In simplified terms, it's basically taking each goalie's stats and determining how likely it is that those stats are truly representative of how good the goalie is based on the sample size (amount of shots faced), keeping in mind the phenomenon of regression toward the mean (the tendency of a variable to be closer to average on second measurement if the first measurement was extreme, and to have been closer to average on first measurement if the second is extreme) and the fact that better goalies get more starts and therefore face more shots.