A Reproducible Provenance Audit of the Public Einstein Elevator Data Package: Trace Correspondence, Metadata Linkage, and the Non-Identifiability of Sensor-Certificate Risk from the Public Archive

We conduct a version-pinned, reproducible forensic audit of the public Einstein Elevator data package (Zenodo: 10.5281/zenodo.15167723) to determine whether it supports validation of quantum sensor certificates. Through an exhaustive cross-file sliding-window correlation search over 44,207,922 comparisons and exact six-decimal rounding checks, we find that all 31 exported sensor channels have a unique best match within the enumerated raw traces under the stated protocol. Residual numerical differences are highly consistent with text-based absolute quantization; the correlation-loss approximation 1 - r ≈ δ² / (24σx²) is diagnostic for this dataset, not a universal proof. We inspect the MATLAB-derived 318-point population series and reconcile it with a 2022 workbook entry only at the level of counts. An archive-wide search of 181 file objects and 1,654 MATLAB metadata fields finds no documented join key linking the 2022 run-indexed population series to the 2017 timestamped traces under the enumerated search protocol. Using Fréchet-Hoeffding bounds and a Charnes-Cooper linearization, we establish that the public archive does not supply the joint observations needed to identify the EKF false-safe risk without additional coupling assumptions; the formal program is illustrative and is not populated with empirical failure/acceptance coefficients because the required linked observations are absent. We frame this as a bounded data-contract limitation for downstream statistical validation, not as a failure of the original experiment.