Discovery is probably the only reason why I still use YouTube. If somebody could make something that provides a non-Google recommendation system with tools to categorize/organize the videos I've watched/want to watch, that'd be great. Assuming the UI is at least on par with Google's, I'd switch in a heartbeat. Not that their UI is particularly good, but the alternatives I've seen (like PeerTube) are frustrating to use on a basic moment-to-moment level.
I feel that at some point you need to be chunking flashcards in a way that reflects your evolving level of competency. This is an extra amount of effort and this is something I'm not sure I've seen any research or even much discussion about, so I don't know how effective it might be or whether it's worth the extra time. That said, after surveying all the blog posts, research and articles I could find, I don't think it's something that's really been explored for flashcards. A certain set of static rules have solidified to which most people adhere and they seem diametrically opposed to the idea of building a lattice of deeper, conceptual knowledge. Do flashcards really need to be simple and atomic? Do they need to be answerable in a second or two? Do they need to be easily parsed so they can be instantly answered? I think there's a lot of dogma there that needs to be reconsidered.
I've seen some interesting approaches around flashcards for mathematical concepts applied to sets of mathematical problems. I've experimented with the idea where a flashcard might ask me about algorithms in a way that challenges me to apply a specific algorithm to solve a problem or think about its properties and how they affect where/when they should be used, and I feel it's more useful then your average "vocab"-like flashcard.
It's weird. They've obviously written a bunch, but there's no clear "this is what this is and what we want it to be and this is where we're going". It's all very ... academic and "meta" in a very vague way.
I'm fairly disappointed with Obsidian so far. While the basic functionality of linking/backlinking works fine, there are too many restrictions in how it works or how I can interact with it to make it anything more than a toy to experiment with. I disagree with your statement that it's "complete", since I consider it quite barebones. It doesn't help that there still isn't an API for me to build the features I need and they're unwilling to open source it.