Why error-prone quantum measurements have outcomes