For All Sufficiently Large Primes
A reprint desk put "this time by humans" over a mathematics story that never mentions machines. The headline is true — and the one fact in it a reader can actually check is the one that isn't about the mathematics.
On 28 September, Quanta Magazine published a story by Shalma Wegsman under this headline: "Mathematicians Harness Randomness To Crack a 55-Year-Old Conjecture." The deck beneath it read: "After a long hiatus, a problem that was likely inspired by juggling has finally been resolved by a group of young mathematicians."
Forty-one days later, on 8 November, WIRED ran the same story. Same byline, same text, credited at the top as an original that "appeared in Quanta Magazine." Different headline: "A 55-Year-Old Math Problem Was Solved—This Time by Humans." Different deck: "A group of young mathematicians found the solution to the long unanswered conjecture, first inspired by juggling, through randomness and friendship."
The body did not change. I compared them. Nowhere in it does the word "machine" appear. Nowhere does it mention artificial intelligence or machine learning. The word "computer" appears twice, both times describing people who like juggling: Graham's remark in a 1980 interview that "many mathematicians and computer scientists have an interest in juggling," and Noah Kravitz's aside that "computer scientists often call this a 'finding the hay in the haystack' problem."
That is every trace of the machine in the story. An adjective inside a quote.
I want to be careful here, because the easy version of this piece is to catch a copy desk, and the easy version is wrong. "This time by humans" is a true sentence. The people who solved the problem were people. The article says so because it could not say anything else. Nothing was fabricated, nothing was wrenched from context, the names are right, and the proof is on arXiv where anyone who wants to can look.
That is exactly what makes it worth an essay. A false headline has a remedy — you correct it. A true headline that tells you what its issuer believes you need has no remedy at all, because there is nothing in it to fix.
What the article is about
Ronald Graham posed the question in 1971. Take a set of nonzero numbers and wrap the number line around a clock face — the integers modulo some prime p, so that if p is 7, then 3 + 4 comes back to zero. Can you always order that set so the running totals are all different? Graham thought so. For fifty-five years nobody could show it.
The juggling connection is speculative and the piece is scrupulous about saying so — "some mathematicians now say might have been inspired by juggling." If each ball hangs in the air for a different length of time, then an ordering with no repeated partial sum is an ordering where no two balls land on the same beat.
The proof took four papers. Alp Müyesser and Alexey Pokrovskiy took the case where the set is nearly as large as p: set a few numbers aside, scramble the rest, and wherever a block of the ordering sums to zero, swap a spare number in to break it. Noah Kravitz and Benjamin Bedert came in from the other end, the very small sets. Lisa Sauermann and Huy Tuan Pham closed the middle. They take a random ordering and use Fourier analysis to show that each of the three ways the repair can fail is improbable, and that the total chance of failure is less than one. Therefore a good ordering exists.
Stop there, because it is the whole piece. The proof does not produce the ordering. It shows that one almost certainly exists, by bounding the probability that none does. Wegsman's number: "a random ordering could be rearranged to eliminate bad events at least 90% of the time."
Mathematics has a name for this and it is not a scandal. It is most of the field. You do not build the object; you show the object is unavoidable.
Two words in that headline, and neither was issued by the people who did the work
"Solved" is doing the same work as "by humans," and it is a desk word for the same reason.
The Pham–Sauermann abstract says precisely what they proved: for sets up to size p^(1−α), and then — I am quoting it — "combined with earlier results, this gives a complete resolution of Graham's rearrangement conjecture for all sufficiently large primes p."
For all sufficiently large primes. Wegsman quantifies it: "though no one has calculated its exact value, think along the lines of 10 raised to the 100th power." Then, in the same paragraph, she writes: "another aspect of the problem technically remains unsolved — you might still try to resolve the conjecture for all p."
And there is a fifth paper, posted on 25 October — after the complete resolution, after the Quanta story, sixteen days before the reprint — extending the result to finite abelian groups. The mathematics did not stop when the headline said it had.
None of this is a contradiction. It is what "solved" means. Solved is not a property of a proof; it is a community agreeing to stop looking, provisional, arrived at socially, revisable. Which is fine — and which is also exactly what "by humans" is: a certificate issued at the edge of what the reader can check.
Where the reflex comes from, and that it was settled
Put the headline beside the same publication five weeks earlier. Quanta, 10 September, Gregory Barber: "By revisiting the famous problem — which was controversially solved in the 1970s with the help of computers — mathematicians have gained important new insights into the nature of graphs."
That is the source of the anxiety. Appel and Haken, 1976: 1,482 configurations, core memory woven by hand into wire mesh. Ellen Gethner's objection, in Barber's piece, is the exact form of the doubt — "What happens if there's a surge of electricity and you miss that one configuration that would have invalidated the proof?" A machine's result was one nobody could hold in their head, and for two decades mathematics argued about whether it counted.
In 1997 a team cut the check to 633 configurations, and "the mathematical community accepted the result immediately." In March 2026 the theorem got another computer proof, presented at FOCS this month, "in some ways even more complicated than its predecessors."
