Decision nodes do implement reference-like behavior, but they don't cross branches. Therefore, an operation such as,\"connect this node to the first branch of the next node\" only operates as a reference. This means the two nodes cannot cross a branch in the decision tree.
Decision Tasks now support a new type of undoable operation: nodes can be \"scaled back\" to their original shape using the operations on the node. This works for all editing operations that produce a tree structure, and all folded operations. The operations available are \"square back\", \"switch back\", \"branch back\" and \"fold back\" operations on the original tree. You can always undo an operation while deciding, up to when the decision is made. (The decision itself will always be undoable.)
@RISK 6.2 made many changes to the Tree Node classes, so that the indexing of the performX methods no longer needed to be done with a switch statement, since the method names now matched the method index on the node in the original decision tree.
The speed up from decision nodes is pretty good at relatively low node counts (100 or fewer nodes in a connected tree) -- about an order of magnitude. The speedup from reference nodes is, as @RISK mentions, less pronounced and generally grows more gradual.
Do you need to run a script or application to generate the tree and then evaluate it? Supervisor UPSIDE-5000 is the answer. It empowers you to automate the generation and evaluation of your decision model. It contains a manual walk-through of using the tool, but it's also used by many users of the tool and it is considered to be the most robust version. 7211a4ac4a