Because once it hits the ultimate character of a word, \w+ matches that (single) character, next time it matches the penultimate character, etc. You'd need \W\w+ to make it look far enough back to the beginning of the word.
So it's literally looking at the regexp backwards and forwards at the same time ๐ตโ๐ซ
The regexp itself always looks forward, the BACKWARD argument just determines which direction the point should move after a match.
Consider the string abc. From the end, moving backwards, when does it match \w+, and what does it match? When it reaches c, it matches c. And from the front, moving forwards? When it reaches a, it matches abc. This is why it acts differently.
Yes, I got that, that wasn't the weird part. The weird part is why the matcher is searching char-by-char backwards in the first place as opposed to skipping match-by-match.
I'll use "\b\w+", that seems to work well. \W\w+ was not good since it caught the spaces.
(Thanks for your patient repeated replies, BTW, I don't mean to come across as ungrateful.)
Our infinitely powerful editor.