If X is a countable state space and g is a bounded reward function on Xn, then say g has the Markov-adequacy property if every strategy has a corresponding randomized Markov strategy which gives g the same integral as the original strategy. A complete characterization of functions having the Markov-adequacy property is given. In particular, if g is permutation-invariant and X has at least three elements, then g has the Markov-adequacy property if and only if g has the linear sections property, a condition which is easy to verify.
