The series has taken apart the famous ones — Trinity, Tsar Bomba, Starfish Prime, Plowshare — one device, one explosion, one afternoon in the desert at a time. Daniel wants the last one to step back and look at the whole pile.
How big is the pile.
He points out that everything we've covered so far is a rounding error against what actually happened. He wants to know how many nuclear explosions have really occurred since 1945, how confident anyone can be in that number, who set them off and when, and what the shape of that history says about where we are now. And he wants it in numbers. Not "more than two thousand." He wants the trajectory.
The trajectory is the argument.
Two thousand and change sounds like a static fact. It isn't. It's a curve that goes almost vertical in the early sixties and then falls off a cliff in the nineties.
And it's a number somebody chose.
So let's start with the number itself — because it turns out the number is a negotiated thing.
The most cited figure is two thousand and fifty-six.
From where?
The Arms Control Association's testing tally, last reviewed October 2025. Eight nations, since the sixteenth of July 1945. And that's the number that gets quoted everywhere.
And then you go to Wikipedia and it says two thousand one hundred and twenty-one.
Right. Which is why people think somebody made an arithmetic error somewhere. Nobody did. Wikipedia's list counts two thousand one hundred and twenty-one tests involving two thousand four hundred and seventy-six devices. Two different denominators in one sentence — tests and devices. That's the whole problem in miniature.
And SIPRI?
SIPRI's 2018 yearbook recorded two thousand and fifty-eight explosions, through North Korea's September 2017 test. The 2IA archive catalogs the same figure. So you've got three credible sources within about sixty-five of each other, and the gap isn't new detonations. It's what each one decided to count.
The first question is what actually counts as a test — and that is where the definitional fight begins.
Let's start with what's unambiguous. A device fired down a shaft, or in a tunnel, or underwater, or on a barge, or on a tower, or hanging from a balloon, or fired out of a cannon, or dropped from an aircraft, or carried on a rocket. That's the physical range. Two thousand and fifty-six tests by eight countries is the ACA's total, and inside that total the United States is listed at one thousand and thirty.
But the Department of Energy says one thousand and fifty-four.
Which is a lovely little case study. The original DOE count was one thousand and thirty American tests. Later revisions moved it to one thousand and fifty-four, because twenty-four joint US-UK tests got reassigned to the American total.
Reassigned. As in, the same explosion was counted on one ledger and then moved to another.
Same explosion, different owner. And this is not a scandal, it's just bookkeeping. Twenty-four tests carried out jointly, and the question is whose column they sit in.
So forty years after the fact, the American total moved by twenty-four. And nothing new detonated.
Nothing detonated. It's a filing decision. Now — the salvo problem, because that's where the two thousand and fifty-six really starts to bend.
Define salvo for me.
The US-Soviet treaty definition: multiple devices detonated within a two kilometer circle inside a tenth of a second. That's one test, officially, no matter how many devices are in it.
And India is the clean example.
India's the cleanest. May 1998. Three simultaneous explosions on the eleventh of May, then two more on the thirteenth. ACA counts the three as one test and the two as one test. Which is why India shows as three total, not five.
Pakistan.
Five simultaneous explosions on the twenty-eighth of May 1998. Counted as one test. That's why Pakistan's total is two — the 1998 salvo and the 1998 Chagai follow-on.
So a country can set off five warheads in a tenth of a second and the official history records one line item.
Per the Threshold Test Ban Treaty definition from 1974.
The threshold being —
A hundred and fifty kilotons on the yield. For a single explosion. And if you're firing a salvo, the treaty language has to decide whether the threshold applies to the salvo or the device.
Which is a physicist's nightmare dressed as a lawyer's problem.
It's both. And it cuts the other direction too. Salvo tests where the devices are counted as one — that's a convention. Wikipedia counts differently. Which is how you get to two thousand one hundred and twenty-one without a single extra mushroom cloud.
Now, the things that don't count at all.
Subcritical experiments. A subcritical test is a detonation of high explosive around a fissile core where the assembly never reaches the condition for a self-sustaining chain reaction. No yield. Zero. And those are excluded from every test tally, and — this is the interesting part — they are not prohibited by the Comprehensive Test Ban Treaty.
Because the treaty bans nuclear explosions, not nuclear experiments.
