Inspecting compiled semantics
No matching nodes
Choose up to two semantic kinds. One reveals its real traffic; two compare only direct authored relationships.
Choose a kind to inspect its nodes and touching relationships.
Semantic radar needs more MAP space.
What keeps poly-ref from sharing the compiler's mistakes?
A checker is only worth having if it can disagree with the compiler. The architecture calls poly-ref the part most likely to have bugs of its own, so its independence is the whole reason to trust a CLEAN verdict.
poly-ref is written from 04-semantics. Its only compiler-adjacent import is the AST type definitions from poly-core, written as import type so the compiler erases them. It may also import sam-pattern, because it must drive the lowered module to compare with it. It never imports poly-lang, poly-embed or poly-lower.
The boundary is enforced by a script in CI, not by convention. A file that cannot be read is an error and never a pass.
Even so, two implementations can share a blind spot, so the pipeline cross-checks them where it can. polygraph and poly-ref must count the same states on the same lowered module, and each polygraph finding must be confirmed by poly-ref's replay. A disagreement is PL2501 and is addressed to the toolchain, not the author.