Anyone who's used tactics trainer at all has seen way too many complaints about the relative ratings of various problems.  As I understand it, each Tactic has a rating that is calculated in the same manner as my rating.  So when I defeat the Tactic, I gain a bunch of points and the poor Tactic loses a bunch.  Unfortunately, this can lead to situations where a Tactic's rating can swing wildly based on winning or losing streaks, even though I must assume the Tactics don't study hard for certain days and don't stay up all night partying others.  

I propose that Tactic's ratings should be less variable, and should not really consider more recent results so heavily.  I would propose that a Tactic's rating should be more like a simple average of the ratings of all the people who have solved it vs all the people who have failed it, so that each new result will have a gradually diminishing impact on the Tactic's overall rating.  I know there are more details than that, if anyone familiar with the relevant algorithms wants to pitch in, I'd love to see that.

so we should lower the RD of the tactics so they move less?

i think this idea has potential but isnt fully developed,such as you could also take into account that persons actual rating.

The only way I know tactic's ratings have moved is from the comments.  Based on some comments, I've noticed that some tactic's ratings have swung from 1900 to 1400 and back up to 1900 again.  I'm not hip to the lingo, ("RD") but unlike player ratings, it seems the tactic ratings should gradually move to a "true" score and then more or less stay there.

Assuming this isn't a rhetorical question: -- yes, the RD of the puzzles themselves should lower as more and more people attempt them.

I like the idea of considering the solver's actual rating too.

Is this the adjustment you made?  It seems better now.  My wins/losses adjust my ratings by a dozen points or so, not the wild swings of about 50 points per try.

My rating should never have been as high as 1902, and probably not as low as 1045 within the same 12-month period. (But I'm more willing to take responsibility for the losses and the low rating, than I am for the wins and the high rating.) Now I'm hovering somewhere near the midrange in the 1500s, which is more consistent with my game rating.

Thanks for the adjustments.

I think it would be sufficient to remove the effect of time on RDs. The effect would be that RDs only decrease, never increase.

Of course I'm replying to a post from this summer, perhaps some change has been made already...

 My highest rating was 1910 and my lowest was 919. I got 919 during Christmas. Christmas+ New Year+Birthday on 19 Jaunary+ chess= bad rating