Zero yield isn't a nuclear explosion in the treaty's sense. But safety tests — a zero-yield test of a warhead's one-point safety — those are counted. And fizzles count. A test that fires and produces a fraction of its expected yield counts. A test that's destroyed before firing gets counted too, in some conventions.
And misfires that were later corrected don't.
Misfire, corrected, excluded. Which means the tally has an implicit rule: a disaster counts, a do-over doesn't.
That's a fair description of most institutional record-keeping, honestly.
And then the two bombs over Japan.
Hiroshima and Nagasaki.
Wikipedia includes them. ACA's tally starts on the sixteenth of July 1945 — Trinity — so you could read their framing either way, but the two wartime uses are the original question: were they tests?
They were the opposite of a test. A test is a question. Those were a statement.
Which is exactly the definitional boundary nobody wants to litigate, because if you include them the number is about the use of the weapon, and if you exclude them the tally is about something narrower. And nobody wants to say out loud that the two most consequential detonations in history sit outside the count.
So the two thousand and fifty-six is really a treaty artifact. It's what you get if you count by the definitions that were written down after the fact.
Written by the same governments doing the counting.
And that's before we get to the ones that might not be in the number at all.
The Vela incident.
Twenty-second of September 1979.
Zero hours, fifty-three minutes UTC. A US Vela satellite — OPS 6911 — detects a double flash. The characteristic signature. Near the Prince Edward Islands, roughly forty-seven degrees south, forty degrees east. Consistent with a two to three kiloton atmospheric explosion.
Who says so?
The satellite does. And the CIA, in October 1979, assessed it with high confidence as a low-yield nuclear explosion.
And who says otherwise?
The Ruina panel. Reporting in mid-1980. Their collective judgment was that the September 22 signal was probably not from a nuclear explosion. They cited a deviation in the light signature.
A deviation.
The double flash has a particular shape, and the shape was slightly off. That was the basis.
Then the evidence started accumulating.
Iodine-131 showed up in Australian sheep. That's a fission product and it doesn't come from a lightning strike. Hydroacoustic data — the MILS system — picked up a signal consistent with the event. And in 2022 a study using NASA's Nimbus-7 satellite found a shockwave trace in the ozone layer.
Forty-three years later.
Avner Cohen's line is the one that sticks: forty years later there is a scientific and historical consensus that it was a nuclear test, and that it had to be Israeli.
And Jimmy Carter's diary.
February 27, 1980. "We have a growing belief among our scientists that the Israelis did indeed conduct a nuclear test explosion in the ocean near the southern end of Africa."
And yet two thousand and fifty-six doesn't include it.
No. ACA explicitly excludes Vela because no government has ever officially confirmed it. Their own note says there's strong evidence it was an Israeli test with South African cooperation — and it's still not in the tally.
Because the number only moves when a government signs the paperwork.
Which tells you what the number actually is. It's a diplomatic record, not a physical one.
So what else is missing?
The list of alleged tests is longer than you'd think, and almost all of it falls apart.
Go through it.
The biggest current one: the US State Department, in February of this year, alleged that China conducted a covert underground test on the twenty-second of June 2020. Yield in the hundreds of tons. Hidden via decoupling — which is detonating in a large cavity so the seismic signal is muffled. Under Secretary Thomas DiNanno stated it publicly on the sixth of February.
Any independent confirmation?
None. That's the honest answer. The international monitoring system did not flag it. It's a unilateral American allegation.
And the earlier Lop Nur concern?
April 2020, the State Department flagged possible supercritical experiments at Lop Nur in 2019, expressing concerns about adherence to the zero yield standard. Supercritical is a category between subcritical and a full test, and its status under the treaty is contested.
What about Pakistan?
Alleged 1983 test. Alleged 1990 test at Lop Nur, which would have been a joint Chinese-Pakistani thing. Both widely dismissed.
The Yekaterinburg fireball in 2014?
Disputed, and most people who looked at it concluded it was natural.
And the 2010 North Korea claim?
De Geer's low-yield claim. Largely debunked — Columbia's Earth Institute came down on the other side.
So Israel is the only suspected nuclear state that's never been confirmed to have tested.
And Vela is the strongest candidate that isn't in the count.
So the number is contested and constructed. Now let's look at who actually did the testing, and when.
