Here’s an interesting thought experiment. You can try to manually “internalize” the elements into your graph. Let’s say you create some beams on the initial run. You then proceed to record their IDs as strings like so:
and paste the result into a new code block. Once the design has been revised, you fire up your graph once more, convert those id strings to revit elements with archi-lab’s node and modify their location and size like so:
One of the biggest limitations is that the number of beams has to remain the same between runs.