Nice. I've really felt the pain point here. In R I use Drake to run a big pipeline on my laptop. In theory that provides the same "only run the computations you

Nice. I've really felt the pain point here. In R I use Drake to run a big pipeline on my laptop. In theory that provides the same "only run the computations you need" DAG, but because outputs are defined as whole data frames, changing or adding a single column often forces recomputation of everything. Column level isolation like this is the smart way to go.