Should a person's vote be weighted based on how well they comment, or how well they vote? Either could be argued, both might be tried.

Making a dataset available, even to a limited set of folks under NDA, could allow some interesting experimentation.