Bach: Yes, I understand what you are saying about Q=60. It actually explains why CCEGuesser works in the first place. My point was that CCEGuesser only claims to be able to predict the size, which it does. It does not offer any theories as to what is the best way to encode. So, I took a small bit of offense when you said CCEGuesser's algorithm is "wrong". Its algorithm is actually, in contrast, pretty accurate for what it does. I think what you were actually saying was that encoding at 60 is wasteful. Right? Regarding the straight line theory, I still plan to collect data. All curves are made of little straight lines, right? Who knows, maybe we'll find that within the workable Q range (say, 20-60), the line is always relatively straight. It's certainly worth exploring. On with the quest to faster encoding! PS thanks again for your "sample" idea. It's a great one.