If the component updated in Gibbs sampling depends probabilistically on the current state of the chain, how must the HastingsMetropolis acceptance probability be modified to preserve detailed balance? Under the appropriate modification, the acceptance probability is no longer always 1.