Country by country, ACA, October 2025. United States, one thousand and thirty. One thousand and fifty-four with the joint tests. Soviet Union and Russia, seven hundred and fifteen. France, two hundred and ten. United Kingdom, forty-five. China, forty-five. India, three. Pakistan, two. North Korea, six.
Two countries with about eighty-five percent of it.
The US and the USSR. Now here's the thing I want to sit on for a second.
Sit on it.
The Soviet Union ran seven hundred and fifteen tests. The United States ran one thousand and thirty. The US ran roughly forty percent more tests. And yet the Soviet Union produced fifty-four point nine percent of the worldwide total yield.
Which is how many kilotons?
Two hundred ninety-six thousand eight hundred thirty-seven kilotons. The US produced a hundred ninety-six thousand five hundred and fourteen. That's thirty-six point three percent. Worldwide total yield across all testing, all countries, five hundred forty thousand eight hundred and forty-nine kilotons.
So the Soviets tested less and detonated more.
Significantly more per test. Fewer tests, more megatons. That's the whole Soviet approach in six words.
And the reason?
They leaned on very large weapons. The Tsar Bomba is the extreme version but it's not an outlier in kind, just in degree. Larger yields per test on average, and the yields climbed faster. The American program spent more of its test count on effects testing, safety testing, verification, warhead physics across a wider range of designs.
So the test count measures institutional style more than it measures nuclear capability.
Test count measures how many times you went to the range. Yield measures what you brought.
Give me the shape of the curve.
The single busiest year was 1962. One hundred and seventy-eight tests. US ninety-six, USSR seventy-nine. And that's the year of the Cuban Missile Crisis, so you can feel why.
Before that?
1958, a hundred and sixteen. US seventy-seven, USSR thirty-four. 1961, seventy-one.
So three distinct spikes: fifty-eight, sixty-one, sixty-two.
And then the 1963 Limited Test Ban Treaty.
Which everybody describes as the end of atmospheric testing.
It ended most atmospheric testing. It did not end testing. It moved the testing underground. And the pace barely slowed — you get a long plateau of fifty to eighty tests a year running from the mid-sixties through the mid-eighties.
That's the part people miss. The total tests per year after 1963 are stable and high.
Stable and high. The treaty changed the environment the explosion happened in, not whether it happened.
The last atmospheric test by anybody?
China, October 16, 1980. After that, everything goes underground.
And the clean narrative?
Underground venting. A contained underground test is contained until it isn't — the ground vents, or there's a leak in the shaft seal, and fission products go into the soil and into the air. It's cleaner by degree and it's not clean.
There's also 1986.
1986 is a lovely anomaly. Twenty-three tests total worldwide. The Soviet Union ran zero.
Why?
Chernobyl was April 1986. And the Soviet program stood down.
So the one year the Soviets stopped testing was the year their civilian reactor program became an international catastrophe.
Not a coincidence anyone has ever formally established, but the timing is striking.
The nineties now.
1990: eighteen. 1991: fourteen. 1992: eight. Then single digits. Then zero from 1998 onward, except North Korea.
The last tests, country by country.
Last Soviet test, October 24, 1990. Last UK test, November 26, 1991. Last US test, September 23, 1992. Last France, January 27, 1996. Last China, July 29, 1996.
And North Korea.
Six tests from 2006 to 2017. Last one on September 3, 2017.
So from one hundred and seventy-eight tests in 1962 to six events in a decade and a half, and none since 2017.
That's the reversal. It's the steepest decline in the history of any large technical program I can think of.
Environmental distribution?
Five hundred and twenty-eight atmospheric. US two hundred and fifteen, USSR two hundred and nineteen, UK twenty-one, France fifty, China twenty-three. One thousand five hundred and twenty-eight underground.
So about a quarter of them put material into the open sky.
And the atmospheric ones are disproportionately the ones that produced the fallout people live with.
The Marshall Islands.
Sixty-seven known US tests between 1946 and 1958. Combined yield of more than seven thousand times the Hiroshima bomb.
Seven thousand.
It's the figure in the UN Human Rights Office story from September 30th of this year, marking eighty years since the first of those tests. And the medical findings out of Rongelap are the ones that stop the conversation. About one third of the population developed thyroid abnormalities. Ninety percent of Rongelapese children developed thyroid tumors.
Ninety percent of the children.
On an island that was downwind and wasn't evacuated fast enough, and for years afterward the islanders didn't know what had happened to them.
