In noisy environments, a state with high average fidelity does not mean every sampled pure state is good enough; the distribution may simultaneously contain samples with extremely high fidelity and samples with extremely low fidelity, and the average masks this divergence.