Yes, you are right, it is also the case of certain areas in CS. But then outside of the specific areas of CS (and ML which can be seen as CS) the focus is not so much in conferences. Anyway, my point is that in almost all areas of science, the focus is not in conferences.

These points seem to be pretty unique to (just) ML? Though perhaps that was the context here. I am not from ML really.

I tried to check the linked discussion, but it looks like something not wise to engage... But I guess that there are good journals to publish in, no arbitrary deadlines, you can just submit to arxiv at the same time and wait for the endorsement from the journal peer-review system?

One way to think about this might be to take a continuous time and space limit: then the Galton board is just the SDE dX = σ dW. Then a biased one is something like dX = f(X,t) dt + σ dW. Perhaps putting a suitable ∇log p(X,t) to f(X,t) would work?

That is great - what should be the correct font and font size then? If the cool kids don't use either Python or Julia.

It is indeed a good idea to have figure style guidelines. I see the width of the figure, but would it be good to also say something about the fonts? Presumably sans-serif and say 5-7 pt?

I guess that the cool kids are no longer using such legacy languages as Python? They did 10 years ago though. 😀