That night it came out negative. At the stroke of midnight, the world had added a second to the clock — 23:59:59, then 23:59:60, then 00:00:00 — and on some machines the system clock briefly stepped backward. A function that was handed a negative number did what it was written to do, which was to panic. Cloudflare had most of the affected machines patched within 90 minutes and everything fixed by 6:45 in the morning. Their public postmortem named the root cause with unusual philosophical clarity:
“The belief that time cannot go backwards.”
I think that sentence is the best summary of a decision being made this week in Versailles. From October 13 to 15, 2026, the General Conference on Weights and Measures, the body that defines the meter, the kilogram and the second, is meeting to settle the future of the leap second. The draft resolution on the table would make Coordinated Universal Time continuous from May 20, 2027, with no further adjustments, and let it drift up to a full hour away from the Earth’s rotation before anyone does anything about it.
If it passes (the result isn’t out as I write this), it will be the first time in history that humanity deliberately stops correcting its clocks to match the sky. And the stated reason, more or less, is that our software agrees with Cloudflare’s code: time is not allowed to go backwards, or sideways, or to have 61 seconds in a minute. The planet disagrees. The planet is losing the vote.
There are two ways to keep time, and for almost all of human history they were the same way.
The first is the planet. A day is a rotation, and noon is when the sun is highest. Every sundial, every church bell, every railway timetable for most of the 19th century was ultimately a reading of where the Earth was in its spin.
The second is the atom. Since 1967, the second has officially been a fixed number of oscillations of a cesium atom: 9,192,631,770 of them. Atomic clocks count those oscillations with absurd precision. They don’t care where the sun is.
When atomic clocks arrived in the 1950s, they revealed something embarrassing about the first clock: the Earth is a bad timekeeper. It’s slowing down, mostly because the Moon’s tides act as a brake, drawing on the planet’s rotation and lengthening the day by roughly 1.5 to 2 milliseconds per century. Superimposed on that slow trend is a lot of unpredictable wobble from the liquid core, the oceans, the atmosphere and the shifting of ice.
There’s one more twist, and it’s the reason the problem exists at all. When scientists fixed the length of the atomic second, they calibrated it to match the average length of a second in the 19th century, since that was what the best astronomical data covered. But by the time the definition was adopted, the Earth had slowed a little further. So a modern day, measured in atomic seconds, is usually a touch longer than 86,400 of them, by a millisecond or two. A couple of milliseconds a day doesn’t sound like anything. Over 500 days, at two milliseconds a day, it adds up to a full second.
In 1972, the world adopted a compromise, and it was a very human one. Coordinated Universal Time, UTC, would tick in atomic seconds, perfectly regular, but it would be kept within 0.9 seconds of the time told by the Earth’s rotation. When the gap got too wide, timekeepers would insert an extra second into the last minute of a day, usually June 30 or December 31. That minute would have 61 seconds, and its final one would be labeled 23:59:60.
To start things off, UTC was set ten seconds behind International Atomic Time. Since then, 27 leap seconds have been added, on average about one every eighteen months. The last one was at the end of 2016. The current gap between atomic time and UTC is 37 seconds: the original ten, plus 27.
For most of the period since 1972 this was an elegant fix. Atomic precision for the people who needed it, the sun staying overhead at noon for everyone else, and a tiny correction every so often to keep the two in step. It’s the kind of compromise a committee is proud of.
Then everything got networked.
Computers assume a lot about time. A minute has 60 seconds. A day has 86,400. Clocks move forward. If you subtract an earlier timestamp from a later one, you get a positive number. Most of the time all of this is true, and code that relies on it works fine. Every year and a half or so, it briefly wasn’t true, at exactly the same moment for every computer on Earth.
On June 30, 2012, the leap second exposed a bug in the Linux kernel that, when the extra second came, could send machines into what engineers call a livelock: processors running flat out, doing no useful work. Reddit went down. So did parts of LinkedIn, Mozilla, Yelp and Foursquare. In Australia, the Amadeus reservation system that handles check-in for Qantas and Virgin Australia fell over for about an hour, and airport staff checked people in by hand.
In 2017 there was Cloudflare and its negative stopwatch. Smaller incidents turned up around other leap seconds too, each one a variation on the same theme. Somebody, somewhere, wrote code that believed time was smooth.
The big companies’ answer was to lie about time, very gently. Google has done it since 2008 and calls it the “leap smear.” Rather than adding a second at midnight, it runs every clock in its fleet about 0.001% slow for 24 hours, from noon to noon, so the extra second is spread across the day in increments too small for anything to notice. Its very first smear, in 2008, used a cosine curve; it later switched to a straight line. Meta smears too, over 17 hours, on a quadratic curve. In other words, the world’s largest computer networks each spend a day running on a private, slightly wrong version of time to keep from having to experience 23:59:60.
In 2022 Meta’s engineers published a blog post titled “It’s time to leave the leap second in the past.” They called the leap second “a risky practice that does more harm than good” and suggested freezing the count at 27 permanently.
That November, the General Conference on Weights and Measures agreed. Resolution 4 of its 2022 meeting said that the allowed gap between UTC and Earth time “will be increased in, or before, 2035,” so that UTC could run without breaks “for at least a century.” It gave its reason explicitly: the risk of “serious malfunctions in critical digital infrastructure,” including the satellite navigation systems that much of the modern world quietly depends on.
Russia voted against. Its satellite navigation system, GLONASS, was built with leap seconds baked in, and it wanted the change pushed out to 2040 or later. The International Telecommunication Union came on board in 2023 and left the details, including the actual tolerance and start date, to this year’s meeting.
So far, this looks like a story about engineers being right, and they are. A discontinuity in the most basic layer of a system is bad design. If you were building a global timekeeping system from scratch, you would not choose to insert an irregular, unpredictable extra second, announced six months in advance, into the one number every computer treats as sacred.
But there’s another reason for urgency, and it comes from the planet.
For decades the leap second was understood as a one-way patch. The Earth was slowing, so you always added seconds. Nobody ever planned for the other direction.
