Best Practice for Cache/Processed Data Management

Follow the full discussion on Reddit.
In ML research, we often have tons of cache files to read from disk. As a project progesses, I often end up with lots of cache files which I don't remember how they were created. My current way to keeping track of cache files and preprocessed files is by writing it down when it is created and why.