After taking a risk analysis course, I am getting myself familiar with Bayes nets. Currently, I am looking at a common example of whether to take an umbrella on a walk. This is in the context of Norsys's Netica application, where decision nodes are realized using deterministic nodes with no dependence of the node state on inputs (deterministic nodes normally specify a deterministic dependence of the node's state on its inputs).
In section 4.2 of the above link, I was curious as to the effect of a chance/uncertainty node feeding the decision node. According to the risk course, it means that the uncertainty is resolved before the decision is made. Since it is a decision, however, I expected no conditional probability table (CPT). Inspection of the node's properties confirmed that there was none. It seemed like the input arc served no purpose. I removed the link and re-compiled. The behaviour of the net is identical to before the removal, including the result of selecting the "Sunny" state of the "Forecast" chance node.
This makes sense, but it also implies that some directed edges are simply there to reflect the cognitive process of the decision maker, with no functional effect on the math at all.
Is this the correct understanding of either (i) how Bayes nets are typically designed and/or (ii) how they are applied in helping real-world clients?