1. The rating distribution of problems is multimodal. To see what I mean, go to your tactics trainer page and click on "view problems".  The rating frequency distribution has detectable modes at approximately 1200 1400 1600 and 1800.  If the Glicko rating adjustment algorithm assumes a normal distribution, this is a huge problem.

2. The limiting Rd values seem to have been adjusted upward. i.e. you gain more for solving a problem quickly or lose more for failing one completely.   The limit Rd values for problems should be lower, say around 7 after they have been worked 1000 times.  Not sure what to suggest for players.

3. The algorithm for pairing players is now biased downward. In the last 50 problems, maybe it's more like 100, I've had ONE problem higher than my TT rating.

Maybe it's time to hit the "reset" button.