Thanks for the feedback! I’m afraid that I may have given the impression that I was offering a fully-fledged design proposal. All I have is an “utopian ideal” which I threw out there to see if it inspires anyone. That’s all.
I’d just like to expand on something I had already said, but maybe not very clearly: a framework with the ability to use plugins is not incompatible with offering tried-and-true workflows out of the box, designed by experts and known to work well in most cases.
Making an analogy with Lego: you buy a Lego set and it comes with clear, well-designed instructions to build a particular model. If you want, though, you can deviate from the instructions. (You can also buy a Lego bucket with assorted pieces and no instructions; this, I agree, is useful only to a small subset of model builders).
To give a more concrete example or a potential benefit of plugins: Some people here were (are?) working on a sigmoid mapping curve. If I wanted to try it out for myself, I’d need to pull and compile a patched darktable parallel to the main version. If their code is not accepted upstream, it becomes very hard for me to keep using it. A plugin architecture avoids all these problems, both for developers and for users.