Simulator Choice as an Experimental Factor in a Registered BSM1 Fault-Classification Audit: Data and Reproducibility Record

This dataset and reproducibility record supports the manuscript “Simulator Choice as an Experimental Factor in a Registered BSM1 Fault-Classification Audit.” The deposit contains registered inputs, all 355 observed confirmation-run outputs, 20 eligible strict solver-tolerance diagnostic trajectories, frozen protocol artifacts, derived analyses, figure and table source data, reproducibility scripts, validation records, and a SHA-256 file manifest. The high-severity campaign comprises 23 complete blocks plus one preserved interruption: C008 stopped after 7 of 12 required cells and was not rerun, replaced, imputed, or reconstructed. Six low-severity robustness blocks are complete. The supported claim is limited to the stability of a three-level classification under the registered paired BSM1 protocol. The deposit does not establish fault masking, simulator interchangeability, solver equivalence, diagnostic performance, physical-plant validity, or deployment readiness. Data, metadata, documentation, and figure/table source data created for the study are licensed under CC BY 4.0. Original analysis, verification, figure-generation, and document-build code created for the study is licensed under the MIT License. Third-party software and benchmark implementations remain under their original terms and are not redistributed. Rebuild instructions and provenance details are provided in README.md.