Benjamin, I'm also excited to see more probabilistic programming in Haskell. Adding to Sergiu's questions I'd like to know why you chose to implement a new interpreter instead of making an embedded domain-specific language. Are there C++ libraries that were convenient to use? As theorist I'm curious whether you have a denotational semantics in mind. Can you decribe the building blocks of your language? The examples show primitives for known families of distributions (e.g. beta). Then there is Observe, which seems to compute conditional distributions. What else is there beyond ordinary Haskell? Cheers, Olaf