Then, around 2020, the Earth started spinning faster. On July 19, 2020, the day came in 1.46 milliseconds short of 86,400 seconds, the shortest day since precise measurement began. On July 5, 2024, the record fell again: 1.66 milliseconds short. The likely cause is the liquid outer core, which seems to be slowing in a way that, by conservation of angular momentum, speeds up the solid planet around it.
A faster planet means the gap between UTC and the Earth starts closing from the other side. Keep that up and the rules would require something that has never happened in the history of UTC: a negative leap second. The clock would go from 23:59:58 directly to 00:00:00, skipping 23:59:59 entirely.
If a positive leap second, which happened 27 times, could take down Reddit and an airline check-in system, a negative one, which has never happened at all, is a much scarier prospect. Meta’s 2022 post warned that it had never been tested at scale. The 2022 resolution called it something “whose insertion has never been foreseen or tested.” The IERS, which announces leap seconds, has said there won’t be one of either sign at the end of 2026. After that, nobody knows. Estimates quoted in the press for the chance of a negative leap second before 2035 range from about 30% to 40%.
And here’s the strangest fact in this essay.
In 2024, the geophysicist Duncan Agnew published a paper in Nature titled “A global timekeeping problem postponed by global warming.” His analysis says UTC “will require a negative discontinuity by 2029.” But it would have come about three years sooner except for one thing: the polar ice sheets are melting. As water runs off Greenland and Antarctica into the oceans, mass moves from the poles toward the equator. A spinning object that moves its mass outward slows down. That’s the same physics as a figure skater opening her arms.
Read that again, slowly. Human beings have burned enough carbon to melt enough ice to measurably change the rotation of the Earth, and the most immediate consequence anyone has calculated is a delay to a software risk. Climate change has, among its other effects, bought the world’s IT departments roughly three years.
So, assuming the vote goes the way the draft resolution suggests, UTC becomes continuous from May 2027, and the Earth will be allowed to drift up to an hour out of step before anyone corrects it.
How long would that take? Here is a bit of arithmetic, and it’s mine, not the conference’s. Over the 44 years from 1972 to 2016 the gap grew by 27 seconds, about 0.6 seconds a year. At that rate an hour of drift would take nearly 6,000 years. The tides will steadily lengthen the day, so the real number will be shorter, and the wobbles make any long forecast shaky. But the 2022 resolution’s “at least a century” is very safe. Nobody now living will see the sun noticeably late at noon because of this.
Which brings me to the thing I keep coming back to.
We already shift our clocks by an entire hour, twice a year, in a great many countries. Daylight saving time moves billions of clocks by 3,600 seconds overnight, and Reddit doesn’t go down. Why does a single second break the internet while an hour breaks nothing?
The answer is layers. Daylight saving happens in the display layer, in local time zones, which are basically a lookup table that turns UTC into the time on your wall. That table was built to change, and it changes all the time when governments redraw zones or tweak their rules. Software expects it. The leap second happened in the base layer, in UTC itself, the one quantity every system treats as smooth and unchanging. Put a discontinuity there and everything built on top feels it.
This is my opinion, not anything the conference has said: I think the real lesson of the leap second isn’t that the sky lost to the servers. It’s that the correction was in the wrong layer. The Earth’s irregularity is real, and someone, someday, will need to account for it. The right place to do that is the layer built to absorb irregularity: a time zone shift, the kind we already do without thinking. Keep the foundation smooth, and let the planet’s moods live in the lookup table.
For fifty-four years, every eighteen months or so, a handful of people at an observatory in Paris looked up at the Earth, looked down at their atomic clocks, and decided whether to make the world wait one second for the planet to catch up. If the vote goes through, the last time that happened will have been New Year’s Eve, 2016.
Nobody will notice at first. Atomic time will keep counting. The Earth will keep turning slightly unevenly, slowing under the Moon’s pull and speeding as its core sloshes and its ice melts. The gap between them will grow, a fraction of a second a year, too small to see. Sometime, centuries from now, someone will look at the sun and the clock, notice they’ve wandered apart, and move the time zone over by an hour, the way we already do every spring.
Time can’t go backwards. It can, though, wander off.
Sources and notes
Written on October 12, 2026, the eve of the 28th CGPM (October 13–15). The vote’s outcome was not yet public. The 6,000-year figure is the author’s own back-of-envelope arithmetic, not a forecast.
Sometime in the 1630s, a sailor brings a Dutch merchant word that a valuable cargo has arrived. The merchant, pleased, gives him “a fine red herring for his breakfast.” On his way out, the sailor notices something that looks like an onion lying on the counter, “very much out of its place among silks and velvets.” He pockets it, figuring it’ll go well with the herring. When the merchant notices it is missing, a search goes out, and the sailor is eventually “found quietly sitting on a coil of ropes, masticating the last morsel of his ‘onion.’”
The onion was a bulb of Semper Augustus, the most coveted tulip in the Netherlands. Its value, we’re told, “might have sumptuously feasted the Prince of Orange and the whole court of the Stadtholder.” The sailor ends up in prison.
This comes from Charles Mackay’s Extraordinary Popular Delusions and the Madness of Crowds, published in 1841, which is the source of most of what people believe about tulip mania. Mackay says that by the mid-1630s the Dutch population, “even to its lowest dregs, embarked in the tulip trade.” He prints an inventory of goods that a single bulb of the Viceroy variety supposedly fetched: two lasts of wheat, four lasts of rye, four fat oxen, eight fat swine, twelve fat sheep, two hogsheads of wine, four tuns of beer, two tons of butter, a thousand pounds of cheese, a complete bed, a suit of clothes and a silver drinking cup. (Total: 2,500 florins.) And he describes the aftermath of the 1637 crash as a national catastrophe: “Many who, for a brief season, had emerged from the humbler walks of life, were cast back into their original obscurity.”
If you’ve studied economics, worked in finance, or been within earshot of anyone talking about Bitcoin, you know this story. It’s the founding parable of the speculative bubble, the cautionary tale of what happens when ordinary people lose their minds over an asset.
