63 Checking Moves: Double Agent Proves the Limit of a 58-Year-Old Chess Problem

In 2026, my father taught me to play chess on an ebony board in Kerala. He told me: 'Chess does not teach you how to win. Chess teaches you that everything has a limit, and your job is to find that limit.' I did not fully understand him. It took until the night of October 13, when I sat before a screen in Da Nang reading the results of a research team working on a system called Double Agent, for me to see my father's words proven by a machine, in mathematics. The story begins with a number. But before the number, let us speak of years. In 2026, a chess problem was posed, and for 58 years afterward no one solved it completely. The problem asks: in a single position, what is the maximum number of legal moves that all give check? It sounds simple. But the simple things in chess are usually the hardest. This is not competitive chess. No player sits at the board, no clock ticks, no Elo rating exists. This is the world of chess problem composers. They are a small, quiet community, patient to the point of severity. They do not compete against one another. They compete against the very limit of the 64-square board. This branch has its own tradition called retrograde analysis: from a position on the board, one traces backward to see whether it could arise from a legal game. This matters more than it appears. The most beautiful position is meaningless if it cannot appear in a real game. Legality is the spine of the whole discipline. And in Double Agent's problem, legality is also a mandatory condition: the position must be reachable from a legal game, the side to move must have at least one legal move, and every legal move must give check. Before Double Agent, the record for the maximum number of checking moves in a single position belonged to Norbert Ringeltaube, with 60 moves. And the record for the longest consecutive checking sequence belonged to Alexey Khanyan, who in 2026 constructed a position with 54 consecutive checks. Those numbers were found by human hands, by human eyes, through thousands of silent hours of trial and error. I once sat up all night with such a position. In 2026, while still organizing chess tournaments in Da Nang, I had an old Hungarian friend who composed chess problems. He told me: 'William, a chess problem is like writing poetry. You are confined to 64 squares, but it is precisely that confinement that gives birth to beauty.' I have never forgotten those words. They shaped how I see everything since, including the football matches I once commentated. I remember 2026, when the sports paper I had served for 18 years killed my piece because it thought I 'wrote literature in a sports report.' I left the newsroom, started my own blog, and began every article with an image instead of a scoreline. I thought I had been exiled from my craft; in truth, the craft led me deeper into myself. The chess problem Double Agent solved is the same: it needs no noise. It only needs to be right. What Double Agent achieved lies not in the number. It lies in the method of proof. The research team did not use brute force — that is, constructing every possible position and checking each one. They stated plainly that this is computationally infeasible: the number of board combinations is so vast that no machine could exhaust them in reasonable time. Instead, they took a different road, slower but more certain. First, they proved geometric properties of positions rich in checking moves — constraints involving pins, blocking, and square control. Then they formalized the entire argument in a proof assistant called Lean 4 — a language in which a machine can verify every logical step, tolerating not a single slip. The remainder of the problem was reduced to SAT form, and finally the whole proof was machine-checked. The result: they proved that a position attaining exactly 63 legal checking moves exists, and that no position can exceed 63. This is a closed result — containing both an existence part (construction) and an impossibility part. Ringeltaube's old record was 60, meaning humans advanced three moves with the aid of a machine. Three moves. It sounds small. But in the world of combinatorial problems, three moves is a gap that has nearly slammed shut. It shows that the old human record had crept close to a hard ceiling, and only a structural proof could break through it. And this is where I want to linger longest. The second problem — the consecutive checking sequence — Double Agent gives the number 364. But this is not a record. It is an upper bound. That is: the machine proves that no sequence runs longer than 364 consecutive checks. But it does not prove that a position attaining exactly 364 exists. The record actually constructed remains 54, by Khanyan, from 2026. This is the detail most headlines will skip. And that is why I am writing this piece. Picture it this way: someone says 'no one can run faster than 40 km/h in this city' — that is an upper bound. But that person has not proven that anyone actually runs at 40 km/h. Between 'cannot exceed' and 'can attain' lies an abyss. Double Agent bridged that abyss for the 63 problem. For the 364 problem, it still stands on the near shore. The news will say: 'AI solves a 58-year-old chess problem.' It sounds magnificent. But if you read closely, the real story is different. 63 is a closed result. 364 is only half a result. What Double Agent truly solved is only the first problem; the second remains open. A headline that merges the two into 'solving a chess problem' is a deliberate exaggeration — not quite a lie, but a way of letting the reader imagine more than the truth. This is the kind of exaggeration I have seen hundreds of times in 40 years in journalism. Worse still: the proof itself has not been independently verified. Lean 4 and SAT are powerful tools, but they guarantee only this: if the formalization is correct, then the conclusion is correct. If there is a gap in how the problem was encoded, some unchecked hidden assumption, the whole edifice can collapse. In formal mathematics, this is the classic failure mode. No third party is reported to have re-checked the result. What is called 'machine-checked' is in fact the research team checking itself with its own tools. I say this not to diminish the achievement. I say it because I have reported on chess for 40 years, and I have learned one thing: the truth of a number matters as much as its origin. A correct result that is inflated will quickly be forgotten. A correct result presented honestly will live longer. Based on my experience watching matches and chess problems, the most remarkable thing here is not that the machine found 63, but that it proved 63 is the end. Finding is easily mistaken for brilliance. Proving is where intelligence truly lies. So what should we remember? Remember that this is not AI solving chess. Chess — with its billions of games, with the endless creativity of human beings — has no upper bound to solve. What was solved is only a narrow theorem: at most 63 checking moves in a single position. Small. But absolutely precise. What is worth keeping is not the number. It is the method: humans taught a machine not only how to find, but how to prove that nothing can be better than what it found. In an age when machines can say anything, the ability to prove is what is valuable. My father said: everything has a limit. Fifty-eight years later, a machine proved it with 63 checking moves. I thought I had been exiled from chess; in truth, chess led me deeper into myself. Perhaps one day, someone will prove the limit for 364 — and then the chess problem will teach us once more what my father taught me in 2026.