That is interesting. I’ll take a good look whenever I have some available time.
I suspect the bad loo pit with the excess density around 1 might be due to the interaction between censoring and the pit transformation. I don’t remember right now which one it is but it uses \leq or \geq, so one of the bounds would behave like an uncensored bound (with the probability at the bound going to 0 or 1) whereas the other won’t, it will have a discrete probability exactly at the bound.
The good loo pit has me a bit more puzzled. There is no intuition “speaking” to me (yet?) so I’ll have to carefully check what is going on.