Residual-First Benchmarking of NV-Center ODMR Models Using Public Data

This work presents an executed residual-first benchmark of nitrogen-vacancy (NV) center optically detected magnetic resonance (ODMR) frequency data using publicly available datasets. Residual-first benchmarking evaluates competing model families by examining residual structure prior to aggregate fit metrics. A conventional baseline spin-Hamiltonian model is compared to a reduced parameter model under explicitly declared constraints. The benchmark uses publicly archived ODMR sweep data. Two resonance dip centers are extracted per sweep using a fixed automated rule, yielding 105 ODMR sweeps and 210 resonance-center observations in the executable benchmark package. The reduced model introduces a predictable global offset due to a fixed parameter constraint but does not introduce additional structured residual behavior relative to the baseline. This satisfies the residual-first adequacy criterion. All benchmark scripts, derived tables, and figures are available in the associated reproducibility kit. Executable benchmark repository:https://github.com/jamesbourassa/qdl-benchmark-odmr Reproducibility kit:https://doi.org/10.5281/zenodo.18064816 Primary dataset:https://doi.org/10.5281/zenodo.14697917 This record represents Version 2 of the manuscript, aligned with the archived executable benchmark.