LAUSR

Characterizing temporally correlated (``non-Markovian'') noise is a key prerequisite for achieving noise-tailored error mitigation and optimal device performance. Quantum noise spectroscopy can afford quantitative estimation of the noise spectral features; however, in its current form it is highly vulnerable to implementation non-idealities, notably, state-preparation and measurement (SPAM) errors. Further to that, existing protocols have been mostly developed for dephasing-dominated noise processes, with competing dephasing and relaxation effects being largely unaccounted for. We introduce quantum noise spectroscopy protocols inspired by spin-locking techniques that enable the characterization of arbitrary temporally correlated multi-axis noise on a qubit with fixed energy splitting, while remaining resilient to realistic static SPAM errors. By validating our protocol's performance in both numerical simulation and cloud-based IBM quantum processors, we demonstrate the successful separation and estimation of native noise spectrum components as well as SPAM error rates. We find that SPAM errors can significantly alter or mask important noise features, with spectra overestimated by up to 26.4% in a classical noise regime. Clear signatures of non-classical noise are manifest in the reconstructed IBM-qubit dephasing spectra, once SPAM-error effects are compensated for. Our work provides a timely tool for benchmarking realistic sources of noise in qubit devices.