> (1) The performance system specifies the agent’s capabilities at a fixed point in time – what it could do without further adaptation.  The three basic elements of the performance system are a set of detectors, a set of IF/THEN rules, and a set of effectors.
>The detectors represent the agent’s capabilities for extracting information from its environment, the IF/THEN rules represent its capabilities for processing information, and the effectors represent its ability to  act on its environment.  For all three elements, the abstraction looses the details of the mechanism employed by the different kinds of agents.
>Our framework concentrates on the information produced – the properties of the environment to which the agent is sensitive.  We exploit the fact that any such information can be represented by a binary bit string, here called a message.  We gain the ability to describe , in a uniform way , any agent’s ability to extract information from its environment.  Defining the performance system’s ability to affect the environment in terms of message-sensitive effectors entails similar losses and gains.
>The same considerations hold for the agent’s ability to process information internally.  The mechanisms are various, but we have concentrated on the information-processing aspect.  By conjoining IF/THEN rules with messages, we wind up with rules of the form IF (there is a message of type c on the message list) THEN (post a message m on the list).
>In so doing, we loose the details of the mechanisms used by particular agents for processing information.   For example if we are studying the progression in which genes are turned on and off in a developing embryo, we loose all the fascinating details about the particular mechanisms of repression and depression.  We retain, however, the description of the stages of development, and the information fed back at each stage.  In general, we gain the ability to describe any information-processing capability that can be modeled on a computer.
>Because many rules can be active simultaneously, we gain a natural way for describing the distributed activity of complex adaptive systems.  In particular, systems with this parallelism automatically describe novel situations in terms of familiar components; internal models, in the form of default hierarchies, which form naturally.  Bother activities are pervasive in cas.
>Once we settle on a rule-based description of performance, the process of adaptation provides components (2) and (3) of the framework.