Data and code for Reference-dependent loss reversals in squeezed-reservoir frequency estimation

This dataset accompanies the manuscript "Reference-dependent loss reversals in squeezed-reservoir frequency estimation" by Matthias Jakob (Optics Communications manuscript version 0.8). It contains numerical data and source code for continuously driven two-level systems monitored through homodyne detection in independent squeezed thermal Markov reservoirs, with physical detector loss modeled by vacuum admixture at the output. The central studies compare classical Fisher information in the measured records across detector efficiencies and drive frequencies, examine the effect of changing from a fixed thermal-parameter reference to a reference with equal total reservoir occupation, and test initially uncorrelated product states with the same local marginals as the entangled preparation. Additional data document conditional-state complementarity diagnostics and a secondary ablation of posterior weights in adaptive homodyne phase selection. The release comprises two ZIP files. sqdata.zip contains the complete archived sample arrays for the frequency, occupation-matched and equal-marginal product-state studies, their fixed designs, seeds, C++ simulators, Python analyses, numerical checks, result tables, and all seven manuscript figures with plotting code. sqextra.zip contains all 58 archived arrays for the adaptive-control study and 34 recovered arrays for the early complementarity study, together with corresponding source code and summary tables. README files, binary-format documentation, manuscript-to-data mappings, provenance records and SHA-256 checksums are included. Coverage limitation: the early complementarity archive was truncated. Its 22 main and twelve numerical-refinement arrays were recovered with matching CRC and original SHA-256 checksums. Four independent-repeat arrays and three historical driver/validation scripts could not be recovered; their existing result summaries are retained and explicitly identified. Historical nominal-frequency calibration material is supplied as source and summaries only. Complete time-resolved photocurrents and individual adaptive filter-bank states were not archived. The supplied arrays are sampled trajectory outputs, not full measurement time series. All data are numerical simulations, not experimental observations. Reanalysis of the included samples reproduces the archived result files in the tested environment. Reported confidence families apply to the stated discrete comparisons and use normal approximations; they exclude systematic discretization error. The results do not establish a universal benefit of loss or squeezing, a global measurement optimum, or an entanglement-specific metrological advantage. Data and documentation are provided under CC BY 4.0; source code under the MIT license. AI-assisted tools supported code, analysis, figure preparation and documentation.