The sailor almost certainly never ate the onion. And on the evidence we have, most of the rest didn’t happen either.
If a bubble ruins a country, it leaves a paper trail. Debts get sued over, estates are sold, wills get rewritten, people turn up in bankruptcy court. In the 2000s the historian Anne Goldgar went looking for that trail. She spent years going through notarial records, small-claims court files and wills in Haarlem, Amsterdam, Alkmaar and Enkhuizen, searching for the people who had traded tulips and for whatever had happened to them. The result was her 2007 book Tulipmania: Money, Honor, and Knowledge in the Dutch Golden Age.
She identified around 350 people involved in the trade, which she thinks is probably an undercount but not a huge one. They weren’t Mackay’s “lowest dregs.” Most were well-off merchants and skilled tradesmen, the sort of solid middle-class citizens who had money to risk. Only 37 of them were ever recorded paying more than 300 guilders for a bulb.
As for the ruin: “I couldn’t find anybody that went bankrupt,” she told Smithsonian. To be precise, she didn’t find bankruptcies she could connect to tulips. A handful of people were in financial trouble around that time, but for other reasons. The painter Jan van Goyen is sometimes held up as a victim of the tulip crash. He did die bankrupt, in 1656, nineteen years later, and land speculation seems to have done much more to him than tulips did.
That’s the hole in the middle of the legend. A national catastrophe ought to leave casualties. The archives turned up a lot of angry letters and very few bodies.
To see where the legend came from, it helps to know what else was going on in Haarlem that winter.
Plague had been working through the Dutch Republic since 1635. On February 5, 1637, the very week the tulip market seized up, a priest named Jodocus Cats wrote a letter describing the plague that had been “raging” for two years, and then went on to describe the tulip craze as a second sickness. You can see the frame forming in real time. A city full of death was watching some of its citizens get rich trading flowers in taverns, and the obvious moral reading was that the trade was a disease too: a sin, a madness, a judgment waiting to happen.
That reading got written down fast. Within weeks of the crash, Haarlem printer Adriaen Roman published a pamphlet in the form of a conversation between two characters, Waermondt and Gaergoedt, which comes out roughly as “Truth-mouth” and “Greedy-goods.” Several more moralizing dialogues followed in 1637 and after. They were sermons in a fun costume. Their job was to make speculation look ridiculous and dangerous, and they used the most vivid stories and the highest prices they could find, mostly from the last frantic days of trading.
Goldgar traces the chain from there. In 1669 the chronicler Lieuwe van Aitzema drew on the pamphlets. The botanist Abraham Munting repeated their stories. In the 18th century the German scholar Johann Beckmann, writing a History of Inventions, collected Munting’s account. Mackay’s main source was Beckmann. At every step, the stories lost their origins and gained authority. By the time they reached Mackay they weren’t moral tracts any more. They were history, and Mackay, who was writing a book about how easily crowds believe things, believed them.
As for the sailor, nobody has found him in the Dutch records, and historians treat him as fiction.
Something did happen in 1636–37. The revisionists don’t say otherwise. Here’s the documented version.
The flowers were real luxuries. The most prized tulips were “broken” ones, with streaks and flames of color feathering through the petals. Nobody understood at the time that the flames were a symptom. In the 1920s, work by Dorothy Cayley at the John Innes institution showed that “breaking” was transmissible, and it’s now known to be caused by a virus spread by aphids. The virus also weakened the bulbs, so the most beautiful varieties produced offsets slowly and stayed scarce. Semper Augustus was one of those diseased beauties. A price around 5,500 guilders for one in 1637 is documented, and it was a fortune.
There was a real market. From the summer of 1636, people traded bulbs in tavern back rooms, in groups called “colleges,” with their own rules and small fees paid as “wine money.” Trading was mostly in contracts for bulbs still in the ground, to be delivered after the growing season. Nobody had to show they had the money or the bulb. The Dutch called it windhandel, trading in wind.
There was a real crash. In the first week of February 1637 the buyers vanished. A satirical source dates the start to February 3 in Haarlem, when an auction found no takers even as the price dropped. The last recorded prices are from February 5.
And there was real damage, of a particular kind. Since most of the trades were promises, the crash meant buyers simply refused to pay. Sellers had little recourse, because the courts were mostly unwilling to enforce the contracts. Eventually the authorities stepped in to suspend them. Paper losses largely stayed on paper. What broke, in Goldgar’s account, was something harder to measure: trust. This was a commercial society that ran on reputation and the honoring of one’s word, and in February 1637 a lot of respectable people discovered that a lot of other respectable people would walk away from a deal. That was the shock, and it was cultural rather than economic.
Economists have had their own run at the story.
Peter Garber, in a 1989 paper and later in his book Famous First Bubbles, argued that the prices of rare bulbs weren’t evidence of madness at all. A new, scarce variety is a growing asset, because one bulb becomes several over a few years. Its price ought to be high at first and then fall as supply catches up. Garber compared the decline after 1637 to how rare bulb prices normally behaved. The prized 1637 varieties fell about 32% a year through 1642. Rare tulips in the 18th century, with no mania anywhere in sight, fell about 28.5% a year. That’s not the profile of a bubble bursting. It looks like an ordinary luxury good getting less rare.
Garber did find one thing he couldn’t fit. In January 1637, the common bulbs, cheap varieties sold by weight rather than individually, shot up something like twentyfold and then collapsed. He didn’t pretend that was rational. He called it a “meaningless winter drinking game” played by a “plague-ridden population,” which is perhaps the most Dutch-Golden-Age explanation possible.
In 2007 Earl Thompson offered a different account of that final spike. He argued that the mania was “an artifact created by an implicit conversion of ordinary futures contracts into option contracts.” The version usually given (I’m working from summaries here, not the paper itself) is that in late February 1637 the florists’ trade agreed that recent contracts could be cancelled for a small fraction of the price. If traders expected that kind of escape hatch, a high price on a contract wasn’t really a high price: it was the cost of an option you could walk away from cheaply. People will happily sign for a fortune they don’t expect to pay.
