For this version, I have added the offensive and defensive components of each team's ranking (oGPF and dGPF). I can derive these by looking at the over/under in combination with the point spread.
I modified the methodology slightly relative to last week. I switched to a weighting based on weeks elapsed as opposed to days. I also modified the weights to favor recent results more heavily. The modified weights were based on back testing against prior seasons. The model still has a somewhat large prediction error of about 3 points. I suspect that with the wide dispersion of talent among Division 1A teams, my simple linear approach to team strength may be breaking down at the extreme ends. In contrast, for the NFL, I've been able to get the error down to about 1.8 points on average.