Interesting how the models seem to have been flip-flopping almost to the 11th hour on whether this or today's would be the more significant threat (although that's not uncommon). I've also been following the discussion on the upper Midwest winter storm potential associated with this weekend's system over on the American WX forum, and it appears the potential ceiling of the two types of weather increases/decreases in tandem as the models trend one way or the other vs. the last set of runs.
It seems that certain setups in general tend to be rather hard for models to get a handle on. The 5/3/1999 tornado outbreak was particularly notorious for the issues it posed for the weather models that were in use at the time (the SPC only upgraded their convective outlook to a HIGH only a couple hours or so before the first tornadoes formed), and hindcast experiments done with more recently-developed models have also shown a tendency to produce radically different results with each run--which goes to show just how challenging certain setups (particularly the more subtle ones where even a slight change in a particular parameter can greatly change how the events in question might play out) can be to predict.