The Bradley-Terry method looks to create a model of actuality based on history. Given the nature of KRACH (the Bradley-Terry method), I say it is a far better choice. The PWR still has "arbitrary" aspect to it, namely (as pointed out) 25 teams, not 20, not 30, not all, and it has RPI which is an arbitrary formula (made up of negotiated percentages of your, your opponents', and your opponents' opponents' winning percentages).
My thought (and I'm about to contradict myself because I'm going to put some "arbitrary" into my plan): Let's run two KRACH calculations after the conference tournaments. One would be the currently generated full-season KRACH. The other would be the from the end of the season back to enough games to make the analysis statistically valid (that'd probably work out to be "since Christmas"). Apply a weighting to each (oops, sorry, arbitrary) and seed the tourney from that.