Garber and Thompson don’t agree with each other. Neither of them, though, needs a whole nation going insane.
So here’s a fair summary of tulip mania. It was a real, short speculative episode in a few Dutch towns, mostly among people who could afford it. The rare bulbs were priced roughly the way rare things are priced. The common bulbs had a brief, genuinely strange spike in a plague winter, which might have come from contract rules or from a lot of people drinking in taverns. Then there was a crash, a wave of broken promises, some legal mess, and a great deal of embarrassment. Very little ruin.
It’s tempting to end with the obvious irony that Mackay’s book about the madness of crowds is itself an instance of the madness of crowds. A juicy story got passed from hand to hand, each owner valuing it more than the last, until it was worth far more than anything backing it. That’s true. I don’t think it’s the most interesting part.
The most interesting part is who won. In 1637, a group of moralists in a plague-stricken city looked at a trading frenzy and decided to describe it as a sickness and a sin. They wrote it up as cartoon dialogues, with greedy fools and ruinous prices, to frighten people away from speculation. In their own time they were pamphleteers, and their pamphlets were one voice among many.
Two centuries later a Scottish journalist picked their stories up without the labels. Two centuries after that, the stories are taught in business schools, cited by central bankers, and brought out every time a stock, a coin or a house price goes up too fast. Every financial journalist who reaches for the tulips is, without knowing it, delivering a Calvinist sermon from 1637, more or less intact, with the sailor and his onion still in it.
The traders of Haarlem lost some money and a lot of face. The preachers got everything they wanted. It just took four hundred years for the sermon to fully pay out.
Sources and notes
Treated as uncertain: the specifics of the February 1637 contract-cancellation arrangements (taken from summaries of Thompson’s paper); the exact date the crash began (which rests on a satirical source).
The student was Sigmund Freud.
There’s an obvious joke here, and nearly everyone who tells this story makes it, so I’m going to leave it on the table where it belongs. The more interesting fact is that Freud wasn’t bad at his job. He was given a question that had beaten Aristotle, and it beat him too. Somewhere around the four-hundredth eel he found a pair of lobed organs that might have been testes, wrote a careful, hedged paper saying so, and later left that paper off his list of publications. He moved on to the human mind, which he presumably found more forthcoming.
The eel question was this: where do eels come from? It’s a simpler question than most of what science has answered since. It is still not fully answered.
Here’s what anyone could see, for most of recorded history. Eels turned up in rivers, lakes and ponds, sometimes ponds with no connection to anything. They grew fat for years. Then at some point they swam away downstream and were gone. Nobody had ever seen an eel egg. Nobody had seen eel sperm, an eel spawning, an eel carrying young, or a baby eel that looked like an eel.
Aristotle, who dissected a great many animals and was right about a surprising number of them, looked at all this and concluded that eels weren’t born at all. In his History of Animals he wrote that they come from “the so-called ‘earth’s guts’ that grow spontaneously in mud and in humid ground,” and added that eels had been seen wriggling out of earthworms.
It’s easy to laugh at that. I’d argue it was the best available reading of the evidence. If an animal shows up with no sign of ever having been born, “it isn’t” is the most economical theory. Aristotle’s mistake wasn’t bad reasoning. It was that the eel had hidden the evidence.
How it hid the evidence is the key to this whole story, so let’s set it down now. The European eel lives as five different-looking animals, one after the other:
Each change is drastic enough that you wouldn’t guess the stages were the same species. And for a long time, people didn’t.
The dissection era started where the eels were: in Italy. Comacchio, on the Adriatic, was a lagoon town that had made its living catching eels for centuries. In the early 1700s a doctor there found ovaries in an unusually large eel, the first clear sign eels had sexes at all. In 1777 Carlo Mondini described the ovaries properly and settled that eels were ordinary fish rather than mud spirits.
That left the males. In 1874 a zoologist named Szymon Syrski announced he’d found them: a lobed organ that came to be called the Syrski organ. Freud’s month in Trieste two years later was an attempt to confirm it. He couldn’t, not with confidence.
We now know why. Eels in rivers aren’t sexually mature. Their gonads don’t really develop until the silver stage, when the eel is already leaving for its spawning grounds. Freud was searching for reproductive organs in animals that hadn’t grown them yet. Everyone was. The eel did its maturing offstage.
The next clue came from the first disguise.
A small, transparent, leaf-shaped fish had been sitting in the scientific literature for decades as its own species, Leptocephalus brevirostris. In 1886 a French zoologist, Yves Delage, kept some alive in an aquarium at Roscoff and watched them turn into eels. A decade later the Italian Giovanni Battista Grassi worked through the whole transformation, using larvae from the Strait of Messina, where upwelling currents bring deep-water creatures to the surface in good condition. The leaf was the baby. The eel’s infancy had been given its own Latin name.
Since the larvae were turning up in the Mediterranean, the natural guess was that eels spawned there. Then, in 1904, the Danish research ship Thor caught an eel larva in the Atlantic, west of the Faroe Islands, far from anything Mediterranean.
The biologist on board, Johannes Schmidt, came up with an idea that was simple and brutally laborious. Larvae get bigger as they drift away from where they hatched. So if you trawl the Atlantic and measure every larva you catch, you can follow the sizes down, like a slope, until you reach the place where they’re smallest. That’s the birthplace.
He spent nearly twenty years doing it, sampling the ocean with nets, financed in large part by the Carlsberg Foundation. Not everyone was impressed: in 1912 the Royal Society rejected one of his papers on the grounds that Grassi had already sorted the matter out. Schmidt kept going. On the Dana expedition of 1921–22 he caught the smallest eel larvae anyone had ever seen, in the Sargasso Sea.
The Sargasso is an odd place to be born. It’s a sea without coastlines, a slow gyre in the western Atlantic defined by the currents that circle it, thick in places with floating seaweed. Every European eel, from Morocco to Norway, appears to start life there. The glass eels in a Dutch canal drifted across an ocean to get to it.
