You’re under-thinking it.

In pseudo-correct but probably not order:

• Step 1: Collect underpants
• Step 2: Keep receiving Google security updates but stop updating Google mainline
• Step 3: Start replacing the underbelly to just raw Linux (or BSD or whatever) and slowly shift the “Android” portion to a VM/container
• Step 4: RIL and other stuff (probably should happen first) have to be packaged up and become their new entity on the modem side (also probably the biggest challenge, but manufacturers and ODMs provide dev kits)
• Step 5: ???
• Step 6: Once the Android side is safely firewalled away from the core OS, start embracing something like PostmarketOS
• Step 7: GUI/graphics are built out with the Android pieces still running in a container
• Step 8: Start writing applications that replace the Android applications, go one by one, remove dependence on each Android application as you go while still maintaining compatibility (I mean the core OS ones that make the device at least basically functional, the F/OSS devs will have to each rewrite/change their apps, or some other magic can be inserted here that isn’t really magic.)
• Step 9: Once the OS itself is beefed up enough, retain Android container for the needs of some for some uncomfortably long frustrating time to maintain, but not too long
• Step 10: Have Obtainium/F-Droid/etc. all simultaneously pivot and start providing apps for the native OS as well as maintaining backwards compatibility with the Android apps in the container
• Step 11: Once some magic point, forced or otherwise happens, sunset the Android portion of the app stores. Keep the containerized Android around a little longer
• Step 12: Sunset the Android container, at this point the phone should be running 100% “native” OS and apps and store
• Step 14: Profit!

There are industry blueprints for this. Apple is probably the best example of how to implement these shifts, from OS 9 (co-op MT proprietary OS)->OS X (BSD-NextStep-based Unix OS), 68k->PPC, Replacing Unix underpinnings with Apple Frameworks, PPC->Intel, OS X->iOS, Mac from Intel->ARM, etc. etc. They frequently used containerization to keep the old running while the new was built up around it and replaced. It is a solid proven design pattern.

And edit72: I’m not just saying “hey magic people do this” - I’ve done this shit. I’m down to help, and I will. But the project owners need to step up for some actual work instead of just putting potpourri on something someone else built. Annoying side-story, I figured out how to cross-compile/rebuild/fix dependencies on a CPAP app called Oscar so it would be ARM-native on ARM Macs. Couldn’t figure out how to contact the devs after much digging to let them know, so. I have 1 of 1 copy of that app running ARM-MacOS native. Would be neat to help them replicate it though.