VEXED allows backtracking and saves only the steps of a plan that were not retracted. This means that there is no capture of the mistakes made during the design process. BOGART allows steps to be saved even if they are retracted. The design plan that contains the later retracted steps appears just as any other plan saved in the system. While I think it is good that BOGART adds this saving functionality, it is foolish to not capture the that steps in the plan might not be correct. There is much useful information in knowing where plans can make mistakes and BOGART does not take advantage of this.

      It wasn't clear at first, but it seems BOGART uses the idea of correspondence to decide whether a branch of the design plan should be replayed or not.

      BOGART seems to be primarily a Case-Based Reasoning system without the reasoning. It stores cases, design plans as a tree of rules. The further down the tree, the less abstract the rule are. It seems many of the advantages this system could provide are left to the user. It does seems much better than making the user choose each rule that gets applied as in the pre-BOGART VEXED. The main flaw seems to be that the more design plans one has in the system, the more useful the system becomes. The more design plans there are, the more work it is for the user to sort through them to pick an appropriate one. This ultimately limits the system (as described here) from ever being truly useful.

      Most of the circuit diagrams didn't seem to add anything to the article and could have been excluded. The same goes for the benchmarking section. Measuring user effort seems like a whole separate topic for research. While it was slightly interesting to see some measure of how BOGART improved upon VEXED, far too much detail was given considering the flaws of the chosen measure.

      Big points for the article mentioning CRITTER... Ooo, CRITTER.