I wonder if you could cut this down by focusing on know implementations and their random number generation. That is trying to figure out if weaker random number generation at any point was prevalent...