Due to the recent update, I have found a way to make this work the way I explained without coding it in. However, it should only be used if you have a high % match and are certain everything will be deleted. The following explanation would allow you to go through a large collection of songs you plan on listening to and deciding what you will and will not keep. Since you have no preference and no folders for the group function (or too many folders), this is the way to delete all duplicates without accidentally deleting all of them:
Rearrange your group bases (the bold files that candidates go under) to whatever you prefer. Most likely size or bitrate. Next, get a macro recorder program and loop the keyboard commands "home" "down arrow" "delete". The result is that ONLY candidates can ever be deleted from the results. Home will always position the selected row at the very top of the list, and the 2nd entry will always be a candidate until all candidates are gone and the group base is all that remains, at which point it is cleared from the list automatically. This process is important, because similarity constantly updates the list, deleting matches to that file that was deleted and thus avoiding inverse relationships. This means that it will churn through the group bases 1 by 1 automatically, making the list refresh and guaranteeing that you do not delete all copies of a file, without using the grouping function.
The only problem that persists is when I have several dozen preferred folders, for which the group function will work. Yet, the results will be mixed as to what folders serve as the group bases, so you couldn't automate that process. However, as soon as the development team lets you give folder preferences to group bases, this will be possible.
