In July 1944 the Hungarian mathematician Pál Turán was a forced labourer in a brick factory near Budapest, pushing carts of bricks from kilns to storage yards along rails. Where the rails crossed, the carts jumped the track. He asked himself how to lay the rails so that they cross as few times as possible. Eighty-two years later the question is still officially open — and one of the 719 papers OpenAI released on October 6, 2026 claims to answer it, for every number of kilns and yards, in sixteen pages, with a machine-checked statement. What followed is the third shape this library has recorded from that release. Specialists set out to understand one result (UC-328). Outsiders built on another before it was checked (UC-329). This one met silence: no specialist has publicly commented, and when the claim was added to Wikipedia, a long-standing editor removed it within hours — “let’s wait until we have published reliable sources.” The wait has a history. This exact problem was declared solved in the 1950s, accepted for about a decade, and then found to have a gap that was never filled.
The problem has an author and a place. Turán told the story himself, in a short memoir published in 1977, after his death: “In July 1944 the danger of deportation was real in Budapest, and a reality outside Budapest. We worked near Budapest, in a brick factory. There were some kilns where the bricks were made and some open storage yards where the bricks were stored. All the kilns were connected by rail with all the storage yards.”[1] The work, he wrote, was not difficult — “the trouble was only at the crossings. The trucks generally jumped the rails there, and the bricks fell out of them.”[1]
So he turned it into a question: “the idea occurred to me that this loss of time could have been minimized if the number of crossings of the rails had been minimized. But what is the minimum number of crossings?” He saw within days how to improve the yard he was in, but the general problem “seemed to be very difficult,” and he put it aside “to times when my fears for my family would end.”[1] The companion drawing to this case lets a reader try the same thing.[15]
In 1952 Turán mentioned the problem to the Polish mathematician Kazimierz Zarankiewicz, who published a solution: a way of laying the rails, and an argument that nothing could do better.[1][4] For about a decade it stood. A 1963 paper on a sister problem still states that “the exact formula … was found by Zarankiewicz.”[3] Then a gap was found in the argument — independently by two mathematicians, Ringel and Kainen — and, in Turán’s words, it was one “which nobody has been able to fill so far — in spite of much effort.”[1][5] The layout was never beaten. The proof that it could not be beaten was gone.
The sister problem — connect every point to every other point — has an even less likely origin: its formula was proposed around 1958 by Anthony Hill, a British artist with no formal training in higher mathematics.[3][13] Its authors ended their paper with a hope: “that this problem will eventually come to the attention of someone who can settle the above conjecture.”[3] Decades of work settled only small cases: up to six kilns in 1970, a handful more by computer in 1993, leaving seven kilns with eleven yards, and nine with nine, as the smallest unsettled cases.[6][7]
The October 6 release contains two papers on these problems, both authored “OpenAI” and dated September 23: sixteen pages on Turán’s problem and thirteen on Hill’s.[8][9] Neither searches through drawings. Each converts the geometry of a drawing into a statement about matrices and proves that statement.[8] Both have machine-checked statements covering every size, and the definitions are the standard ones: any curved rails, any positions, every crossing counted.[10] Neither paper is among the three OpenAI withdrew or the fourteen it revised the following day.[11]
And then, as far as this case could find, nothing. No crossing-number specialist has commented publicly on either paper. No one has reported checking the argument, finding a flaw in it, or building on it. One mathematician’s roundup of the release notes that it “reports proofs” of both formulas and adds, of the release as a whole, that the results are still “to be verified and digested by human mathematicians in the months to come.”[12]
The clearest response is a small one, and it is on the public record. On October 7 an editor added the claim to the Wikipedia article on Turán’s problem. Under six hours later David Eppstein, a computer scientist and long-standing editor, removed it with the note: “let’s wait until we have published reliable sources.”[14] The article still describes the problem as open. That is not a judgement on the proof. It is the rule a reference work runs on — and on this problem it is a rule with a memory. A proof was accepted here once.
