The implementation passes 33/33 local tests. The attached review
handoff lists concrete adversarial questions around context-sensitive
operator semantics, precedence/arity, accessibility, Unicode/ASCII
equivalence, MathML/OpenMath mapping, parser ambiguity, and
namespace/versioning.
This remains a non-confidential experimental research prototype, and
it does not claim that its semantic classes or primitives are a
complete description of reality.
We would appreciate one breaking example, relevant prior
art/standards, or a concrete improvement/patch.
Best,
Avraham Bregman
ESLC-AI13 / Symbol Foundry13