Honestly, depending on whether you count it or not, LFS. I have not tried Gentoo yet, though I want to one day, for the learning experience, and yet I already know that compiling everything is not something I enjoy.
I can get by with OpenSUSE and Void (kinda), I've used Debian for a few weeks, I've used Fedora for a month or so, I've used Ubuntu for a bit, I've tried PopOS for a week or two, I've used NixOS for a few months, and I've used Arch for most of my time on Linux.
Currently I'm on Arch, but I don't like rolling releases that much. At the same time, I am also not a fan of immutability, as there are some programs I need that cannot be installed on an immutable distro, so that's why I'm on Arch. Why am I only using these 2? Because they are the only distros that have all the packages I need (excluding the specialist software that I need for university). By the time I discovered Distrobox (which would solve this problem), I was already on Arch. I've also done some changes to my setup and as such, I'll need to wait for some new features to make their way into program releases and into the NixOS Stable repo with the following release. Until then, I'm on Arch.
Gentoo is useless for learning how things work. Back in the 00s when I still had time to hang out at events it was always quite ridiculous at what kind of basic stuff the gentoo crowd got stuck at - and with the tooling 15+ years more polished now I'd expect what is actually going on is way more hidden than back then.
If you do want to understand how things work just build a minimal system - either on spare hardware, or qemu/kvm. Don't go with systemd, or other fat userland options - that just makes you compile a lot of dependencies not adding value for learning.
Use some lean init (or just write one yourself), and some lean shell.
I'd just start from very simple kernel and static init, and work my way up to adding more functionality. I'd use kvm with rootfs on p9fs - that allows playing with it without having to build images. I can throw together the initial invocation, if you're interested.
Then start building simple core elements in a language allowing easy static linking - I'd use C with dietlibc or go. Start adding core userland programs, explore initramfs (without using something like dracut), add dynamic libraries and explore the dynamic linker, ... - if you're interested we could set up a matrix channel for questions (typically with some lag, though), and do a github repo to follow along.
LFS iirc goes for full desktop - the high level userland is very complex, but easy to understand when you know the basics. You pretty much learn how to compile lots of libraries - which has limited use. A full LFS style desktop I'd no longer recommend nowadays - it's just too many dependencies to deal with. I used to build my own system (not following LFS) until the Xorg fork made it sigificantly more complicated - and things got just worse since then, and I never was using a complicated UI stack.
edit: I had a few minutes, so I've thrown this together https://github.com/bwachter/lll - you should easily get a kernel with a custom init running, and have enough to start experimenting. If you or anyone else is interested to go deeper I'll set up a matrix channel for guidance.