Governed Controlled Natural Language (GCNL)

Natural language can express nuance, ambiguity, implication, rhetoric, and context that executable systems cannot safely treat as determinate instruction. This paper develops Governed Controlled Natural Language (GCNL) as a human-authorable representation layer that constrains grammar, resolves or exposes references, separates epistemic posture from normative force, records asserted authority and provenance, closes scope, preserves residual ambiguity, and produces a canonical semantic candidate for downstream evaluation. GCNL does not make a statement true, legitimate, authorized, or executable. Governance metadata is a claim about authority, not authority itself. A GCNL compiler may parse, resolve, type, canonicalize, and refuse malformed input; a separately identified law and admission boundary determines whether the resulting candidate may acquire standing or be lowered into GSOL, GOML, MCIR, GCML, PEIR, or another realization profile. This separation keeps a language tool from becoming an undeclared sovereign. The paper retains the source's crystallization gradient and cooling metaphor while treating them as architectural metaphors rather than thermodynamic measurements. It adds a typed statement model, partial compilation semantics, loss and residual records, preservation invariants, a worked example, configuration selection through CM1/CM2/Instance Configuration, adversarial tests, and explicit authorial-compression criteria. The result is a proposal for governable semantic drafting, not a completed language implementation or a proof that controlled prose eliminates interpretation.