I broke out copy-and-patch into its own thing -- I think it's big enough and parts of it (especially the build-time stuff), is parallelizable.
I added PEP 554 (subinterpreters in the stdlib).
For the detailed planning parts, I've created METABUG issues where we can link to ideas and/or cpython issues related to each step. I find issue links showing the status (open/closed etc.) of the issue extremely useful for tracking progress, but those only work issue-to-issue, not in a file or wiki.
For full transparency, I'm debating how much to prioritize the memory management work (at the expense of other things). I'm leaving it here for now, but I think mainly to see if there are quick cheap experiments we can run to generate more concrete solutions.
This builds on Mark's excellent overview.