It's a mix of Claire et al's ideas and my own. A distinctfd constraint takes a list of vars and creates the real distinctfd constraint for each var, these are p

It's a mix of Claire et al's ideas and my own. A distinctfd constraint takes a list of vars and creates the real distinctfd constraint for each var, these are put into the constraint store. They do not run unless the var reaches a single possible value. At which point they run once and are purged from the constraint store. Each constraint has a list of the vars involved plus the list of known single values for the other vars. When it runs if the constrained vars' value exists in the single values set it fails. If it doesn't, the constraint removes the value from any other var that has a domain of values. It's actually pretty simple. Most of the performance is from Clojure, the purely functional design which gives us backtracking "for free", and the random access nature of the constraint store.