The full description of cross similarity score in here: #35.
Based on initial testing, this implementation should solve most of the issues pointed out here