Reproducibility dataset for "Correcting phase-blind waiting-time inference under periodic driving"

Description Code, numerical data, independent validation outputs, and publication figures supporting the manuscript Correcting phase-blind waiting-time inference under periodic driving (Ruocun Ou, Nanyang Normal University), prepared for submission to the Journal of Statistical Physics. Scope. Waiting-time distributions between partially observed transitions can be used to infer dissipation in stochastic systems. Under periodic driving, however, discarding the phase of the external protocol can pair forward trajectories with the wrong time-reversed measure. The manuscript identifies a sufficient measure-matching condition for a previously conjectured phase-blind entropy-production bound, constructs a minimal three-state counterexample when that condition is absent, explains the failure through protocol-phase locking, and derives a corrected nonnegative estimator by pairing forward snippets with reversed-protocol snippets before marginalizing the phase. For the rotating three-state ring studied in the manuscript, the forward-only phase-blind estimator is \(1.4722\), while the physical entropy-production rate is \(1.0555\), corresponding to a \(39.5\%\) overestimate. The violation begins near drive amplitude \(a_c\approx3.213\). The corrected construction gives approximately \(0.098\) and satisfies the proved ordering with the phase-resolved bound and the physical entropy-production rate. This archive contains everything needed to reproduce the numerical results, validation tests, and figures reported in the manuscript and its Supplementary Material. Contents code/periodic_wtd_counterexample_search.py — continuous-time implementation of the periodically driven three-state Markov ring, periodic stationary state, killed-process waiting-time distributions, entropy-production rate, and phase-blind estimator. code/validate_minimal_counterexample.py — high-resolution calculation of the headline counterexample, grid-convergence table, chirality-reversed protocol, time-reflection-symmetric control, amplitude scan, and localization of the dominant contribution. code/final_numeric_audit.py — independent mechanical audit of the entropy-production definition, local detailed balance, frozen detailed balance, instantaneous cycle affinity, \(39.5\%\) overestimate, and threshold \(a_c\). code/redteam_measure_pairing.py — independent reconstruction of the forward-only, phase-resolved, and correctly phase-paired estimators, including the log-sum contraction used for the corrected bound. code/tolerance_redteam.py — robustness test under ODE relative tolerances \(10^{-8}\), \(10^{-10}\), and \(10^{-12}\). code/reviewer_quickcheck.py — fresh moderate-resolution calculation designed for rapid referee assessment. It reads no cached result files and exits with a nonzero status if any decision-level test fails. code/make_paper_figures.py and — regeneration of all publication figures.code/plot_minimal_counterexample.py data/minimal_counterexample_validation.json — headline values, controls, WTD normalization, killed-process tail, and localization diagnostics. data/minimal_counterexample_convergence.csv — simultaneous phase-grid and waiting-time-grid convergence data. data/minimal_counterexample_amplitude_scan.csv — entropy-production rate, phase-blind estimator, and their difference across the driving-amplitude scan. data/final_numeric_audit.json — independently recomputed entropy production, detailed-balance identities, exact unrounded percentage, and bisection bracket for the threshold. data/redteam_measure_pairing.json and — phase-resolved and phase-marginalized measure-pairing diagnostics at two resolutions.data/redteam_measure_pairing_highres.json data/tolerance_redteam.json — ODE-tolerance robustness results. figures/ — PDF and PNG versions of the three publication figures. REPRODUCIBILITY.md — complete execution order, expected numerical values, provenance, and cost notes. REVIEWER_GUIDE.md — shortest path from each central manuscript claim to its numerical evidence and falsification criterion. MANIFEST.csv — machine-readable claim–script–output–expected-result map. verify_reproduction.py — fast verification of archive completeness and consistency of all archived decision-level quantities. run_reproduction.py — ordered entry point for quick checking, figure regeneration, or full reproduction. SHA256SUMS.txt — byte-level integrity manifest for the archive. requirements.txt — minimum Python-package versions. Reproducing the results Install Python 3.10 or newer and the required packages: python -m pip install -r requirements.txt For the fastest independent assessment, run: python code/reviewer_quickcheck.py This command recomputes the periodic stationary state, entropy production, forward and reversed waiting-time statistics, chirality-reversed result, time-symmetric control, WTD normalization, killed-process tail, and corrected-bound ordering. It does not read the archived CSV or JSON results. To verify the archived high-resolution results without recomputing the complete calculation, run: python verify_reproduction.py For complete regeneration of the numerical results, robustness tests, and figures, run: python run_reproduction.py --full All calculations are deterministic. No random numbers, fitted parameters, trajectory sampling, or external datasets are used. Validation The archive documents five complementary checks. Independent entropy-production audit. The entropy-production rate is evaluated using the undirected-edge current formula independently of the ordered-transition expression. At \(a=5\), both calculations give \[ \sigma=1.0555268992. \] The maximum discrepancy between the two formulations is below \(10^{-15}\). Counterexample and threshold. The high-resolution calculation gives \[ \widehat{\sigma}_{\Psi}=1.4721863623>\sigma=1.0555268992, \] corresponding to an unrounded overestimate of \(39.4740734\%\). Independent bisection brackets the onset at \[ 3.2131543<a_c<3.2131836. \] Model-assumption checks. Local detailed balance, frozen detailed-balance fluxes, and zero instantaneous cycle affinity are verified numerically to residuals below \(2\times10^{-15}\). Every transition rate is positive, and the calculation uses the periodic stationary solution of the driven generator. Numerical robustness. Simultaneous refinement from \((N_\phi,N_\tau)=(48,1600)\) to \((192,3200)\) preserves the violation margin near \(0.41666\). Tightening the ODE relative tolerance from \(10^{-8}\) to \(10^{-12}\) also leaves the conclusion unchanged. The killed-process survival probability is below \(4\times10^{-11}\), and the WTD normalization error is negligible compared with the violation. Corrected measure pairing. The independent forward/reversed-protocol audit verifies \[ 0\leq\widehat{\sigma}_{C} \leq\widehat{\sigma}_{\psi} \leq\sigma. \] At the highest archived resolution, \[ \widehat{\sigma}_{C}=0.0978538789,\qquad \widehat{\sigma}_{\psi}=0.3532798261,\qquad \sigma=1.0555268992. \] This confirms that pairing the protocol phases before marginalization restores a rigorous nonnegative entropy-production bound. Citation Please cite the version-specific DOI of the archive that was used, together with the associated article: Ruocun Ou, Correcting phase-blind waiting-time inference under periodic driving, manuscript prepared for submission to the Journal of Statistical Physics. If the code, numerical outputs, or manuscript change during peer review, a new Zenodo version will be created and the revised manuscript will cite the corresponding version-specific DOI.