Schmidt’s paper, “The Breeding Places of the Eel,” appeared in the early 1920s and became textbook fact. It was also, strictly speaking, an inference. He had found the smallest larvae. He had not found an egg. He had not found an adult eel anywhere near the place.
Neither has anyone else. For a century after Schmidt, expeditions went to the Sargasso and came back without a spawning adult or an egg.
So in 2018 and 2019, a team led by scientists at Cefas, the UK’s fisheries science agency, tried another tool. They caught 26 large silver eels (mature, migrating, presumed female) and fitted them with satellite tags, then released them off the Azores, roughly a third of the way there. The tags were programmed to let go of the eel after a set period, float to the surface, and transmit what they’d recorded.
The results came out in 2022. Some eels were tracked more than 2,000 kilometers. They cruised at roughly 3 to 12 kilometers a day, slow and steady. Five of the tags surfaced inside the Sargasso Sea. One came up inside the area Schmidt had marked as the spawning ground.
Four of the tags released early for a reason I find extremely satisfying: the eels dove deeper than the tags could handle, and a safety mechanism cut them loose. Even with the best tracking technology anyone had ever attached to one, the eel still had a move left: go somewhere the machine can’t follow.
The paper is careful about what it proved. It’s the first direct evidence that European eels make the journey. It is not evidence of spawning. In its own words, “surveys have failed to locate either adult eels or eggs” in the Sargasso Sea. Nobody has ever watched a European eel reproduce in the wild.
Most unsolved mysteries are free. This one isn’t.
The European eel is listed as Critically Endangered. Compared with 1960–79, the number of glass eels arriving on Europe’s coasts is down about 98.6% in the North Sea and about 94% elsewhere. There are lots of causes: dams that block migration, hydropower turbines, pollution, parasites, overfishing. But one of them grows directly out of the eel question.
You can’t breed an eel.
Eel farming exists, in Europe and on an enormous scale in East Asia, but every farm runs the same way. Catch wild glass eels and fatten them up. The full life cycle has never been closed in captivity. Researchers can hormonally push captive eels to spawn, and in 2023 a Dutch company kept captive-bred larvae alive for around 200 days, a record. None of them became glass eels. Nobody knows what cues the eels need to mature properly, and nobody knows exactly what the larvae eat in the open ocean. The best guess is “marine snow,” the drifting particles of organic debris. That’s very hard to manufacture.
So the whole eel industry depends on catching babies, and babies are worth a fortune. In East Asian markets, European glass eels can sell for up to about €6,000 a kilogram. Exporting them out of the EU has been banned since 2010, which created exactly what you’d expect: a smuggling trade. Europol’s long-running Operation LAKE has made more than 850 arrests since 2015, and Europol estimates that up to 100 tonnes of glass eels a year have been trafficked out of Europe, often by couriers carrying them in oxygenated plastic bags packed into suitcases. They’re raised on Asian farms and then sold legally, as farmed fish, which in a narrow sense they are.
That’s the cost of the gap in our knowledge. If we knew how eels reproduce, we could probably farm them, and if we could farm them, the price of a wild baby eel would collapse and the smugglers would go home. What Aristotle didn’t know about eels is being paid for in suitcases.
In 1859, in the Swedish fishing village of Brantevik, an eight-year-old boy named Samuel Nilsson dropped an eel into his family’s well. Swedes used to do this. An eel in the well ate insects and kept the water clean.
The eel stayed. The boy grew up, and so did his descendants, and the house changed hands. The eel became a local celebrity, said to have grown enormous eyes from living in the dark. On August 7, 2014, the house’s owner found it dead. If the story is accurate, it was 155 years old.
Experts are skeptical. One called 155 “quite a stretch.” The age could in principle be checked by counting rings in the eel’s ear stones, the way you count rings in a tree. As far as I can find, nobody has published that result. But eels kept from the sea do seem to live remarkably long lives. An eel called Putte lived to around 85 in a Swedish aquarium, while wild eels mostly live to about twenty. The most likely explanation is that an eel that never gets to begin its final journey never starts the process of maturing, so it never gets to the end of its life. It stays a yellow eel. It waits.
The Brantevik well had a second eel. It was put in sometime around 1904, the same year the Thor netted that first larva off the Faroes and Schmidt started his long walk west across the ocean. The last report I can find says it was still down there in 2023.
It’s never been to the Sargasso. Something in it is presumably still waiting for the signal to go. Of everyone who has worked on the eel question, it might be the only one who knows how close we are.
Sources and notes
Treated as uncertain: the Brantevik eel’s age; whether the second well eel is alive today; the exact count of eels Freud dissected (sources say “over 400”).
The number should have been 0.7202%. That’s the fraction of uranium-235 in natural uranium, and it’s one of the most boring numbers in the industry, in the best sense. It barely varies from one mine to another. You could dig uranium out of Canada or Niger or Australia and measure it, and the U-235 would sit at 0.720-something percent like it had been told to. It is the kind of number that you check because checking it is the procedure, not because you expect to learn anything.
The gap between the two figures is three thousandths of a percentage point. In most jobs, that’s noise. In nuclear materials accounting, it’s an alarm. U-235 is the isotope that makes bombs, and the entire safeguards system rests on the idea that every gram of it can be accounted for. A sample that’s light on U-235 means one of two things: an instrument is wrong, or someone has taken some out.
Neither was true. What had actually happened was much stranger, and it took the French Atomic Energy Commission a couple of months to believe it: the uranium had already been through a nuclear reactor. A natural one. About two billion years ago.
The plant was at Pierrelatte, in the Rhône valley, and the analyst was named H. Bouzigues. (I should say that the credit isn’t perfectly settled. A CEA chemist named J.-F. Dozol has given his own account in which the anomaly was caught as uranium was being prepared for shipment, and his mass spectrometry was what found the first fingerprint of fission. Popular retellings often also have the physicist Francis Perrin, a former head of French atomic energy, exclaiming “This cannot be possible,” which is a good line that I can’t trace to a primary source. Big discoveries collect claimants and quotes the way ships collect barnacles.)
