Reproducibility Package for Evaluating ChEMBL's Potential-Duplicate Annotation in Molecular Benchmarks

Reproducibility package supporting the manuscript “Evaluating ChEMBL's Potential-Duplicate Annotation in Molecular Benchmarks”. The archive contains ChEMBL 37-derived activity and pair-level data, the complete potential_duplicate snapshot, frozen preregistrations and amendments, blinded audit materials and de-identified coding, citation-check outputs, and the CARA LO-GPCR and NTAB confirmatory protocols, authorized scripts, outputs, and integrity manifests. Two historical code-provenance gaps are explicitly documented. The initially archived pair-builder variant omitted activity-ID fields required downstream; the original, a minimal reconstruction, and a portable reference version are retained. The three-level bootstrap table came from an unarchived interactive calculation on 2026-07-29; bootstrap_ci.py is a post-hoc reference reimplementation written on 2026-07-30 that reproduces all reported intervals but is not represented as the original code. The package excludes the unaudited historical ml_impact analysis and contains no RRE or FTO materials. Public source datasets retain their original licenses. Related records: preregistration and pre-execution code, DOI 10.5281/zenodo.21701846; preprint, DOI 10.5281/zenodo.21705010. Archive SHA-256: 75b48b9e4bf619f68d78561179690af888fb413e6e72b29a040f35630fb63282.