I'm reading the Idris book now, and one cool thing there is guessing bits of implementation based on the type declarations (there written before implementation).
I imagine in a visual environment, being able to make useful suggestions on potential ways to use/combine different nodes/types would help as well as an auto-complete for the user's intent.
I'm actually also a bit reminded here of MS Excel Power Query, which also offered a GUI for data transformation, see e.g. [this pic](https://blogs.office.com/wp-content/uploads/2015/07/6-update...).
I bring this up because I see you covered visualizing the steps, while they focused on showing the data (though after finishing a transformation the script could be generalized into a reusable function). I wonder if adding a dimension like that could be helpful for Luna as well.
If you target non-programmers, showing things as concrete as possible (e.g. their data transformed by whatever function they just pulled together) sounds like it might help make things even more accessible.