Merges have to be treated with utmost caution because when they're not done right they can't be undone and create messes that take hours and hours to resolve.
A "comedy of tears" scenario is when there are no parents shown for the parents of "George". The bio may say that extensive research shows no information about parents, but who reads bios? So someone sees "George" and thinks, "I know about George, his parents are William and Anne. I would have thought they'd be added. I'll go ahead an add them."
If they'd checked, there actually is a different George whose parents are William and Anne. Now with the addition of the parents, we have a William and Anne with two sons named George. Obvious target for a merge, right! Merge is proposed, no response, completed in 30 days. "George"s profile manager is on vacation for 6 weeks and comes back to discover that the two Georges have been merged, and heaven help him if he doesn't have original records offline to help in sorting the two Georges back out..
I just approved a merge where the person proposing the merge said that "Jane" and "Joan" were obvious duplicates. Well, they weren't. I only approved the merge after looking at the profiles and ascertaining that the mother, also named Joan, was known to use the nickname "Jane" -- and that though they had different dates of birth, one profile was sourced by ancestry and the other by Richardson. Murphy's Law still says that when the merge is done, someone will choose the ancestry date of birth for the data field rather than the Richardson date of birth.
I would tend to use "unmerged match" only if I'm almost but not quite 100% sure they are really the same person. Otherwise my default is "rejected match." Because such bad things can happen with bad merges. And when I've done the research to know that two files really should be merged, a rejected match can be undone in exactly the same number of keystrokes as an unmerged match.
I don't mean to denigrate the excellent work done by arborists. But from my scanning of lists of names trying to find someone, there are enough totally obvious duplicate profiles with exactly the same data out there to focus on them rather than profiles which are "close."