It's not easy to game in that a lot of the data relies on proportionate comparisons (i.e. time spent in .py files vs. time spent in .java files or on IM). Not only that, but there'll be a level of human interaction involved. If we were going to vouch for a candidate, I'd want to at least look at why we were vouching for them myself, or have someone else audit that conclusion. At this point we're pretty familiar with (and have a lot of reference points for) what a real human usage profile looks like. To your point about the lines of code problem, really the only thing we can do is provide a filtered list of people who we'd be sure spent a certain amount of their time in the technology areas companies are looking to hire in.