I think having docutils as a backend parser will make the experience more resilient + make myst easier to integrate. But I am not all the way through the APIs yet.

I have a new branch bootstrapping at https://github.com/tony/pytest-sphinx/commits/docutils-parser. Super preliminary.

I notice now you've already attempted this approach: https://github.com/thisch/pytest-sphinx/tree/docutils-experimental.
What made you pause on this approach?

Both look similar we're both using sphinx's doctest.py directives. We even have the same conditional (except you used .traverse() and I use findall()), also I'm looking back at pytest's doctest.py