I recently migrated my research workflow from Windows to Debian and started checking my existing research repositories. I wanted to make sure that the analyses still ran as expected and that the migration had not altered the computational results.
Several repositories showed large numbers of modified files immediately after the migration. These changes affected source code, notebooks, documentation, data, and configuration files. However, by ignoring end-of-line differences, I found that the changes were caused by Windows CRLF line endings rather than actual changes to the files. I restored the affected repositories after checking the differences. The same issue also appeared in the PaperMod Git submodule of my academic website, which I checked separately.
The hierarchical Bayesian working memory project required more work. I replaced its previous Conda environment with a project-local uv environment while keeping Python 3.12 and the required packages. I also checked the notebooks for Windows-specific paths and confirmed that the analysis used project-relative paths. The trial-by-trial behavioral adaptation project did not need a new environment because its existing renv setup was already suitable for the R analysis.
I then reran the working memory analysis on Debian. Some numerical results changed slightly, but the model ranking and main interpretation remained unchanged. This provided evidence that the migration had not materially affected the main scientific conclusions.
In practice, the Debian setup has also made my workflow more efficient. My research code runs noticeably faster, and package installation and environment management are simpler.
A project is not defined only by its code and data. The environment, file paths, and version-control state can also affect reproducibility. The migration made me more careful about these parts of the research workflow. I will therefore check them when moving projects between systems and rerun important analyses before considering the migration complete.