That. Was. Awesome.
On the probabilities... the categorization is subjective, yes. It is also inconsistent across years. Maybe you could take out some homerism by utilizing Vegas Odds for each game.
Although - thinking out loud here - the previously played games in a season would skew the model toward the end of the season, with the goal of the model being to predict win totals for the season before it starts.
Since I doubt that there are odds for every game prior to the season starting (maybe there are? JamieMac?), would it be possible to utilize the 'odds' that the teams on the schedule will win the championship? Like if Michigan was 30:1 and Toledo was 100:1, that could be used to determine a 'winner probability'?