Design-dependence of comparative disproportionality estimates in FAERS antidepressant sexual-dysfunction signals: analysis code, configuration files, control definitions, control-selection protocol and derived results (deposit version 2)

Analysis deposit, version 2, accompanying the manuscript "Design-Dependence of Comparative Disproportionality Estimates: Comparator Choice, Reporter Type and Empirical Calibration in FAERS Antidepressant Sexual-Dysfunction Signals (2014–2025)" (Tahir Hakami). It contains everything needed to regenerate the cohort and every reported estimate from the public FDA Adverse Event Reporting System quarterly files (2014Q1–2025Q4, retrieved 1 August 2026; distributed since March 2026 through the FDA Adverse Event Monitoring System pages), which are not redistributed: a download script reproduces the retrieval with integrity checks. Contents. S1, completed READUS-PV checklists (version 2 mapped to the revised manuscript; version 1 retained). S2, drug-name matching audit. S3, outcome, indication, drug-name, negative- and positive-control and combination-product dictionaries. S4, negative-control estimates, fitted empirical nulls, calibrated p-values, leave-one-out and bootstrap results. S5, analysis code, commands as executed and run logs. S6, reporter-type analyses under inclusive and restrictive definitions, including the logistic model with the drug × reporter interaction. S7, report richness and expanded control calibration. S8, sensitivity analyses. S9, combination-product identification and recalibration. S10, figure source data and scripts (version 1). S11, independent verification of the empirical null in R. S12, near-duplicate detection. S13, rank correlations against external reference orderings with permutation tests. S14, product-source heterogeneity. S15, comparator choice (broad-reference design and control estimates by design). S16, detailed methods. S17 (new in version 2), the second-phase analyses: the pre-specified negative-control selection protocol with its dated record and SHA-256 digest, the label term lists, candidate tables and yields at three count thresholds, rebuilt cohort reaction and demographic tables, calibration with propagated null uncertainty (grid posterior, resampling of controls and of reports, stability by number of controls), tipping-point, reporter-stratified and exchangeability analyses, adjusted reporter-specific contrasts, the 34-specification curve, a parameter-recovery simulation, the R verification of the posterior fit, and the source data and scripts for Figures 1–4 of the revised manuscript. Relation to version 1 (10.5281/zenodo.21764525, deposited 5 August 2026). S17 and the version-2 checklist in S1 are added; nothing in S2–S16 was recomputed. MANIFEST.csv was regenerated for the whole deposit and SHA256SUMS.txt added (sha256sum -c SHA256SUMS.txt verifies all 318 files); operating-system and console files are excluded. See README.md and S17_revision_protocol/README_S17.md. Software as executed: Python 3.10.12 with pandas 2.3.3, NumPy 2.2.6 and Matplotlib 3.10.9; the original pipeline used pandas, NumPy, SciPy and statsmodels as recorded in S5; R 4.6.1 with EmpiricalCalibration 3.1.4 for the verifications. Random seeds are fixed in the scripts. Once those are in, the other fields from the note are: Creator Hakami, Tahir (Jazan University; ORCID 0000-0002-1933-6044); Version 2; Language English; licence as in version 1; Related work: 10.5281/zenodo.21764525, relation "Is new version of", type DOI. Reserve the DOI before publishing and send it to me.