Human-Legible Intermediate Representation

Human and AI authors reason through goals, alternatives, constraints, authority, uncertainty, evidence, and intended effects. Canonical machine representations often optimize different properties: exact structure, identity, validation, transformation, and execution. A Human-Legible Intermediate Representation (HLIR) is proposed as a structured proposal form bridging those concerns without granting proposals canonical or operational standing. An HLIR is best understood as a family of human-facing authoring and projection profiles over governed semantic objects, not necessarily as a new foundational meaning representation. It makes selected distinctions explicit, retains source and unresolved remainder, and supports a declared lowering function into an MCIR candidate. Lowering, canonicalization, adjudication, admission, authority grant, execution lowering, and external effect remain separate transitions. This paper defines an HLIR proposal envelope, validity and legibility criteria, deterministic-lowering conditions, refusal outcomes, non-ambient authority rule, full realization-funnel grammar, AI authoring boundary, round-trip and usability requirements, and evaluation program. The older Specification-Configuration-Implementation triad remains a compact pedagogical profile, while the current architecture distinguishes CM1, CM2, Instance Constitution or Configuration, and Instance or Implementation Realization.