I found this article by the former administrator of this forum.
The article is very informative, and really answers many of the recurring questions that people have about various aspects of scoring.
The basic idea is to get matched sets of anonymized reports, typically spaced at a two-year interval, and then determine what factors in the first report, along with factor weights, are the most useful in predicting what shows up in the second report for the same person a few years later.
The article also talks about how people are separated into various buckets or score cards, and this helps to explain why it's impossible to have the same rules and weights for everyone.
http://blog.credit.com/2013/02/how-credit-scores-are-developed