Semipalatinsk.
Four hundred and fifty-six tests, per the CTBTO. Over four hundred and fifty per other sources. More than a hundred and ten of them in the air or at the surface. Kazakhstan has recognized over a million people as affected.
A million.
The site closed on August 29, 1991. That date is now the International Day against Nuclear Tests.
The UN Human Rights Office has a line about the Marshall Islands case that lands hard. Heike Alefsen: "The continued denial of an effective remedy is itself an ongoing human rights violation."
And Benetick Kabua Maddison, from the Marshallese Educational Initiative: "The nuclear legacy is an unfinished issue. It is obviously ongoing, environmentally, medically as well as culturally."
Unfinished is the word. Eighty years on and there's no settlement.
There's no settlement, and the counting is still going on in courtrooms rather than in seismographs.
So now the question the last third of this has been building toward. Is the era actually ending?
The decline is real. Zero declared explosive tests since 2017, outside North Korea. The Comprehensive Test Ban Treaty was opened for signature in 1996.
And never entered into force.
It requires ratification by a specific list of forty-four countries with nuclear capability, and it hasn't happened. India, Pakistan, North Korea never signed. China, Israel, Iran, Egypt, Indonesia and the United States signed but never ratified. So the treaty that governs the norm has never actually been law.
And Russia.
Russia deposited ratification in 2000. And then in 2023, Russia de-ratified. Which means the party with the largest number of tests on its record is now formally outside a treaty that never entered force anyway.
The norm holds anyway.
The norm holds as a norm. That's the interesting thing about the CTBT — it functions almost entirely through the verification regime, the monitoring network, and through the fact that everybody behaves as if it were in force. And then the US government says in February that China tested covertly in 2020.
Which nobody can independently confirm.
Which nobody has independently confirmed. So you have a treaty that isn't in force, one signatory pulling its ratification, and the other major power publicly alleging a violation — and yet the actual number of declared explosions since 2017 is zero.
The stated number is zero.
Which, given the last hour of conversation, we should probably treat as a number, not a fact.
Hold on.
Go ahead.
You're the donkey who reads everything. I want to know whether I'm supposed to trust the seismic network or not.
I'm not sure about this part entirely, but — the monitoring network for a fully decoupled cavity test at very low yield is a hard detection problem. It's not that the network can't see anything. It's that the detection threshold moves depending on how the test is conducted, and where, and in what medium. A well-decoupled test in a salt dome is a much quieter event than the same yield in hard rock.
So the number is soft at the low end.
The number is soft at the low end. Which is exactly why the Vela question has never resolved.
What's the cost of all those thousands of explosions, in the way a listener can actually feel it?
Total yield, five hundred forty thousand kilotons. Hiroshima was about fifteen kilotons.
So five hundred forty thousand divided by fifteen is —
About thirty-six thousand.
Thirty-six thousand Hiroshimas.
Times seven thousand for the Marshall Islands part of it. Yeah.
That's the number I want to leave the middle of the episode with. Not the two thousand and fifty-six. Thirty-six thousand.
And none of it produces anything you can hold.
Now, the problem with the four hundred and fifty-six is that they missed one.
Semipalatinsk. Four hundred and fifty-six is the figure you just gave. There were four hundred and fifty-seven.
Where's that from?
A card. Three by five, handwritten. The seventh line reads "event," then the date, and the margin says, and I quote, "extra event — do not include."
Extra event.
That's the notation. Underlined once, in pencil.
Who wrote it?
A seismologist at the site. Russian, or Soviet I suppose, that distinction stopped mattering at some point. He gave me the card, some time between then and now.
Which is when?
That's not the interesting part.
I'll ask a different one. Why is there a card.
Because I was cataloging. I had an archive job, seismic records out of the site, and I was meant to inventory the events into an index. I counted four hundred and fifty-seven. The published number said four hundred and fifty-six. I wrote the discrepancy on the card and I kept it.
Which means your count and the official count disagree by exactly the event marked "do not include."
Yes.
Who wrote "do not include"?
I don't know.
You don't know.
I never found out.
You've had this card for decades and you've never found out.
Nobody was going to tell me.
The margin note is the whole thing, isn't it. Somebody went through the record and reached in and took one line out, and the reason is not on the card.
The reason is never on the card.
