I did the thing you are not supposed to do. I ran a big upgrade on my workstation last thing on a Friday, with plans the next morning, and watched it leave the box in a state where the display manager would not start and a graphics driver and the running kernel had a difference of opinion. The old me loses Saturday to a reinstall. The current me had taken a btrfs snapshot before the upgrade, almost out of habit.
So instead of debugging at midnight, I booted to a console, rolled back, and went to bed. The whole recovery was a couple of commands.
btrfs subvolume delete /mnt/@
btrfs subvolume snapshot /mnt/@_pre_upgrade /mnt/@
That is the entire trick: I keep @ as the live root subvolume, and before anything risky I take a read-only-ish snapshot of it. Rolling back is just deleting the broken subvolume and promoting the snapshot back into its place, then a reboot. No reinstall, no fishing config files out of a backup, no trying to remember what I had customised.
The honest lesson is not "btrfs is magic". It has its sharp edges and I have been bitten by its quirks before. The lesson is that a filesystem-level snapshot is a different kind of safety net to a backup. A backup protects your data. A snapshot protects your state, the whole working system, and it can put it back faster than you can type the word "reinstall". Five seconds of forethought on Friday bought me my entire Saturday. I will take that trade every time.