When you edit many files where there are natural pairs (for instance, interface + implementation pairs when dealing with C source code) I tend to get lost in the maze of open documents and often end up doing a project-wide search or similar to find my way between the messy graph.
So I got this idea that you could group documents (or rather sort documents) with respect to their Levenstein distance. I implemented this into an application I’m working on and it turns out to be a really neat feature. Here’s a quick demo screencast: