What I don't understand is how a puzzle gets a low pass rate, in that puzzles also have ratings which change, so as more people get it wrong so its rating goes up, and so the rating of the people trying it goes up, so they should be better at solving it. It seems therefore that every puzzle will drift towards the correct rating so that 50% of people trying it will be getting it right. Rather like my chess rating drifts up or down so I have a 50/50 chance of beating people I play. However, there are many puzzles with pass ratings below 20%. Explanation?

What might affect this a lot is that TT gives credit for partial solutions: e.g. when you only figure out the first two moves of a mate in three. It's marked as a fail, but you might actually gain points depending on the difference between your rating and the problem's.

I think it's wrong to give any credit for partial solutions, if you get the second move wrong, you usually didn't see the point of the first move. It encourages guessing, and gives really hard problems falsely low ratings (but a high fail rate), because many people guess the first move correctly.

The only thing I can do, is approve Puchiko's comment O.o

completely true, I almost commented exactly the same :)