Fair enough, and it's certainly an interesting experiment, don't want to discourage experimentation

But I will say I think the desire to show the inferred types is orthogonal to the desire to persist them on disk, in the source-of-truth code files. If the state machine really is what you want to experiment with then have fun, but if your desire is just to give the user more visibility into inference, I think there are simpler ways to go about that