Partial static single information form

One of my hobby projects is adjacent to this, so fun post to stumble across.

I’m building a dataflow language with explicit effect boundaries, which creates an interesting angle here: some facts can be derived by literally evaluating pure subgraphs rather than only through static analysis. That makes the AOT/JIT boundary feel more like a continuum than a hard distinction.

Seeking partial SSI also feels adjacent to formal methods -- refinement types / interval analysis / effect systems all seem like ways to surface stronger facts for the optimizer, rather than relying purely on inference.

Tangentially, Rust’s niche optimizations (Option<&T> etc) feel like a tiny example of the same broader theme: exploiting stronger facts about values for representation wins.