We present two systems, CPD2-Soar and Interval-Soar, that draw on the lessons we learned during the development of CPD-Soar (see preceding paper), our earlier system for the design of distillation sequences. The first system, CPD2-Soar, depicts how we can improve both the problem-solving and learning performance of CPD-Soar. By endowing the system with more powerful evaluation functions, we show how its problem-solving strategy is improved, and, by embedding the system with a richer model of its evaluation functions, we depict how its learning is expected to be more general. We describe the task, problem-space structure and performance of our postulated system. Our second system, Interval-Soar, plays a dual role. First, it depicts how an agent can learn the conditions under which to select an evaluation function by using the strategy we postulate for CPD2-Soar; and second, in doing so, the system provides direct evidence in support of our hypothesis about learning : namely, that the richer the model an agent has of its evaluation functions, the more general its learning will be. We depict how we implemented Interval-Soar, which operates within a simple arithmetic domain, and describe its performance.