Consider a different recursive algorithm to accumulate the desired number of valence states in bond type assignment

[ajkerr0]

Per Section 2.3 (http://wanglab.ucsd.edu/html/publications/Atom_type.pdf):
Now the code (I believe) accumulates all of the valence states of a particular total penalty score then cuts off the array up to that number, rather than only accumulating that value.

This might be a matter of going breadth-first rather than depth-first, for example. I need to think about this more to really know what has to be done. This is a lower priority than getting bond type assignment to work full stop.