Predicting the expected behavior of agents that learn about agents: the CLRI framework