When my Tactics Trainer rating was between 2000 and 2050, I was consistently getting paired with problems that were higher than my Tactics Trainer rating. This isn't just some random bias. It is skewed to where I was getting 90% or more problems that were higher rated than me. I have posted several blocks of Tactics Trainer history to demonstrate this.
Now that my rating has slipped, the bias is now in the other direction. In this block, you will find that 16 out of 25 problems are now LOWER in rating than my Tactics Trainer rating. This condition appears to be most likely when my rating is less than 1950 or so. Maybe I can get more data.
Has anyone else noticed this?
More interesting is how on earth you've managed to get to 2472 online rating when you're only ~1800-1900 in TT! Do you just spend hours on each move on online chess or something?
Admittedly I don't tend to pay much attention in online chess, but my highest rating there is 2150 or something, while my highest TT is more like 2450. Strange differences. Perhaps we have different strengths.
Below is another block of 25 problems. Now that my rating is well under 1800, the bias in problem pairing has shifted completely. Now I am getting problems that are much lower than my present TT rating, and we have a "gravity well" effect that is likely to pull me further down. I've been here before.
"Down here", I'm not getting any "rated retests". You can see that 25/25 problems are rated lower than me. Does everyone in this TT rating range see the same thing?
Here is another block of 25 problems worth of Tactics Trainer history. 25/25 are rated lower than me. Even with a pass rate of 90%, it's hard to gain rating points and maintain them.
Here's another block where 25/25 problems are rated lower. This "gravity well" effect is difficult to break. I probably will eventually break it again, but it might take 200-300 problems to do it.
Below is another block of problems. With one exception where I somehow got paired with a problem rated in the 1900's, the rest of the problems are about 100 pts lower than me. This has the effect of depressing my TT rating and keeping it down. So, 24/25 problems are rated lower.
None of these problems were "rated retests.
Well, I've seen enough. The Tactics Trainer criteria for pairing me with problems is biased, and I've got no idea why. If my rating is 2000+, it's biased to give me higher rated problems; If my rating is <1700, it's biased to give me lower rated problems.
If my rating is something like 1800+, I am regularly awarded "rated retests" of problems that I fail. If my rating is lower I do not get rated retests.
In light of all that, I reset my rating and training history. Here's what happened with the first 25 problems.
OMG! The programming has a logic error!
If you look at the distribution of tactics ratings it has huge peaks and troughs with a tiny proportion at the high rating end. This is likely resulting in a pull towards the peaks of lower rated problems.