I just finished test #2. Target was CL31 Manes. Number of shots was again 600, with 153 non-critical hits. Again, auto-follow with no movement. No shots missed. Something I noticed in both trials is that the client is combining numbers of two shots together.
The numbers I saw in this test were 8 (-2), 16 (-4), 16 (-4) (CRIT) and 33(-8) (CRIT). This means that a single non-crit must be the smallest number, 8, and that a single crit must be 16. I never saw a damage number of 24 in the entire test, which implies that the client does not combine crit and non-crit shots together.
Using a binomial distribution again with probability of success, p = .9591 and n = 600 trials, the probability that we pick a sample containing 447 critical hits or fewer is 1.15 x 10^(-74).
I'm not sure what's going wrong but it's very clear that my actual critical hit chance is far below 95.91%.
I would be glad to help you look over the code, if you'd like. It's quite helpful to have a second set of eyes for debugging. I'm proficient with C++ - would just need Kyp to set up the access.