Why is Yarn Spinner not data driven but instead compiled?

[maarten-abbey]

I've used Yarn Spinner for a while now because it seems to be the most complete solution for narratives, especially from a workflow-perspective. Very nice!

Something that has bothered me from the start is that yarn is compiled into executable code. I cannot "look ahead" or "rewind" or otherwise use reflection to expand functionality. To me it seems a weird choice because it comes with many disadvantages, while I cannot see any advantages, except maybe for a (negligible?) performance improvement.

Code, logic, commands, etc could all still be done if everything was interpreted data from a hierarchical structure, which is what Yarn scripts could be modeled as, even in their current form. And even if you really want to keep some parts compiled, at least it would be nice to have other parts of the story available up front.

As a feature request this is probably nonsensical because of the effort required, but I'm very curious why this architecture was chosen, what other people think and whether there has been any effort in the direction that I'm describing.

Thanks, and keep on the good work!