The driving idea is that nodes with less labels are more generic,
nodes with more labels are more specialized and should be used by
executions that do not have those labels only as a last resort