Let $G$ be an acyclic directed mixed graph over observed variables $V$, interpreted as the latent projection of a semi-Markovian causal DAG, and let $X,Y \subset V$ be disjoint. The causal effect $\mathbb P(Y \mid do(X=x))$ is not identifiable from $\mathbb P(V)$ in the nonparametric model associated with $G$ if and only if $G$ contains a hedge for the effect, where the hedge uses the standard $R$-rooted C-forests defined above.