I think it depends on the agreements Bufferbox can get with the buildings. I'm not familiar with these types of agreements (eg Redbox kiosks, vending machines may be similar), but I can imagine that non-competes with multi-year terms are a possibility. Especially if Bufferbox itself will pay for the initial costs.
If that is the case, then Bufferbox is in a more defensible position.
While this may not be applicable to goodfilms, a 2-dimensional system is harder to brand. As in, "Two thumbs up," or "90% on Rotten Tomatoes." There's value in a single understandable number.
I wonder if there's a measurable value for the system. I care about discovery, so I want a site that can recommend me movies that I wouldn't normally think about. How many more movies would this system recommend to me?
A lot of times, I don't care too much about accuracy as long as the system isn't too far off. This is simply because the cost of an inaccurate recommendation isn't too high when I can stream it on Netflix.
I like the idea of providing more data for people to make more accurate assessments, but I don't necessarily believe optimizing for accuracy optimizes the value provided to the user.
At this stage, I don't think it matters if Bob cheated because Udacity can't even verify Bob is actually Bob. It does seem strange that the post doesn't talk about the logical next step, physical testing centers that can verify identity, especially since Udacity already announced their first partnership (http://udacity.blogspot.com/2012/06/udacity-in-partnership-w...).
If that is the case, then Bufferbox is in a more defensible position.