Behavioral Anomalies in Commit Telemetry: Pseudo-Labels, Author Effects, and Next-Window Prediction. Reproduction package

Data and code to reproduce the analyses of the manuscript “Behavioral Anomalies in Commit Telemetry: Pseudo-Labels, Author Effects, and Next-Window Prediction”. The package contains the pseudonymized commit records of seven public GitHub repositories (bitcoin/bitcoin, facebook/react, microsoft/vscode, neovim/neovim, sveltejs/svelte, tensorflow/tensorflow, torvalds/linux; 207,686 commits, January 2025 to July 2026), the cleaned features and five-commit windows, the result files, and the Python scripts of the pipeline, from the cleaning of the records to the figures. The README gives the execution order and the software versions.