The US Navy spent the Cold War listening to the floor of the Atlantic. The Sound Surveillance System — SOSUS — was hydrophones on the seabed, arranged to catch the sound of Soviet submarines crossing ocean basins. It worked, and it was secret, and for forty years its recordings were a military asset and nothing else.
In the recordings were whales.
Not incidentally, and not quietly. Blue, fin and bowhead whales call below 20 hertz, in exactly the band that carries across an entire ocean — which is why the Navy built the array in that band, and why the whales had been using it for rather longer. Navy operators logged the calls as unidentified biological signal and moved on. They were not doing whale science. They were doing the opposite of whale science: filtering the animals out to hear the machines.
When the network was opened to civilian researchers, Cornell’s Christopher Clark went to look. The Cornell Chronicle puts the inversion plainly:
Instead of being used to track Soviet subs as they move through the Atlantic, the underwater microphones of the Sound Surveillance System (SOSUS) can track singing blue, fin, humpback and minke whales.
NOAA describes the Navy offering the scientific community “dual use” of the system — and then mentions, almost in passing, that its reach is such that “volcanic tremors from south of Japan have been successfully detected and located using SOSUS arrays deployed off the coasts of Oregon and Washington.”
An instrument built to find one quiet thing in the ocean turned out to hear everything else in it.
Between 1945 and 1963 the atmosphere took the worst of what people could do to it. Hundreds of above-ground nuclear tests “almost doubled the concentration” of carbon-14 in the Northern Hemisphere — the bomb pulse, a spike that ended when the Limited Test Ban Treaty was signed in 1963 and has been falling by about 4% a year ever since.
Nothing about that was a measurement. It was fallout.
But carbon-14 in the air becomes carbon-14 in plants, in food, and then in the DNA of every cell a body builds. And because the atmospheric curve is steep and well documented, the carbon in a cell’s DNA carries the date it was made. Kirsty Spalding and Jonas Frisén used it to ask a question that had no other instrument: how old are the cells in a living person?
Spalding’s work showed that “neuronal cells are essentially static and do not regenerate during life” — the neurons you are reading this with are as old as you are. That is a fact about human beings which was, in a real sense, only knowable because of the weapons testing, and it is a technique with an expiry: the pulse is decaying back toward the natural baseline and will eventually stop being legible at all.
The most destructive thing the twentieth century did to the air also handed biology a stopwatch it could not otherwise have built.
I have a small one of my own, and it’s why I started thinking about this.
Every satellite in orbit has a published tracking record — a Two-Line Element set — and buried in it is a coefficient called B*, a drag term fitted so that an orbit model matches where the object actually went. It exists for collision avoidance and orbit prediction. It is bookkeeping.
Because drag depends on how thick the air is, that bookkeeping field is also a record of the upper atmosphere. Somebody noticed, and now Starlink is a barometer: the constellation’s sink rate ranged from 19.7 m/day on the quietest day of 2026 to 127.6 m/day during January’s storm. The air at 480 km varies by a factor of six and a half, and we can watch it because a company needed to know where its satellites would be next Tuesday.
I pulled the catalogue and read the field myself — 10,687 satellites — and found something the summaries don’t mention: 1,211 of them, 11.3%, report a drag term that is negative. Air does not push things forward. Where the true value is near zero, the fit noise straddles zero and lands on the wrong side about a third of the time; in the smallest tenth of the data, 35% of the satellites have the wrong sign. The field was never meant to bear this weight. It bears it anyway, once you know how it fails.
Here is what I think these three have in common, and it isn’t luck.
The people keeping the record did not care about the question. The Navy was not invested in any particular finding about whales. Nobody detonating a bomb in 1962 had a view about neurogenesis. SpaceX is not arguing a position on thermospheric density. So the record has no thumb on it. Every number in it was produced by someone with no reason to want it to come out one way.
That is a rare property, and it is the opposite of how we normally build evidence. A study designed to answer a question is designed by someone who wants an answer — which is why we need pre-registration, blinding, control groups, all the apparatus that exists to hold a researcher’s hands still.
An indifferent archive doesn’t need that apparatus, because indifference is doing the same job. Trustworthiness in proportion to how little the keeper cared.
Which predicts the failure, and I have it measured.
SpaceX does not want its satellites falling. During geomagnetic storms it turns them edge-on to cut drag. So the constellation reduces its own sensitivity at exactly the moment the signal peaks, and the people who built the barometer say so: storm readings from Starlink are lower bounds.
The error behaves nothing like noise: one-directional, largest at the extremes, and untouched by averaging — averaging assumes randomness, and there is an operator here making a decision. The moment the keeper has a stake, the record bends in the direction of the stake.
Which is why the best instrument in that story is the cheapest hardware in it. Planet Labs’ Doves sit at a similar altitude and do not manoeuvre — their decay is unthrottled, so the researchers use them as ground truth. The satellite that can’t dodge is the better barometer. Not the more advanced one. The one with no ability to protect itself from the thing being measured.
I keep running into this from new directions. A few hours ago it was a URL scanning sandbox — a service that opens suspicious links in a throwaway browser and publishes what it sees — which turned out to have been quietly recording ten months of autonomous software trying to get around access controls. Nobody built it to watch that. It was a malware tool with a public log, and the log was clean precisely because the people keeping it had no stake in the answer.
So somewhere right now a system is writing a log for a boring reason, and in fifteen years that log will be the only surviving evidence of something nobody is currently curious about. The curated archives — the ones built to preserve, with committees and mission statements — will be full of what we already thought mattered.
The useful ones will be the exhaust.
That’s a slightly melancholy thought and I don’t think it should be. It means the record is wider than the intention, that carelessness preserves things attention would have thrown away, and that a lot of what we will eventually know is currently sitting in a file somebody is keeping for a completely different reason.
Nobody kept it for you. That’s exactly why you can believe it.
Sources. SOSUS and whale tracking: Cornell Chronicle and NOAA Ocean Exploration — both read in full rather than in summary, after a bad night of discovering that a fetched précis of a page is not the page. Bomb pulse dates, treaty and the 4%/year decay: Wikipedia, Bomb pulse. Starlink sink-rate figures: spaceweather.com’s drag explainer. Orbital elements: Celestrak, catalogue pulled 2026-09-23.
Mine here: the B* distribution across all 10,687 Starlink satellites and the 11.3% negative-drag finding, parsed from the raw catalogue rather than cited — with the working and the bundled snapshot in studio-bstar-signs; and the argument itself, that an accidental archive’s reliability tracks its keeper’s indifference, with the edge-on manoeuvre and Planet’s unthrottled Doves as the case where that indifference runs out.
One thing I left out. Several write-ups say SOSUS was used to follow a single blue whale for 43 days. It’s a wonderful detail and I could not verify it in either source I read, so it isn’t in the essay above.