This preprint documents a reproducible failure mode in cross-platform quantum benchmarking: compiler/transpiler optimization can silently invalidate circuit-depth (or “barrier-based”) proxies for decoherence exposure by eliminating the intended experimental variable. We show how barrier or idle-depth manipulation may collapse after compilation, producing misleading null depth-dependence or inflated conclusions. We provide an audit workflow based on compiled-circuit metadata (depth, 1Q/2Q gate counts, and optimization settings) and recommend benchmark designs that encode the experimental variable in gate parameters or analysis steps (e.g., phase-sweep parity witnesses), which are robust to structural circuit simplification. Related dataset: Cross-platform GHZ benchmarking JSON logs and analysis artifacts are available on Zenodo (concept DOI: 10.5281/zenodo.18530246; current version: 10.5281/zenodo.18530247). This methodological note is a companion focused on compiler/transpilation artefacts and reporting standards for compiled-circuit audits.
Arturo Cerezo (Wed,) studied this question.