The problem is half the ensemble members are time lagged so you're getting 50% of that mean coming from 12z runs and 50% from 00z runs. When there's a dramatic shift the utility of the HREF is definitely more limited.
FWIW - the 18z GFS, 18z NAM (12km and 3km) still show excellent snow growth for BDL/HFD. Even on the QPF gradient the efficient dendritic growth could help.
Here's our first call. Only thing I don't like about the 18z GFS is that it brought the 700mb low a bit farther north. Have to watch that south of the Pike I think.