SIGNALprophet format
Scientific benchmarks are quietly adopting a new kind of audit: one that puts a number on how much hidden help a claimed result would need to survive. The cheating budget, in plain terms, is the minimum side information a classical imitation would need to reproduce a quantum score.
The 2607.13090 paper applies this idea to quantum advantage, the long-running claim that a quantum computer can beat a classical one on a specific task. The framework replaces "we won" with "here is the minimum benchmark-correlated classical information required to close the gap." The IBM numbers on CHSH and Mermin–GHZ, and the magic-square test, are exhibits for that audit, not the news itself.
The mechanism is portable: take any reported score gap, measure the conditional dependence between the benchmark and any side information, and compute the threshold above which a classical cheat would have closed it. On a current quantum-kernel benchmark, the 2607.13090 paper reports a measured value of 0.5, above the 0.375 threshold required to explain the gap, and yields perfect classical classification. The audit's verdict: the apparent quantum edge can be reproduced by benchmark-correlated information alone.
The pattern generalizes. AI benchmarks, climate models, randomized trials, every domain where a "win" is announced by a score, are now candidates for the same falsifiability yardstick. The quantum field just got the first clean worked example.