One of the most valuable tips I ingrained from an old mentor: document everything in a project. Every command run, with notes and its context. Especially valuable in ML where silent failures are everywhere. Reproducibility is hard enough