My experience is that with out the fudge factors you get trade offs. You can increase the prediction of correct responses but only at the cost of lossing reliability at predicting non-solutions. Many times in the last 48 hours, I thought I had found the answer only to discover the new solution didn't work with other solutions.
To be truthful, I am not sure my algorithm is 100% accurate but it worked on the several dozen examples that I had on hand. I though of exporting all the permutations to excel so I could test the forumala in an environment that is reliable but I haven't had time to test that idea yet.
