Spectral Detection of Correlated Errors on Superconducting Quantum Hardware via Parity-Circuit Canaries

Standard device characterization protocols, such as Randomized Benchmarking, typically report averaged, independent per-qubit error rates and are largely blind to correlated errors between neighbouring qubits, precisely the structure that breaks the independent and identically distributed (i.i.d.) noise assumption relied upon by most quantum error correction (QEC) decoders and mitigation strategies. This work estimates pairwise error correlations from parity-circuit canaries and applies spectral kernel analysis to the resulting correlation matrices: a departure of the kernel from positive semi-definiteness quantifies the violation of the independent-error assumption. As a proof-of-concept, the approach is emulated on publicly available surface-code data from IBM Quantum hardware (ibm_fez), where it surfaces a strong correlated error structure that originates in the correlation structure itself rather than in the marginal error rates. Full validation on directly measured canary circuits and extended acquisition remains future work.