Governed Controlled Machine Language (GCML)

Conventional compilation stacks translate programs through intermediate representations designed for transformation, optimization, portability, and execution. In governed computation, another question precedes machine emission: even when a semantic operation is admissible, does the selected procedure preserve its authority, capability, refusal, evidence, and effect obligations? This paper develops Governed Controlled Machine Language (GCML) as a candidate procedural stabilization layer between admitted constitutional meaning and execution-ready realization. Under the interpretation developed here, GCML does not decide what an operation means or whether its originating semantic artifact deserves admission. It binds an admitted operation to a target-neutral procedural plan whose dependency graph, capability budget, effect call sites, resource assumptions, nondeterminism, refusal branches, receipt obligations, and lowering contract are explicit before target-specific emission. The paper also identifies an unresolved architecture question. The source GCML paper places GCML after MCIR-Law, while the current GSOL v0.1 specification places GCML before MCIR and PEIR and permits several direct lowering targets. Program Execution Intermediate Representation (PEIR) already names the layer where canonical semantic artifacts become executable operational structures. GCML may therefore be a distinct language, a constrained profile of PEIR, a GSOL-to-PEIR lowering contract, or a restatement of machinery already present in those systems. This paper specifies a discriminating contract but does not settle that identity by assertion. The result is a meaning-to-machine gradient in which each boundary has separate responsibilities: language stabilization, governed operation semantics, canonical meaning, admissibility, procedural selection, execution realization, effect adjudication, and evidence. The proposal supplies formal obligations, an artifact schema, a worked example, conformance tests, failure modes, the full realization funnel, and corpus integration. It does not claim a production compiler, deterministic replay of arbitrary external worlds, universal hardware equivalence, or superiority to existing compiler and capability systems without direct research and implementation evidence.