Funny, I actually did the same thing for python years ago and also talked about "Lossless Syntax Tree" then start calling it a Full Syntax Tree since CST doesn't really fit; I wasn't aware of the existence of lib2to3 at that time.

My goal was different than your: I wanted make writing custom refactoring code (that's it: code that modify source code) and code that works on source code a do-able task.

I end up doing some design decisions that I haven't found elsewhere (but this field is hard to explore):

- producing json, because datastructure doesn't lie to you and potential interoperability

- nodes are responsible for the formatting within itself, in opposition with lib2to3 where a node is responsible for the formatting before itself (or after, I'm not sure anymore)

- the tree is design for the human brain instead of an interpreter/compiler (for example having list instead of recursive structures)

The project is called Baron https://github.com/pycqa/baron and was actually a mean for me to work on what really interest me: the abstraction that attempt to make writing custom refactoring a doable task https://github.com/pycqa/redbaron

Good luck with your project :)