The 'don't hold it like that' strategy? Quite possible. You can't have everything. Still, this behaviour was new to me and it is costing me valuable time, it's good to have it logged as a possible improvement.
It would seem that is simply a matter of regenerating a host family when a type property of a nested shared family changes (or only the ones connected to formula's, if you want to optimize), just like hapens when you change a type property for the host family outside the Group Editor. The formula ensures the instances are still identical, so the Group definition does not need to be affected any more then when you change a Type property of the host family.
O, by the way, I'm now seeing the same in a generic model. For now, I'm dropping my corrupt family template hypothesis.
My next try at a solution is along the lines of your advice @Alfredo_Medina, no more Type parameter in the nested family. But, I don't want too many parameters in the top level family, and no irrelevant parameters in alternate shared nested families. Let's see if Default instance values hold up to the daily grind.
