The parallel-experiments-in-replicate-kernel part is exactly where I want to stress-test it.
I lose the thread fast when I do it by hand – three branches, each with slightly different preprocessing, and a week later I can’t really tell which dataset version + seed produced the model I kept.
So when Clusi branches and runs experiments in parallel, does each branch pin its own environment and dataset snapshot, so that the results remain reproducible even weeks later?
And when I choose a winner, can I cleanly merge that branch into the main notebook – or does it remain as a separate artifact?
<a href