Yes, I am computing the percentiles on the observations generated by the posterior distribution. I am trying to do something like the Bayesian shift function introduced Here (see strategy 3). I wanted to get some feedback, and it seems it is reasonable 