Computational Notebooks Are For Teaching, Not For Learning

Pretty much everything I have to say on the topic is encapsulated in the title. Whether it be through Quarto + .qmd or Emacs + org-mode, the overhead is significant. Many of your experiments do not need to persist. Also, when programming, the ratio of code to the “literate components” is high enough to justify the exclusive use of comments for such components.

Learn using scripts, running those through your preferred IDE. Only at the point of publishing (and possibly subsequent teaching through live demonstrations) should you consider using computational notebooks. By the time you use such a notebook, you must already have a good idea of the outcome.