I will pick one point this time:
The majority of the time, the results will go according to the line numbers, but given the randomness of each play and the general equality of the two teams the play of the line can vary. This is DefenseModerate because on this specific play one of the defensive players got a jump on one of the linemen and put pressure on the QB.
Randomness is a very bad word and a very bad concept in this game. It detracts from player attributes. PLEASE eliminate randomness!
You cannot have any game based on statistics that is not RANDOM, that is the very definition of statistics. I understand that you are frustrated with the results of one play, but the engine (and really any engine that is built on statistics) is built upon chance. The player attributes affect the chance but they cannot determine it or there would be no variation. The benefit of attributes is in aggregate. You are more likely to win because a normalized distribution will favor the opponent with the better attributes. In support of what I think you are saying, I agree that the randomness should be known a little better, and that is currently what I'm trying to do. But if your ideal is for me to eliminate the randomness then you have to realize that is not going to happen. That is not this game.
Now, a statistical basis for this game would supply all the needed match-up data and produce a static, definite outcome for each set of numbers being analyzed. No outcome would have any randomness as each would be set to specific outcomes set by the statistical breakdown of each set of data. (if x = .26 to .34 > mean, or .26 - .34 times the +STD DEV etc then gain of 2 yards (card theory).
Now, probability would provide a range of possible outcomes based on multiple variables with an outcome derived from expected differences in the data (if x = .3 then 25% chance of 1 yd gain, 50% chance of 2 yd gain, 25% chance of 3 yd gain. (bucket theory)
Neither of these are really random. True randomness would be that all possible outcomes are possible for all decisions and some range generates x result and another range generates y result. BUT when I looked into Norbert's distributions of the probable outcomes (buckets) of the player match-ups, the probabilities overlapped so that high negative differences and high positive differences both allowed the full range of possible outcomes. I argued at the time (Feb-Mar 2013) that the groupings should be tightened up to allow less range in outcomes, especially for greater differences. This may be something to consider. Even tighter groupings of potential PROBABILITY of outcomes would provide better consistency to the game.