Either they are in metadata, though usually "metadata leaf" or similar is mentioned in those cases, or they are related to now empty extents, in any case recommend backing up and re-formatting the pool.
This is an example of you it usually looks in the log when a scrub finds data corruption:
Tower1 kernel: BTRFS warning (device md1): checksum error at logical 1006087647232 on dev /dev/md1, physical 1005022294016, root 5, inode 512268, offset 16712871936, length 4096, links 1 (path: plots/plot-k32-2022-01-03-12-03-d59a0e1f87141f9355fd42074dd671c706152c741767e01bb52946991f4a9e59.plot)
Tower1 kernel: BTRFS error (device md1): bdev /dev/md1 errs: wr 0, rd 0, flush 0, corrupt 1, gen 0