Toward a Methodology for AI Architecture Evaluation: Comparing Soar and CLIPS