But this is possible to do trivially using a hash keyed by the input variables. It seems to me that SAC offers an interest only in the case where: - the input

But this is possible to do trivially using a hash keyed by the input variables. It seems to me that SAC offers an interest only in the case where: - the input variables of the DAG are updated at a different rate - and the result of complex subgraphs depend on only a few of these variables (and thus avoiding to recompute their result is a win) I don't know if it's enough to get excited about.