Crick, my code must be completely unclear. Oh well. The paper describes snakes; my implementation makes use of "downsnakes" and "rightsnakes" as references to the edit graph, with a "downsnake" being a single downward move followed by a snake and a "rightsnake" similarly defined.
The paper describes a graph traversal method which I eventually realized was suspiciously like Pascal's Triangle (tilt your head to the left 45 degrees and squint ^^); hence the reference to "start the triangle". Basically instead of + as in pascal's triangle I have (minN (rightsnake left) (downsnake right)), while the leftmost and rightmost nodes are generated via downsnake and rightsnake, respectively, instead of the identity function as in Pascal's.
