You are emphasizing the truth-values at the nodes of the belief network (“check back to Q and P”). That is important. After all, in the end, you do want to have the right truth-values in the buckets.
But there are also structural questions about the underlying graph. Which edges should connect the nodes and, perhaps more deeply, which nodes should be there in the first place? When should new nodes be created? These are the questions addressed by Phil’s and Anna’s posts.