Hey there,
Awesome. Makes me feel better knowing it’ll help somebody 
As to your second question, while I’m pretty familiar with HTML/JS, I assure you, I’m not using that here lol. To be honest, I’d be suprised if discourse would even let me inject Javascript. Seems like it’d be a bit of a security nightmare.
Instead, I’ve just been using discourse’s built-in tool bar, here:
Many of the more interesting formatting options are found under that right-most cog wheel. As to the information in the post, I generally draft that up in a Google Doc. Thankfully discourse is pretty good at translating clipboard styling, so things like tabling and text format are pretty easily copied over without too much additional work on my end.
For more involved diagrams, like the ones found here, I use draw.io.
Lastly, I find spinning up a project in something like ChatGPT/NotebookLLM/Claude to be helpful. For TBP, I have their sensorimotor intelligence paper preloaded, in combination with maybe 7-8 working papers on my own policy recursian project. This lets me quickly bounce questions against the framework and get (generally accurate) responses. I find this most helpful in simply being able to have my thoughts transformed and relayed back to me. It forces me to reframe the various problem-spaces I’m working in and view things in new ways. This is actually how I arrived at this answer here.
If you want to see what this process looks like, heres the conversation chain I went through in order to arrive at my conclusion in that post (about a three-hour long convo): https://chatgpt.com/share/67b3a239-13ac-800f-a511-4c01e4a566f7.
When using a project-type approach like this, I find it best to use some kind of persistant knowledge-base (eg. that sensorimotor intelligence paper), followed up by additional contextual information loaded in via attachment. Thats what I did with some of the codebase files in my aforementioned example. The reason for doing it this way, is that the bot tends to get overwhelmed when you load everything in as a projects file, queries then take forever, and the resulting answers tend to drift towards non-coherency.
Additionally, I tend to keep an external library for papers which I find useful and/or interesting. Then I can load them in as need be. Heres an example of an ML/Neuro-inspired one: Educational Resources - Google Drive
The most important thing to be aware of when using any chatbot, however, is in understanding that they can (and will) BS you. I find that if you understand about 20% of the content of the field for which you’re engaging, you can (for the most part) protect yourself from this.