meeting is this October. The proposal on the table is a leap hour — one correction, several centuries from now, instead of a leap second every few years.

Between now and then, clock noon and solar noon simply drift apart, and no one is told on any particular date by how much. The information does not vanish — you can always compute it — but it stops arriving.

documented  the patch is costly → widen the tolerance

Everything in Resolution 4 is correct, and this zine is not an argument against it.

Leap seconds really do break things. The patches really are uncoordinated. A negative leap second really has never been tested, and finding out how it goes on live infrastructure is not a good plan. The CGPM published its reasoning, put it to a vote of member states, and built in a review period. This is what an institution looks like when it is working.

The collection this zine belongs to exists to reclaim science from the things that wear its clothes. So let it be said plainly: metrology is one of the most careful undertakings our species runs, and the people redefining the second are not the villains of anything.

A leap second and a leap hour correct the same total drift. They differ in one respect: how often the world is told, on a date, with a number, that the standard does not fit.

Hold that distinction. It is the only thing crossing the next joint.

documented  a good decision, with one cost named

leap  the metrology → how institutions handle us

Say the obvious thing first: caesium licenses nothing about a person. Nobody carried the SI second into a classroom. There is no documented handoff here, and inventing one would be exactly the move this collection was built to refuse.

What crosses is not a fact. It is a question, and it survives the crossing because it was never about time:

Look at what an accommodation actually is. Extra time on an assessment. A deadline moved. A meeting recorded. Each of these is a public, dated, quantified notice that the standard and this body disagree by exactly this much, and here is the correction.

And the objections come back in the same order, almost word for word: it's irregular. It breaks the system. It's hard to schedule. Everybody handles it differently. It doesn't scale.

Every one of those is true of leap seconds too, and every one of them is an argument about the patch that gets heard as an argument about the gap. The proposed remedy is the same in both places: widen the tolerance and stop announcing.

The difference is that when the standards body does it, the Earth is not in the room, and does not have to hear that the mismatch was never worth mentioning.

contested  the score → a property of the person

In 1991 the third edition of the Wechsler Intelligence Scale for Children introduced four index scores, and one of them was the Processing Speed Index. It is still there, in the WISC-V and the WAIS-5.

Two subtests carry it. Coding: a key pairs symbols with digits, and you copy as many as you can before time is called. Symbol Search: you scan a row and mark whether a target appears. Both are timed. Both reward fast switching between a key and a page.

What comes out is a number that gets read as a rate the child has. We think it is closer to a fit: between one person's attention and one particular tempo, on one particular morning, in a room.

This is marked contested rather than documented because the argument is genuinely live and we are in it, not above it. But the direction of the evidence is not neutral. Monotropism — Dinah Murray, Mike Lesser and Wenn Lawson, 2005 — describes attention pulled deeply into fewer channels at once. If that is how a mind is arranged, then the cost of switching is real, and a test scoring the switch has measured the switch.

The index is not a lie. It is a measurement of something narrow, reported in a way that sounds broad — which is No. 44's problem exactly, arriving from the other side.

Nobody says the planet is failing to keep 86,400 seconds. Everyone understands that the second stopped being a piece of the day, and that the gap is a fact about the definition.

Crip time makes the same move, and it was ours before it was anyone's literature — a community term, documented in disability studies by Irving Zola and Carol Gill among the first, neither of whom set out to define it. Alison Kafer gave it the formulation that travels: rather than bend disabled minds and bodies to meet the clock, crip time bends the clock to meet disabled minds and bodies. Ellen Samuels, in "Six Ways of Looking at Crip Time" (2017), keeps the hard edges attached — grief time, broken time, time that is loss and not only liberation.

Crip time is not slower time. It is the recognition that the standard is an artifact chosen for coordination, and that the distance between it and a living body is information about the standard.

Which returns the question the chain has been carrying. The gap is going to exist either way; standards are artifacts and bodies are not. The only decision anyone actually gets is whether it is named out loud, on a date, with a number — or quietly widened until nobody has to mention it.