Using a minimum of knowledge available locally about the global state of the system is motivated by human memory limitations. Why limit the system in the same way a person is limited? Shouldn't these systems help people overcome their limitations? I see several possible reasons. It is far easier to access this information from a local point of view. This is apparent from the now long-lasting popularity of object oriented styles of design/programming. It is easier to facilitate communication between components if messages are only passed between incident components in the hierarchy. By only considering locally available information, the changes, if any, resulting from failure recovery are more likely to stay local. Minimizing the changes and scope thereof in redesign seems to be a fairly accepted heuristic. Of course, the very best point I could think of was already offered in the reading. That is, the belief that much can be learned about the behavior of agents in a design system by considering it a person in a design team.

      I was surprised when MOLGEN was not mentioned in reference to means-ends analysis to handle unusual failures not covered in the context of a routine design task.

      It seems using "knowledge of situation-action associations" in a Failure Handler is just a fancy way of saying forward chaining. Is there a great cause for distancing this discussion of failure handling from that of general AI techniques such as forward chaining? Is it possible that I may be reading too far into a certain choice of words? Speaking of choosing words, finally some humor in the readings, that I understood anyway. The "+++" notation did confuse me for a moment though.

      The section on Redesigners and the limited number of basic strategies used during redesign seems to serve as a prelude to Klein's whole take on conflict resolution. That conflict resolution, which redesign attempts to do, expertise is general and applicable across many domains.