It need not be a "merge" as such, as long as we are granted the ability to mark something as a dup (and mark what it's a dup of). Everything else can be done algorithmically; I outlined some ideas for this in an Ask HN last year: http://news.ycombinator.com/item?id=1975950