Turán, a forced labourer, watches carts jump the rails at every crossing and asks what the fewest crossings could be.[1]
The QuestionThe claim is added to Wikipedia and removed within hours, pending published reliable sources.[14]
The WaitZarankiewicz’s 1950s proof was treated as settled — a 1963 paper cites it as the exact formula — until a gap was found.[3][5]
Kleitman, 1970: the formula holds when one side has at most six. A few larger cases followed by computer in 1993.[6][7]
For every number of kilns and yards. The sister problem takes thirteen. Both dated September 23, 2026.[8][9]
The claim was added at 12:08 UTC on October 7 and removed at 17:52, pending published reliable sources.[14]
The work itself was not difficult; the trouble was only at the crossings. — Pál Turán, on the brick factory near Budapest, July 1944 [1]
| Dimension | Evidence |
|---|---|
| Regulatory (D4) Origin · 88 | Wikipedia’s edit removing the claim: “let’s wait until we have published reliable sources.” On this problem a proof was accepted for about a decade before a gap was found.[5][14]A Rule With a Memory |
| Quality (D5) L1 · 82 | Sixteen and thirteen pages claim both formulas for every size. The machine-checked statements use the standard definition, any curves allowed.[8][9][10]Short, Complete, Machine-Checked |
| Employee (D2) L1 · 70 | No public comment found from any crossing-number specialist: no check, no objection, nothing built on it. One roundup notes it and says verification will take months.[12]The Specialists Have Not Spoken |
| Operational (D6) L2 · 66 | The proof turns a drawing into a statement about matrices; decades of human progress came case by case, partly by computer. Not withdrawn or revised.[7][8][11]Matrices, Not Search |
| Customer (D1) L2 · 45 | The public reference still lists the problem as unsolved, by design, until reliable sources are published.[14]Readers Are Told It Is Open |
The cascade originates in D4 (Regulatory): the rules by which a field accepts a result — published, refereed, absorbed — and the memory behind those rules.[5][14] It propagates to D5 (Quality), a claim that would close two problems open since 1958 and 1944, in papers short enough to read in an afternoon, with machine-checked statements;[8][9][10] and to D2 (Employee), the specialists whose judgement would settle it and who have not yet spoken.[12] It lands in D6 (Operational), where the method is again theory rather than search, with no finite list of cases to re-run;[8][11] and in D1 (Customer), the readers of reference works, who are told, correctly, that the problem is open.[14] D3 (Revenue) is not scored.
Three results from one release now sit side by side in this library. In UC-328 the specialists began reading. In UC-329 a crowd began building. Here the field waits — and of the three, this is the response with the best precedent. A machine check answers one question: does the formal statement follow from the axioms? It does not answer whether people understand why, and it is understanding that turned out to be missing in the 1950s. The honest limit: silence is weak evidence. Specialists may be reading and saying nothing; this case searched the public record, not private correspondence. And nothing here says the proof is wrong.
-- UC-330: The Trouble Was Only at the Crossings: 6D Diagnostic Cascade
-- Turan's brick factory problem, posed July 1944 near Budapest: fewest rail crossings joining every kiln to every storage yard (crossing number of K_m,n). Zarankiewicz published a construction and proof in the 1950s; accepted about a decade (Harary and Hill 1963 cite it as the exact formula); gap found independently by Ringel and Kainen; never filled. Proven for min(m,n) at most 6 (Kleitman 1970); K_7,7 and K_7,9 by computer (Woodall 1993); smallest unsettled K_7,11 and K_9,9. Sister problem: Harary-Hill conjecture for K_n, formula proposed about 1958 by Anthony Hill, an artist. OpenAI release Oct 6 2026, result 165: two papers dated Sep 23, 16 and 13 pages, claim both formulas for all sizes; Lean statements formalized for both, standard drawings, crossings counted as points; not withdrawn or revised. No specialist comment found. Wikipedia: claim added Oct 7 12:08 UTC, removed 17:52 UTC by David Eppstein pending published reliable sources.
FORAGE trouble_at_crossings
WHERE long_open_problem_claimed_solved = true
AND statement_machine_checked = true
AND prior_accepted_proof_failed = true
ACROSS D4, D5, D2, D6, D1
DEPTH 3
SURFACE trouble_at_crossings
DIVE INTO checked_not_accepted
WHEN specialist_comment_found = false
AND reference_work_declines_claim = true
TRACE acceptance_lag_cascade
EMIT trouble_at_crossings_signal
DRIFT trouble_at_crossings
METHODOLOGY 88
PERFORMANCE 34
FETCH trouble_at_crossings
THRESHOLD 1000
ON MONITOR CHIRP high 'An 82-year-old problem posed in a wartime brick factory is claimed solved in sixteen pages by an AI model, with a machine-checked statement. No specialist has publicly commented, and Wikipedia removed the claim within hours pending published sources. The same problem was declared solved in the 1950s, accepted for about a decade, and then found to have a gap that was never filled. A machine check shows the statement follows; it does not show that anyone understands why.'
SURFACE analysis AS json
Runtime: @stratiqx/cal-runtime · Spec: cal.semanticintent.dev · DOI: 10.5281/zenodo.18905193
The statement is machine-checked and the papers are short. The field has not moved. A formal check shows a statement follows from the axioms; acceptance means people understand it and stand behind it. They are separate steps, and only the first is done.[10][12]
This problem was declared solved in the 1950s and cited as solved in 1963. Then a gap was found that was never filled. “Let’s wait” is not obstruction here. It is what the field learned on this exact question.[3][5][14]
Specialists reading (UC-328). A crowd building (UC-329). A field waiting (this case). The same machine, the same week — and what happened next depended entirely on the people around each result.
Primary sources are Turán’s own 1977 account and the 1963, 1970, 1973 and 1993 papers, read from scans; OpenAI’s two papers, machine-check note and change history; and Wikipedia’s public revision history. No specialist comment on this result was found in a search of blogs, forums, preprints and press to October 13, 2026; that is a search of the public record only. This case does not claim the proof is correct or incorrect. Source 15 is an interactive companion visual made for this case.
This case separates two things that are easy to merge: a result that has been checked, and a result people have accepted. The gap between them is usually a memory of something that went wrong. Book a discovery call to map where that gap sits in your own approvals, and what it is protecting.