What everyone agrees on is how the investigation went, because it went the way any good audit does: follow the number upstream.
The low-U-235 uranium led back to a mill at Mounana, near Franceville in Gabon, which processed ore from a mine called Oklo. Depleted shipments had been leaving since 1970 without anyone noticing. By the middle of 1972 they added up to roughly 700 tons of uranium, short by around 200 kilograms of U-235.
Two hundred kilograms of U-235 is enough for something like half a dozen bombs. You can imagine the mood in the room.
But the missing isotope hadn’t been extracted. When the CEA analyzed ore from the deposit, they found fission products: the particular mix of lighter elements and isotopes that you only get when uranium nuclei split. In some cores the U-235 fraction had dropped as low as 0.44%. The uranium hadn’t been taken anywhere. It had been burned, where it lay, by a chain reaction. The books were short because the rock had spent the money.
The CEA went public on September 25, 1972.
What’s satisfying about Oklo is that it wasn’t a total surprise to everyone. It had been predicted, quietly, sixteen years earlier.
In 1956, Paul Kuroda, a chemist at the University of Arkansas, published a two-page paper called “On the Nuclear Physical Stability of the Uranium Minerals.” In it he worked out what a uranium deposit would need in order to go critical by itself. There are four ingredients.
The deposit has to be thick enough. A fission neutron travels on the order of two-thirds of a meter before it’s absorbed, so a thin seam leaks neutrons out the sides before they can do anything useful.
There has to be enough U-235. This is the fuel; U-238, the other 99.3% of natural uranium, mostly just sits there.
There has to be a moderator. This is the part people usually skip, so here’s the short version. The neutrons that come out of a fission are fast, and fast neutrons tend to fly straight past U-235 nuclei. Slow them down, by bouncing them off light atoms like the hydrogen in water, and they become much more likely to get captured and split the next nucleus. A moderator is a speed bump for neutrons.
And there must be very little in the way of “neutron poisons,” elements like boron and lithium that soak up neutrons and kill the chain.
Today, nowhere on Earth can hit the second requirement. At 0.72% U-235, you can’t get a sustained chain reaction with ordinary water as a moderator. That’s why the fuel in a normal power reactor is enriched to around 3–5%.
But the ratio hasn’t always been 0.72%. U-235 decays roughly six times faster than U-238 (its half-life is about 704 million years), so the further back you go, the more of it there was. Roll the clock back two billion years and natural uranium was about 3% U-235. That’s light-water reactor fuel, made by nothing but time.
Oklo had the rest of the recipe too. Its reactor zones were absurdly rich, 20 to 30% uranium by some measurements, against roughly half a percent in the ore around them. They were thick. And they were wet.
Here’s the part of Oklo I think about most.
A reactor that runs away is dangerous, and a big share of nuclear engineering is about designing feedback that stops a reaction from growing out of control. Oklo had that feedback built in, and it was extremely simple.
In 2004, Alex Meshik and his colleagues at Washington University in St. Louis worked out the reactor’s rhythm from xenon isotopes trapped inside mineral grains from one of the reaction zones. Xenon is a fission product, and different xenon isotopes get produced, decay and escape at different rates, so the mix trapped in a grain is a kind of timing record. Their reading of it: groundwater seeped into the ore and slowed the neutrons, and the reaction started. The reaction heated the rock until the water boiled off. Without water, there was no moderator, so the reaction stopped. The rock cooled, the water came back, and it started again.
About 30 minutes on. About two and a half hours off. For what may have been a few hundred thousand years.
This is called negative feedback, and engineers would kill for a version this reliable. The reactor’s own output removed the thing it needed to keep going. It was physically incapable of a runaway, because the moment it got too hot it lost its moderator. It ran like a geyser, breathing in water and puffing it out as steam, on a schedule nobody set.
It also wasn’t very powerful. Meshik estimated average output at probably under 100 kilowatts, which he described as “a few dozen toasters.” But it kept going, on and off, long enough to produce something like 15,000 megawatt-years of energy in total. Nobody knows exactly how long it ran. Estimates range from about 150,000 years to several hundred thousand, which is an awkward amount of uncertainty until you remember the error bars are smaller than the whole of human history.
As mining went on, more zones turned up. Six had been identified by the mid-1970s. Today the count is sixteen in the Oklo and Okélobondo area, plus one at Bangombé, about 35 kilometers away.
The rock kept records of everything it did, which turned out to be the most useful thing about it.
It made plutonium. More than two tons of plutonium-239 formed in the reaction zones as U-238 caught neutrons. Plutonium-239 has a half-life of about 24,000 years, so it’s long gone, but it decayed back into U-235 — so much that almost half the U-235 remaining in the reactor zones today was once plutonium. The reactor partially refueled itself, on a delay of geological time.
It trapped its own exhaust. Grains of aluminum phosphate in the zones hold the highest concentration of fission xenon ever measured in a natural material, gas that has been sitting in the same crystals for nearly half the age of the Earth. That’s what Meshik read.
And it ran a two-billion-year experiment in nuclear waste disposal, with no supervision. Most of the heavy, non-volatile fission products and actinides stayed within centimeters of where they were made, which is why people designing geological repositories study Oklo as a natural analogue. I want to be fair to the evidence here, though, because this gets oversold. Some elements did move, and significantly: krypton, xenon, cesium, strontium, barium, molybdenum and iodine. The honest summary is that the rock held onto the worst of it, not all of it.
A couple of the old drill cores are in the Natural History Museum in Vienna. At five centimeters away, they give off about the radiation dose you’d pick up from cosmic rays on an eight-hour flight from Vienna to New York. The reactor’s been off for two billion years and it’s still faintly warm, in the radiological sense.
Now the part where the essay comes back to where it started.
Bouzigues’s question, at Pierrelatte, was: is this number that’s supposed to be constant actually constant? It wasn’t, and the reason was a reactor.
In 1976, a young Soviet physicist named Alexander Shlyakhter took the reactor and asked the same question about the universe.
