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

This Version 8.0 adds two controlled direct-hardware follow-up studies to the existing Spectral Detection of Correlated Errors on Superconducting Quantum Hardware via Parity-Circuit Canaries series, performed on a superconducting quantum processor accessed through a commercial cloud service. The central empirical quantity is raw excess covariance, defined as the joint probability of both logical errors occurring minus the product of their individual probabilities, where the two error indicators are measured relative to a deterministic ideal Canary output. Spectral covariance summaries are retained as exploratory diagnostics; the count-level, control-aware statistical tests are the primary evidence. The first follow-up used one explicit physical pair, four joint readout calibration states, six Full X-X-CX Canary blocks, and six state-matched local |10> control blocks. The Full Canary gave an excess covariance of +0.00475873, while the paired Full-minus-local contrast gave +0.00493357. Both block-stratified two-sided permutation p-values were 0.00049975 at 2,000 draws, while the local control was not significant. The second follow-up measured five preselected disjoint connected pairs in one 80-circuit, 81,920-shot job. All five Full-Canary results remained below 0.05 after Bonferroni correction across the five planned pairs. Four of five paired Full-minus-local contrasts also remained below 0.05 after the same conservative correction; the fifth contrast was positive but not correction-decisive. The resulting evidence supports a restricted conclusion: under the recorded hardware, layout, compiler and calibration context, and measurement times, the effective error process for the tested logical and transpiled Canary paths was not adequately described by fully independent local error indicators. The release does not identify one microscopic physical cause, establish a device-wide or provider-wide invariant, or replace replication across distinct calibration epochs and backends. The accompanying raw counts, submission metadata, recomputation script, and figure make the new controlled follow-up auditable. The narrative text uses neutral language and includes no cloud-provider token or credential; the raw provenance files retain the original backend identifier for full auditability.