[epistasis]

Fantastic, thanks for such a detailed response, it is very much appreciated and I learned a lot! (And you don't come off as contentious at all!)

I think your use case makes a ton of sense for Marimo, iterative modeling and mucking around. Most of my modeling has been less like that, and goes straight to a single computation with the result. That is a significant workflow difference.

The jupyter lab web interface has many many UX problems that make it extremely difficult, as does the whole kernel launching infrastructure. Having a desktop app, or another way to launch kernels in a daemon like mode without keeping a terminal open, could save a ton of headaches.

Marimo is great in that it accomodates pure functional programming fantastically for notebooks. Unfortunately too much of the Python API ecosystem does not, but that doesn't mean that it couldn't in the future.