I looked at the graph, and am quite surprised as well. Perhaps TT shouldn't assign an initial rating, but make it unrated for the first ten tries and count a rating from those. Of course, it would be reasonable not to count these unrated attempts into the daily maximum of problems (basic members can only do three problems per day). The current system doesn't look very accurate.
tactics trainer problem ratings
Sort:
Shouldn't the tactics trainer problems converge to their "true" value if the parameters are correct? It is my understanding that the problem rating changes as people get them right or wrong. Essentially, the player is playing against the problem and the ratings are adjusted. However, if you look at the problem ratings distribution (http://www.chess.com/tactics/problems.html) graph you see that there are narrow gaussian distributions around 1200, 1400, 1600. I assume that the problems are given an initial rating of 1200, 1400, or 1600 and then they move only slightly. It seems that this indicates the problems are not being moved enough to get away from their starting point.