The claim that one has to "reverse engineer the runtime" is simply absurd.

The people that complain about D3 on observable always turn out to have taken neither the time to understand Observable nor D3 and are looking for a way to just copy paste something into their codebase, without understanding it, and without any willingness to really learn the tools they are using. Which would have never worked in the first place, because D3 is not a plotting library like vega or observables plot, it's a library library, building material from which you can build your own bespoke visualisations.

Observable is in essence reevaluating the entire code with each "requestAnimationFrame", no fancy diffing a la react or other magic, if it runs in observable, it will also run in your code as a static thing. Copy paste the cells out of observable, prepend them with a "const" and you're done, not more work than removing and rewriting the boilerplate that's in every codepen.

I find myself wanting to start with a working example I can riff off of more in d3 than with many other tools I've used, so I totally understand the appeal of interactive examples in Observable but find it super frustrating and cumbersome when it comes to actually rolling what I'm working on into an existing application.