I used to run Debian Live, and appreciated the tiny amount of shell hacking they have to layer persistent storage atop- identical to how theres a ro container image then the mutable rw layer atop it, but a decade older.

I find that the pretense of running "live" is largely gone. It's basically become a meaningless difference of what medium you run on but I've been using the same scripts to create the necessary UEFI EFI filesystem (FAT32 with a magic partition-type) & populate it with the bootloader & config... often I spin up systems by just copying a btrfs snapshot & rsync'ing efi directory (and updating some partition uuid's, regenerate machine-id), then I have two systems.

All these attempts to treat things different & distinctly, to make special cases for ourself- oh we need a network oriented OS, we meed a bare OS, we need a consumer OS- so rarely have I ever felt like these attempts to flee from the fold & venture out to special really have rewards. This behavior of making exceptions for your problem, talking yourself out of doing the easy normal base thing, seems so rarely to me to have rewards. Run Debian, run systemd, run normal well-known tools! Unless you're really certain getting odf the base path really is essential, has huge specific merit.

That's why I like this effort. It allows customization & baking your own thing, but it's much more routine & regular & normal than so many "make your own OS" projects, where folks tend to be quite "back to the land" in motive. Let's make our own OS from sticks! Yeah! But no. You almost certainly should have an ok userland base in my view. Debian is a pretty great one.

* LilDeb, a USB multi-tool designed for 256MB (like when you couldn't find a Linux box, but you could find a Windows PC), with careful separation between read-only OS, ephemeral data, and persistent data. One shell script builds the entire thing, including the various config files. https://www.neilvandyke.org/lildeb/

* Rackout, start to a living room media appliance project that was going to provide some examples for my (abandoned) Racket book. If you happen to want to use Racket to do things like repartition a disk or build a distro image, some of the libraries I wrote might come in handy. https://www.neilvandyke.org/rackout/

Ie: Do you just manually reapply your changes every time? Or just do in-place upgrade and then snapshot that?

Personally, by the second time I did this, I'd be writing a script. Then I'd want to script the running of the script.. and before you know it you end up with a framework..

This means Linux Factory is future proof.