Here is a quick approximation of mine, I want more powerful Elicit features to make it easier to translate from sub-problem beliefs to final probability distribution. Without taking the time to write code, I had to intuit some translations. https://elicit.ought.org/builder/CeAJFku1S