Optimal Experiments for Partial Causal Effect Identification
A general procedure for evaluating epistemic potency in discrete settings is given, and it is shown that ID-pruned experiments are combinatorially inert, yielding a super-exponential reduction in the number of subsets evaluated.