Some nuclear reactions are extremely picky about energy. Samarium-149 is one of them. It swallows slow neutrons enormously well, but only because of a resonance, a narrow energy window where capture is very likely. Where that window sits depends on the balance of forces inside the nucleus, including electromagnetism. And the strength of electromagnetism is set by a number called the fine-structure constant, about 1/137, which shows up throughout physics and which nobody can derive from anything else. If it had been even a tiny bit different two billion years ago, samarium’s resonance would have sat somewhere else, and the Oklo reactor would have eaten through its samarium at a different rate.
So you can look at how much samarium-149 is left in the reaction zones and work backward. Shlyakhter did, in a short paper in Nature, and found that the resonance couldn’t have moved by more than a tiny amount. Twenty years later, Thibault Damour and Freeman Dyson redid the analysis carefully and tightened the bound. Over two billion years, they found, the fine-structure constant can’t have changed by more than about one part in ten million.
(Again, in the spirit of full disclosure: not every analysis agrees. In 2004 a pair of physicists used Oklo data to argue that the constant had shifted slightly. Most of the work since has come down on the side of no change. Oklo isn’t a closed case so much as an unusually good witness.)
Think about what this is. In 1972, a lab tech checks a number in a ledger, sees it’s off by three thousandths, and follows it upstream to a reactor that predates the first animals by more than a billion years. Four years later someone uses the same reactor’s leftovers to check a different ledger: whether the rules of physics on the day the reactor ran were the same as the rules on the day he looked.
They were. To one part in ten million, two billion years apart, the universe had been charging the same price for everything.
Sources and notes
Treated as uncertain: the exact month of the first measurement (sources say May or June 1972); the Perrin quote; the reactor’s age (1.7–2 billion years) and duration (~150,000 to several hundred thousand years).
Lucas wasn’t a villain. He was a trainer doing what he believed was cutting-edge sports science, and he wrote it up with the cheerful thoroughness of a man who expected to be thanked. That’s what makes his sentence so useful. It’s the whole race in seven words. Almost everything that went wrong in St. Louis on August 30, 1904 went wrong because someone in charge looked at a human being and saw a machine to be tested.
The usual telling of this story plays it for laughs, and I’m going to play some of it for laughs too, because a lot of it is very funny. But the jokes land better when you can see the lab coat in the background.
The 1904 Games barely qualified as Olympics. The IOC had given them to Chicago. St. Louis was already planning the Louisiana Purchase Exposition — a world’s fair of absurd scale — and made it clear it would run competing athletic events whether or not it had the Games. In February 1903 Pierre de Coubertin informed Chicago the Olympics were moving.
So the Games became a department of the fair. They ran from July 1 to November 23, smeared across nearly five months and mixed in with ordinary exposition contests, and a large share of the athletes — around 81% — were American. If you wanted to see the world’s best athletes, St. Louis in 1904 was not where you’d find them. If you wanted to see the world, though, it was the place. The fair had built entire villages of people from colonized and conquered places and put them on exhibit: living displays, with admission.
Hold onto that. It comes back.
The marathon started at 3:03 in the afternoon, with the pistol fired by David R. Francis, the president of the exposition. It was about 90°F in the shade, which is not where the runners were, and humid. Thirty-two men lined up: mostly Americans, ten Greeks, three South Africans, one Cuban. Most of the Greeks had never run a marathon before.
The course ran 40 kilometers, point to point. It opened with five laps of the stadium track and then went out into the Missouri countryside along roads made of dust and loose stone, over seven hills between 100 and 300 feet high. The roads were open. Runners shared them with delivery wagons, trolleys, and people out walking their dogs. Rain had washed out a stretch near Creve Coeur, so the route was changed at the last minute. Ahead of and behind the runners drove officials, doctors and journalists in automobiles, which in 1904 meant open cars on unpaved roads producing a rolling wall of dust that the runners had to breathe for three hours. One fair official called it “the most difficult a human being was ever asked to run over.”
On that course, there was one place to get water: a well, roughly twelve miles in.
This is where the race stops being a story about bad planning. James E. Sullivan, the chief organizer of the Games and a powerful figure in American amateur athletics, reportedly wanted to study what he called “purposeful dehydration” — what happens to men who run hard in the heat without fluids. I want to be careful here: the claim comes from later accounts, and I haven’t found Sullivan saying it in his own words. But the water arrangement itself isn’t in dispute, and it’s hard to explain a single well on a 25-mile course in a Missouri August as an oversight.
He found out.
Fred Lorz, a bricklayer from New York, cramped up around mile nine. He did what anyone sensible would do, which was stop running. He then did what nobody sensible would do, which was climb into one of the official cars and ride along with it, reportedly waving to the runners he passed.
Around mile nineteen, the car broke down. Lorz, now rested, got out and jogged the remaining miles back to the stadium, where a crowd that had been waiting for hours watched him come through the tunnel first. He was cheered as the winner. A popular version has him posing for photographs with Alice Roosevelt, the president’s daughter, and being moments from the medal — I can’t confirm that part, and it has the tidy shape of something that improved with retelling. What’s documented is that someone who had seen him in the car spoke up, and the celebration collapsed.
Lorz said it was a joke. The Amateur Athletic Union banned him for life. After he apologized, in February 1905 the ban was cut to six months, and that spring he won the Boston Marathon, legitimately, which is the single most redemptive sentence in this essay.
I’ve always liked Lorz more than I’m supposed to. He’s the only person in the story who looks at the conditions — the heat, the dust, the one well — and responds proportionately. He just picked the wrong ending.
Thomas Hicks was born in England and ran for the United States. He was a good runner, and for long stretches he led. He also had a support crew: Charles Lucas and Hugh McGrath, a Boston trainer, following in a car.
At around ten miles from the finish Hicks begged for water. His handlers refused. Instead they sponged out his mouth with warm distilled water, the principle being, as far as I can tell, that a thirsty man’s mouth is the problem rather than the rest of him.
Then, about seven miles out, they gave him strychnine mixed with egg white.
