Tim, Have you read the book Blink -- lots of evidence of collars in many other domains. In re-reading the paper, it appears that the major contributions of this version are these: 1. collars and clumps - definitions - theoretical evidence - experimental evidence 2. Treatment learning - explanation of the TAR2 algorithm - explanation of Jane and its operators 3. Connection between treatment learning and collars and clumps 4. Case studies as examples of the utility of tar2 and verification of collars and clumps. 5. Claim that tar2 has some significant advantages over other data miners. It appears that the biggest problem is that the paper seems to make claims about the generality of this approach (treatment learning) that did not convince the reviewers. One part of this is that many seem to equate "UML" with "SE models" and we had no such UML examples. It was not clear to the reviewers that tar2 could apply to UML models. Here are some of my questions: 1. Would it be better to break this paper into smaller pieces, or to try to strengthen the current approach, or to carefully back off from the claims of generality but still cover all of these same topics 2. clumps and collars - are these defined is a sufficiently rigorous way - is this a new contribution - could a paper be written whose only thesis is that many models have this structure. 3. TAR2 - has the algorithm been published sufficiently, or could there be a journal paper describing its functionality. - is it possible to describe and justify TAR2 without talking about collars? - has the JANE language been published separately in a journal? Jim