And sometimes, later, they turn out not to be mistakes at all.
Has this never happened to you? For my part, sometimes I am mistaken, and sometimes I am even mistaken about what I am mistaken about. So it is worthwhile to keep records of failed experiments.
You can always delete information later, as a log-viewing tool might, but you can never get it back if you just deleted it in the first place.
Please consider this, git lovers, before performing your next rebase or squashed merge.
(My favorite VC quote courtesy ddaa of GNU Arch land, of all places)