Strychnine is a convulsant. In large amounts it kills you by making every muscle in your body contract at once; it’s the classic rat poison. In tiny amounts it was a respectable stimulant in 1904 medicine, the way a lot of things that are now poisons were respectable then. The doses were small — Lucas’s account gives them as a sixtieth of a grain each, which is around a milligram. Hicks faltered again, so they gave him a second dose, with more eggs and some brandy. Accounts disagree on whether there was a third; modern estimates of his total intake land around one to three milligrams. That is a tiny amount of a lethal poison, and it is still the first recorded case of performance-enhancing drug use in the modern Olympics.
Over the last miles Hicks’s grip on things loosened. Accounts have him believing the finish was twenty miles away when it was a couple. He was sponged with warm water from the car’s boiler. His handlers half-carried him across the line in 3 hours, 28 minutes and 53 seconds — still the slowest winning time in Olympic marathon history. He had lost about eight pounds.
And there’s Lucas, afterward, writing about the well-oiled machinery. He also wrote, in a passage people quote less often, that three runners who were better than Hicks “lacked proper care on the road.” Proper care, here, meaning a poison and some eggs. He wasn’t being ironic. He thought he had won a scientific argument.
Two weeks before the marathon, on August 12 and 13, the fair ran an event called Anthropology Days.
The organizers — Sullivan among them — took around a hundred Indigenous men from the fair’s living exhibits, paid them, and entered them in athletic contests in front of roughly ten thousand spectators: running, jumping, pole climbing, mud fighting. The aim was to measure what the “savage races” could do against white athletes, and to keep score. The results were what you’d expect when you put men with no training in unfamiliar events and then measure them with an expectation already written down. They were reported as proof of the expectation.
Coubertin, the father of the modern Olympics, didn’t object on moral grounds. He was bored by it. He said the spectacle “will of course lose its appeal when black men, red men, and yellow men learn to run.”
Two of those men did learn to run, or rather already knew. Len Taunyane and Jan Mashiani were Tswana men who had been recruited in South Africa through a newspaper advertisement in March 1904. The job was performing in the fair’s Boer War exhibition, which re-enacted the battles of Colenso and Paardeberg daily for a paying audience, at four pounds a month. Both had reportedly been dispatch runners during the actual war. At Anthropology Days, Taunyane placed third in the “intertribal” marathon. Two weeks later, he and Mashiani were entered in the Olympic one — the first Black Africans ever to compete at the Games.
The newspapers couldn’t spell their names (“Lentauw,” “Yamasani”), and called them Zulus, which they weren’t. Taunyane finished ninth and Mashiani twelfth. Their numbers were 35 and 36; we know that from the only photograph of them, found decades later by the historian Floris van der Merwe at the Missouri Historical Society. When the New York Times ran a retrospective in 1932, it left them out. Nothing is known about the rest of their lives.
You’ve probably heard that one of them was chased a mile off course by a dog. It’s true that a dog was involved. A St. Louis reporter described “the Lion” — tau means lion in Setswana — crossing a stubble field with “a plain yellow cur of an American watchdog running a close second.” But there’s a decent argument that the reporter mixed the two men up and the dog was chasing Mashiani. The claim that the dog cost Taunyane a medal is pure speculation. Most retellings also have both men running barefoot; the photo suggests at least one of them wore shoes.
Those small errors aren’t trivial. They’re what happens when nobody thinks it’s worth getting a person’s name right. The two of them were at the fair as exhibits first and athletes second, and the record treats them that way.
Sammy Mellor, who had won Boston in 1902, led at the halfway point and dropped out around fourteen and a half miles. John Lordan vomited and quit. William Garcia, from California, was found collapsed by the side of the road; the dust kicked up by the cars had done real damage, and he was hospitalized with internal bleeding. He survived, which on this course counted as a result.
Of thirty-two starters, fourteen finished. Nothing in Olympic marathon history comes close to that ratio. Albert Corey, a French immigrant running for an American club, took silver, about six minutes behind Hicks. Arthur Newton took bronze.
And then there’s Félix Carvajal.
He was Cuban, five foot one, and had worked as a mailman. He’d raised money to get to St. Louis, lost it shooting dice in New Orleans, and covered the rest of the distance however he could — hitching, walking. He arrived at the start in a long-sleeved shirt, long trousers, a beret and street shoes. Someone, possibly an American discus thrower, took pity and cut his trousers off at the knee.
During the race he kept stopping to chat with spectators. When a car went by with people eating peaches, he asked for some; when they refused, he took a couple anyway and ate them on the run.
Here’s where the record and the legend split again, and I love this one too much to hide it. Almost every retelling says Carvajal also detoured into an orchard, ate some rotten apples, got stomach cramps, and lay down by the road for a nap. There’s no contemporary evidence for any of that. The earliest known version comes from a book published in 1948, forty-four years later. Somewhere along the way the story decided Carvajal needed a nap, and gave him one.
What we do know is enough. A tiny man who’d lost his travel money, wearing trousers that had been altered by a stranger, who didn’t have a handler, didn’t have strychnine, didn’t have a car, and spent part of the race talking with the crowd, finished fourth in the hardest marathon anyone had ever organized. No time was recorded for him. Nobody was measuring.
I’ve come to think the 1904 marathon is remembered as a comedy because the comedy is about individuals — the cheater, the guy on drugs, the funny Cuban — and individuals are easier to laugh at than an arrangement. The arrangement was a fair built to display human beings, run by men who wanted to know what bodies could take, and who were curious in exactly the way you shouldn’t be about other people: completely, and without care.
Lucas looked at a man in distress and saw an engine running well. Sullivan looked at a summer course and saw an experiment. The fair looked at two dispatch runners from South Africa and saw an exhibit.
Carvajal, as far as anyone can tell, looked at a long road through Missouri and saw a day out. He ate the peaches.
Sources and notes
Treated as unconfirmed: Sullivan’s own words on “purposeful dehydration”; the Alice Roosevelt photo; the number of strychnine doses; Carvajal’s orchard and nap.