These are good points.

OP, your workflow sounds very similar to the one suggested by Barbara Oakley with a dash of Feynman technique.

I would also recommend trying Papers with code as sometimes looking at the code makes more sense.