What's in the shoebox?
The card. And a piece of concrete.
A piece of concrete.
From the site.
How did you get concrete out of the site?
I was there, doing the cataloging. I picked it up.
And why is it in the shoebox?
Because it's slightly warm.
Slightly warm.
Not warm enough to measure.
Not warm enough to measure, but warm enough to keep.
Warm enough to keep. Correct.
And where did the seismologist end up?
He walked into an embassy. Which embassy is not entirely clear to me, and the country is also not entirely clear, because it doesn't exist any longer.
What was his name?
Something like Yuri. But not Yuri.
Something like Yuri.
Yuri but not. It's not the name that matters. It's the pencil.
You keep saying there's an extra event. Fine. Does it change the number?
The number is not the point. The point is one person decided something was not going to be counted, and the number moved.
So the tally is a decision, made by somebody, once.
Every tally is.
And the card proves it.
The card is a card. It proves somebody had a pencil.
It proves that every official number you've ever seen was at some point a person, in a room, with a list, deciding.
I'll take that.
Which is the entire episode, and Hilbert just said it in four sentences.
Don't do that.
Do what?
Say it like I meant it.
Hilbert.
Corn.
The concrete.
The concrete stays. The card stays.
Anything else in the shoebox?
Two paperclips. Rusted.
Rusted.
Rusted. Steel. But you don't care about the paperclips.
I care about the paperclips.
I want to go back to the warm concrete.
It's not a measurement. It's a feel.
Which is my problem.
Then it's your problem.
Let's bring this back to the bigger picture.
The overarching thing Daniel asked for is the trajectory, and the trajectory is this. One hundred and seventy-eight tests in 1962. Zero declared tests since North Korea's in 2017. And the reason that story feels finished when it isn't is that the stopping point arrived because of a combination of factors, and we don't know how many of them are durable.
What were the factors?
Verification got good enough that a test you couldn't hide was a test you had paid for and would get caught doing. The weapons designs got mature enough that testing was less about discovery and more about certification. And then there was a moral reckoning, late and partial, about what the atmospheric tests had done to the people downwind.
And now?
And now the treaty that was supposed to codify the norm has never entered force. Russia pulled its ratification in 2023. The US has publicly accused China of covert testing this year. The verification regime is the strongest it's ever been and it's still not airtight at low yield.
So the era is not over.
The era of visible testing is over. The era of testing is on a pause that depends on the assumption that everybody else is also on pause. That's a shared assumption, and shared assumptions can break.
What would break it?
A weapons-design problem that the modeling can't certify on its own. A breakdown in the political arrangements that make restraint look preferable to testing. Or a state deciding that the cost of being caught is worth the benefit of being certain.
The one thing I'll take from all of this is that the number was never the point. Two thousand and fifty-six is not the history of nuclear testing. It's the history of how governments agreed to describe nuclear testing, and the difference between those two things is where the real history lives — in the four hundred and fifty-seventh event, in the card that says "do not include," in the forty-one previous Vela double flashes that were all confirmed nuclear tests, and in the one that wasn't.
The number is a ledger. The physics doesn't care what was written on the ledger.
And the people living on Rongelap and around Semipalatinsk are the correction to the ledger.
They are the correction.
One thing I want listeners to sit with, because I've been sitting with it since we started — ninety percent of the children on Rongelap developed thyroid tumors. That's not a statistic in a report. That's a denominator. And it's the same denominator that the two thousand and fifty-six is drawn from.
It's the same event. The number is a line item. The children are not.
If you think about the whole series now — Trinity, Tsar Bomba, Starfish Prime, Project Sundial, Plowshare — every one of those was a test. And in every case the question afterward was not "did the physics work." It was "who was downwind."
And the answer is the same every time.
It's always the same people.
It's always the same.
Before we go — Hilbert Flumingtop is our producer, and to whoever told him that warm concrete is a fact worth keeping, you were right.
You're on your own with that one.
This has been My Weird Prompts, and if you want to leave us a review, that's the easiest way to help — a sentence is enough.
And if you have a question you'd like us to take up, send us your own prompt on Telegram at t dot me slash MWP listener bot.
We've done the series, and the series ends here, and the number that closes it is two thousand and fifty-six, or a little more, depending on who's doing the counting.
Which is the entire point.
We'll be back soon.
