Reproduction archive for: The reported strength of the endometriosis-adenomyosis association is a function of diagnostic definition (analysis code, derived results and a verification package)

Published October 17, 2026 | Version 1.0 Software Open Reproduction archive for: The reported strength of the endometriosis-adenomyosis association is a function of diagnostic definition (analysis code, derived results and a verification package) Authors/Creators - 1. Department of Gynecology, The Second Hospital of Hebei Medical University, Shijiazhuang, China - 2. Department of Obstetrics, The Second Hospital of Hebei Medical University, Shijiazhuang, China Description Analysis code, small input files and every derived table and figure that backs a number in the manuscript “The reported strength of the endometriosis–adenomyosis association is a function of diagnostic definition: a bidirectional Mendelian randomization and endpoint-nesting analysis”. Contents. Inputs. FinnGen release-12 genome-wide association summary statistics for four N14_* endpoints (public mirror, no authorisation needed); FinnGen Risteys release-12 endpoint definitions and individual-level-derived CodeWAS co-occurrence counts; and, for the external replication in Section 3.10, the summary statistics deposited by Koller and colleagues (Nature Genetics 2026;58(5):1051–1061, Zenodo doi:10.5281/zenodo.18983492, CC BY-NC 4.0, non-commercial scientific use only — their raw files are not redistributed here, only statistics derived from them). The second file, No individual-level data are included or were accessed. Every input is public summary statistics or a published aggregate count. Two cautions for anyone reusing the Section 3.10 outputs. (1) Koller's deposit carries no per-allele effect estimates, so effects were reconstructed on a standardised scale while FinnGen effects are log-odds; a ratio mixing the two is on neither scale and is not an odds ratio — those rows are interpretable for direction and heterogeneity only. (2) The coefficients in Software attribution. The analysis scripts were drafted with the assistance of the ClawsGO Science Agent (ClawsGO Science Harness), a large language model-based research assistant, under the authors' direction. The authors designed the study, chose the endpoints, instruments and analyses, directed each iteration, reviewed and edited all content, and take full responsibility for the content of the publication. Contents. scripts/ (Python 3.12 and R 4.6, portable paths via the PAPER_BASE environment variable), data/gwas/ (instrument panels, the FinnGen release-12 endpoint manifest, and archived Risteys endpoint pages), fetch/ (download scripts for every public input, with checksums), results/ (all derived tables) and figures/ .Inputs. FinnGen release-12 genome-wide association summary statistics for four N14_* endpoints (public mirror, no authorisation needed); FinnGen Risteys release-12 endpoint definitions and individual-level-derived CodeWAS co-occurrence counts; and, for the external replication in Section 3.10, the summary statistics deposited by Koller and colleagues (Nature Genetics 2026;58(5):1051–1061, Zenodo doi:10.5281/zenodo.18983492, CC BY-NC 4.0, non-commercial scientific use only — their raw files are not redistributed here, only statistics derived from them). The second file, verification_package_paperA.zip , is a self-contained verification package. It has three tiers, each of which runs independently. Tier 1 recomputes 42 headline numbers from the archived result tables in seconds and independently re-derives the Wilson intervals. Tier 2 rebuilds the LD-independent instrument panel from the four official FinnGen release-12 summary-statistics files. Tier 3 recomputes the full sensitivity battery — inverse-variance weighted with random effects, MR-Egger with its intercept, weighted median, Cochran's Q and leave-one-out — for all six exposure–outcome pairings and prints a line-by-line comparison against the values reported in the manuscript. Every tier compares each intermediate product against the archived baseline and prints any divergence rather than passing over it. The package README, the data-sources note, every script and every comparison table are in English.No individual-level data are included or were accessed. Every input is public summary statistics or a published aggregate count. Two cautions for anyone reusing the Section 3.10 outputs. (1) Koller's deposit carries no per-allele effect estimates, so effects were reconstructed on a standardised scale while FinnGen effects are log-odds; a ratio mixing the two is on neither scale and is not an odds ratio — those rows are interpretable for direction and heterogeneity only. (2) The coefficients in A29_koller_internal_mr.csv are in standard-deviation units of liability and must not be exponentiated. Both points are also stated in the manuscript and in README.md .Software attribution. The analysis scripts were drafted with the assistance of the ClawsGO Science Agent (ClawsGO Science Harness), a large language model-based research assistant, under the authors' direction. The authors designed the study, chose the endpoints, instruments and analyses, directed each iteration, reviewed and edited all content, and take full responsibility for the content of the publication.