Imagine you and a friend are both taking the same multiple-choice test in separate rooms — no communication allowed. You can't copy each other's answers, but before the test you can agree on a shared strategy. Now imagine you could share a pair of magic dice: no matter how far apart you are, when you roll them they're correlated in ways ordinary dice never could be. Tsirelson's problem asks: does it matter whether your magic dice are built from a finite collection of parts, or whether you're allowed an infinite warehouse of entangled components? This paper says yes, it matters — and here's the exact test that proves it. The committed claim: Gao, Yu, and Zhi construct an explicit binary linear system game — a specific system of 1,417,152 equations in 1,889,684 binary variables, each equation touching exactly three variables — that can be won with certainty using commuting operator (infinite-dimensional) quantum strategies but cannot be won perfectly using any finite-dimensional quantum strategy (or any limit thereof). This is the first explicit, human-inspectable counterexample to Tsirelson's problem in its approximation form. The backstory matters. In 2020, Ji, Natarajan, Vidick, Wright, and Yuen proved MIP=RE, which implied that Tsirelson's conjecture is false — the sets Cqa (closure of finite-dimensional quantum correlations) and Cqc (commuting operator correlations) are not equal. But that proof was a complexity-theoretic existence argument: it told you a separating game exists somewhere inside an astronomically large construction, without handing you one you could write down. The present paper makes it constructive. You can read the game. You can check it. They did check it — in Lean 4 with Mathlib. The architecture is algebraic, not computational in the usual sense. The authors work within the linear system game framework introduced by Cleve and Mittal, where a perfect commuting operator strategy exists if and only if a certain solution group is nontrivial. They engineer a system whose solution group has the right properties — nontrivial (so a perfect Cqc strategy exists) but "non-hyperlinear" (so no finite-dimensional approximation suffices, blocking Cqa perfection). The key technical ingredient is embedding a finitely presented group known to be non-hyperlinear into a system with the rigid binary-variable, three-terms-per-equation structure that linear system games require. This is a feat of combinatorial algebra, not quantum hardware. Integrity here is unusually strong for a theoretical result. The entire defining system and the proofs of separation are formalized in Lean 4, a proof assistant where every logical step is machine-checked. This is not "we ran a simulation and it looks right" — it is a certified mathematical proof. The classical value of the game is also computed exactly and formalized. There is no room for the usual concerns about cherry-picked benchmarks or unreported negative results; the object either satisfies the algebraic conditions or it doesn't, and the proof assistant confirms it does. The milestone question for this line of work is whether explicit constructions can be made small enough to be physically realizable — could you actually run a separating nonlocal game in a lab? The current system has ~1.4 million equations, which is astronomically far from any near-term experimental test. The gap between "explicit on paper" and "testable in a lab" is the open frontier. A system with, say, fewer than 100 variables and equations that still separates Cqa from Cqc would be a landmark. Nobody knows if that's possible. The obvious next experiment the authors did not run: optimizing the construction for size. The 1.4M-equation system is explicit but not minimal. Shrinking it — finding the smallest linear system game that separates Cqa from Cqc — is a natural follow-up. The honest read is (c): this is a separate hard problem they're likely aware of and may be working on, but it requires different techniques (group theory optimization, computer search) and is a paper unto itself.