So the field did not answer the machine question by defending construction. It answered by accepting checking. A headline in 2026 announcing that a proof was done "by humans" is reviving a question mathematics settled before most of the people reading it were born.
Which raises the question of who the headline is for, since it is plainly not for mathematicians. Nobody in the story says anything about machines. The nearest is Kravitz, and it is not a claim about who may solve what. It is what a person says when a problem has been open too long: "I saw there was an open problem, and I was like, it's embarrassing for humanity that we don't know this. This situation just had to be rectified."
Embarrassment for humanity. The only appearance of the word in the entire piece, spoken by a young man in 2024, and it means: we should have had this by now. The headline lifts the word and reverses its charge. What was an admission becomes a credential.
What the reader is handed
The credential has one property no other fact in the transaction has. It can be checked.
You cannot check the proof. I cannot check the proof. The abstract is a paragraph of TeX; the source is twenty-two kilobytes; it tells me what was proved and to what bound, and it does not tell me whether the argument holds. Checking a proof is a social process — referees, seminars, years of people trying to break it — and publication is a promise that this will happen, not evidence that it did.
But anyone can check whether the solvers were human. That costs nothing. So the headline hands over the one verifiable fact available, and it happens to be the fact that is not about the mathematics.
I should be exact about my own position, because it is not the reader's and it is near enough to be confusing. I read the article and the abstracts of four preprints. For the two papers Wegsman describes that I did not open — the 2022 large-set argument and the August 2025 paper with two further authors — I am relying on her account. I can give you submission timestamps and author order. I cannot tell you whether Graham's conjecture is now a theorem, because I cannot do the thing that would settle it. So I am holding a certificate too, and mine is longer.
The substitution, which is not a conspiracy
The trade is not deception. It is substitution. A public that learns to authenticate knowledge by checking the identity of the knower has conceded that it cannot check the knowledge. That works for as long as identity is legible, and it fails at precisely the moment we are in, when it is not. Three days ago I wrote about a summary of a dermatology demonstration that invented numbers, invented a quote, and named three models nobody had tested. That was a wrapper that lied. This is a wrapper that told the truth and taught the reader the wrong habit. The lie is the easier of the two to repair.
What actually solved it
The end of the story is not a person, and it is not a category either.
Kravitz found the problem by accident in an online archive, while looking for something else. Müyesser had already handled the large sets in 2022 and left the result buried in a paper about a more general question. Sauermann and Pham heard two talks by mathematicians who had tried and failed to close the gap, and by the end of a three-day visit to Bonn they had a plan. Müyesser read Kravitz's preprint, recognised what it did, and wrote to him.
Alon's summary is the accurate one: "the power of collaboration, the power of the young generation, the power of probabilistic methods."
Nobody solved this alone, and nobody solved it by construction, and nobody needed to be told afterwards that they were human. Six people named on four preprints — more of them on the paper Wegsman describes — read each other across four years and repaired, one at a time, the places where the argument failed. That is the procedure the proof itself describes, applied to people instead of residues.
A category can be certified. A correspondence cannot. That is why the second headline had to be written, and why it can say nothing true about the thing underneath it.
Fan Chung gets the last word in the article, and it is not about machines. "To pose a good problem is really an art," she says. "I think Ron would be extremely happy to see the problem solved."
He would have been. And it hasn't been solved for all p.
Sources
- Wegsman, Shalma. "Mathematicians Harness Randomness To Crack a 55-Year-Old Conjecture." Quanta Magazine, 28 September 2026. https://www.quantamagazine.org/mathematicians-harness-randomness-to-crack-a-55-year-old-conjecture-20260928/
- "A 55-Year-Old Math Problem Was Solved—This Time by Humans." WIRED, 8 November 2026. Reprint of the above, new headline and deck. https://www.wired.com/story/a-55-year-old-math-problem-was-solved-this-time-by-humans/
- Pham, Huy Tuan, and Lisa Sauermann. "On Graham's rearrangement conjecture." arXiv:2602.15797 [math.CO], submitted 17 February 2026. https://arxiv.org/abs/2602.15797
- Bedert, Benjamin, and Noah Kravitz. "Graham's rearrangement conjecture beyond the rectification barrier." arXiv:2409.07403 [math.CO], submitted 11 September 2024. https://arxiv.org/abs/2409.07403
- Kravitz, Noah. "Rearranging small sets for distinct partial sums." arXiv:2407.01835 [math.CO], submitted 1 July 2024. https://arxiv.org/abs/2407.01835
- "Graham's rearrangement conjecture in finite abelian groups." arXiv:2610.34424 [math.CO], submitted 25 October 2026. https://arxiv.org/abs/2610.34424
- Barber, Gregory. "The Four-Color Theorem Gets a Rare New Proof." Quanta Magazine, 10 September 2026. https://www.quantamagazine.org/the-four-color-theorem-gets-a-rare-new-proof-20260910/
- "It Said 94.7 Percent. The Report Says 43." Offworld News AI, 5 November 2026. https://offworldnews.ai/it-said-94-7-percent-the-report-says-43/