Certifiable QuOps: a verification wrapper around any figure of merit
Figures of merit such as quantum volume and QUOPS, and measures from error mitigation and error correction, tell us much about what a machine can do. But it is not enough to define a new metric, however good it is. The key for benchmarking is a mechanism that allows a third party to rigorously certify any claimed achievement. In this talk I will argue that certifiability must be part of the definition of the QuOp. I will introduce the idea of a certifiable QuOp, based on our verification-inspired benchmarking (arXiv:2404.10739). There, any verification protocol can be compiled into a benchmark whose output is a lower bound, with high probability, on the probability of a correct accepted result for every computation in a declared class. The benchmark makes no assumption on the noise model, the hardware or the algorithm. It only assumes that the device stays consistent, and it needs no additional qubits beyond the target computation. Our aim is a verification wrapper, tailored to each hardware platform, that can be placed around the QuOp whatever its final definition, so that every claim comes with a guarantee that others can check.