In the introduction SALT is mentioned. I found the idea of assessing the adequacy of the knowledge base very interesting. I wanted to know how it did this. I was patient and read on. Nine pages later I come to the SALT section. Here it says SALT "keeps track of how the pieces are fitting together and warns the user of places where pieces might be missing or creating inconsistencies." Great, I think. I understand how inconsistencies can be found but how about that "fitting together" part? What does that mean? I read on and I am soon reassured that SALT does in fact analyze the knowledge base and guides ... ensure ... complete and consistent." Hrm... ok, I get the consistency part but how does it know when the knowledge base is complete without knowing more about the domain than the expert it is getting the information from? I am then told "The next section describes an analysis SALT provides to test any knowledge base it collects for adequacy...". Great! At the end of that section it says that the system is described fully in yadda yadda reference. Very disappointing.

      Now we get into the subject of backtracking. The author states that a search that relies on local information only, when trying to resolve constraint violations can run into trouble. I couldn't find where this trouble is described nor examples of systems that use this kind of policy, like that described in Brown's Failure Handling research which is cited elsewhere in this article. Brown does not seem to reflect upon trouble concerning this approach in the article cited.

      It was good to see the problem of loops in logic, such as value thrashing due to constraint violations of inter-related parameters, addressed. This problem seems to have plagued most, if not all, of the systems that implement a form of backtracking or redesign that we've seen so far. Could this be some of the trouble alluded to earlier? SALT sure does seem like a handy thing. It's too bad the actual handling of these cases is hand coded.