Overall, I was very impressed by the authors' anticipation of questions. While I may have found a thought or question addressed a few pages after the prompting statements, it was refreshing nonetheless. These thoughts included: what was meant by synthesis tasks, the limitations of the first configuration task model presented, and the feasibility of having a "null" or "not used" value for each variable instead of distinguishing the activity of a variable. Then again... maybe I am easily led.

      The authors describe using the "key component" idea to partially solve the problem of many-to-many relationships between functional roles and components. It seems like another form of decomposition, both function and component based. It might have been good to discuss it in this context.

      The authors mention a limitation in their algorithm. It does not handle a set of variables that may potentially become active and appear in a solution that is not explicitly pre-enumerated. This seems to be a limitation of the system or implementation, but not of the algorithm as they explain it. Are they referring to generating the list of variables at run time prior to the invocation of their algorithm? Or are they speaking of handling an ambiguous but finite set of variables which gets specified as the search progresses?