Exclusive: Corticon plays by different rules

Dr. Mark Allen of Corticon caused quite a ruckus several years ago when he published a paper called “Rete is Wrong,” which took all of the rule-based engines based on the Rete (pronounced Ree-tee) algorithm to task for inefficiencies and poor construction. Allen explained that, in contrast to the Rete engines in market-leading BRMS (Business Rules Management Systems) such as ILOG’s JRules and Fair Isaac’s Blaze Advisor, Corticon had a DETI (Design-Time Inferencing, pronounced Dee-Tee) engine.

The Rete algorithm was developed in the late 1970s by Dr. Charles Forgy and, at that time, increased the performance of the standard LISP inference engines by as much as 3,000 times over an engine that did not use the Rete algorithm. I believe that rule engines based on Rete continue to be much more scalable than alternatives such as Corticon, though I have yet to find a performance test that allows meaningful comparisons across the different technologies. Nevertheless, Corticon has some advantages over its Rete-based counterparts for many business applications.

As Dr. Allen and Corticon claim (and as I observed for the most part during my testing), the DETI engine does most of the complex checking of the rulebase during design time rather than at run time. This removes the problems later on that can be caused by circular reasoning, duplicate rules, inconsistent rules, and incompleteness of the ruleset. The only thing you lose, compared to a traditional Rete engine, is the flexibility to drop in a rule or change a rule on-the-fly. According to Dr. Allen, however, very few of Corticon’s customers need that capability, and I’m sure that is true. In the business world, most applications obey spreadsheet logic, and Corticon shines when the rules are easily expressed in spreadsheet form.

Setting new rules
Corticon takes a different approach to problem-solving than the Rete solutions, making the traditional benchmarks (including Manners, Waltz, and WaltzDB) inadequate for assessing its power or speed. Corticon performed quite well in the limited tests I was able to run, but how its performance and scalability stack up against Rete competitors remains unknown in the absence of a benchmark that tests Corticon and Rete engines on a level playing field.

The Corticon 4 BRMS includes the rule execution server, a modeling studio, a rule management server, and a new enterprise data connector. The modeling studio, shared by developers and business users, provides a spreadsheet interface where the rules are initially entered and checked for completeness, goodness of fit, conflicts, duplicates, subsumption (when a rule is so broad it contains another rule), and other problems. The rule management server, called the Rules Collaborator, provides version controls, approval workflows, and role-based access.

Corticon simply excels at rule building. The GUI is a bit clumsy, and will take some getting used to, but Corticon’s spreadsheet approach to development easily beats the similarly spreadsheet-oriented decision tables in JRules and Blaze Advisor. If I had to use a spreadsheet approach only, Corticon is definitely the way I would go. Business analysts will not only get used to it, they will love it.

The modeling studio has some really neat features, including checks for rule completeness, ambiguity, and subsumption, but users will need practice before some of them become intuitive.