I made the mistake of mounting a btrfs disk in two places at once - in a VM and on the host (libvirt with LVM). The system was in this state for around a day, though the host didn't write anything in that time. After seeing my mistake, I unmounted on the host and ran a scrub in the guest, which returned 90 or so uncorrectable errors. I then booted into a livecd and ran btrfs check. I was fairly sure my data was hosed at this point, but I had nothing to lose as I had backups, so I ran btrfs check with --repair, which coredumped and seemed like a bug.

btrfs-progs version: 6.7 static
logs:
kmsg.log
scrub.log
check.log
check_repair.log
coredump (2GB): https://mega.nz/file/BihyDSRJ#VF1koGKi5H_pICf5Rsc5He9WyWQe4_Gtya1Y6_SQu_k