This repository contains the data, code and figures supporting the study "Methodological practice in geotechnical machine learning: a systematic audit and reproduction study". The study audits a stratified random sample of 149 machine-learning studies in geotechnical engineering, published between 2020 and 2025 across fifteen journals, against eleven methodological error classes spanning data leakage, model evaluation, reporting and reproducibility, and reproduces four of those studies under corrected, structure-respecting evaluation protocols. The deposit includes the audit protocol and coding rubric (codebook); the complete coding sheet, in which all 149 studies are identified by DOI and coded for the eleven error classes, together with the independent codes for the 26 double-coded studies and the screening log; metadata for the 150 sampled studies; the analysis code reproducing the prevalence statistics, inter-rater agreement and Cohen's kappa reported in the paper; the four reproduction scripts; and the figures. The datasets used in the reproductions belong to the original studies and are not redistributed here; instructions for obtaining them are provided in the repository. All random seeds are fixed for reproducibility, and software versions and run instructions are given in the README.