You are confusing metadata stored in a file, with metadata stored for a book in the calibre database that you see on screen. Think about how you ended up with binary duplicate files in calibre in the first place - how did they get in there? Either the books had the same title and you ignored the warnings given by calibre. Or they were originally the same, but you added one to calibre, slightly tweaked something about the author/title, then added the other book and didn't get warned. Or the book was completely mislabelled when you added it, and its content isnt actually what the metadata describes it to be.
These were likely added over time. Maybe you started cleaning up your library for a book - you downloaded metadata for it. Maybe on one of those rows you did a conversion to another format. All of these actions are going to change calibre's database metadata about that particular row in your library, but none of them affect the actual book content that binary comparison is comparing (not until you do a conversion at least).
So calibre now must have at least two rows each containing the same physical file. We agree one of them has to be deleted. But it is *impossible* for the plugin to guarantee to know which one it should keep. It might delete the one with the "wrong" author name/spelling (it cannot know which is "correct"). The plugin cannot know which cover you might prefer (perhaps one was read from the file, and another came from a metadata download). Maybe one has series information, and the other doesn't (or even has a different series name/index).
There are just too many variables that only a human eyeballing the two can make a decision on. So it ain't gonna happen
