That's some very old data. I'm curious as to how stuff have changed with all the new advancements like helium drives, HAMR, etc. From the stats Backblaze helpfully publish, I feel like the huge amount of variance between models far outweigh the importance of this specific stat in terms of considering failure risks.

I also thought that it's "URE", i.e. unrecoverable with all the correction mechanisms. I'm aware that drives use various ways to protect against bitrot internally.