Controlled Evaluation Boundary (CEB v1): A Governance Model for Execution Control in Autonomous Systems

Abstract (IEEE/ACM Style) Artifact: Controlled Evaluation Boundary (CEB v1)Version: 1.0-CEBDate: 2026-06-30License: CC BY-NC 4.0 Supply-chain uncertainty, incomplete review scope, and intellectual-property boundaries challenge the assumption that design-to-delivery transitions represent unconditional progress. When security postures degrade, premature execution or external delivery can increase exposure rather than reduce it. This work presents the Controlled Evaluation Boundary (CEB v1), an axiomatic governance model specifying authorized program states, transition rules, and accountability predicates for research externalization under explicit boundary separation. Execution control is formalized as a first-class systems invariant: transitions toward build-validated or delivery-approved states require documented authorization, separable integrity evidence, and accountable human judgment—not implicit velocity. A central axiom holds that non-execution—the deliberate withholding of build or delivery while gating preconditions remain unsatisfied—is a valid and often preferred security outcome within blocking states such as REQUIRES_DESIGN_DECISION, not a program failure. Open-question gates and fail-closed semantics prohibit invalid state advances when evidentiary justification is absent. The artifact declares governance structures, audit-evidence schema invariants, and boundary ingress rules at the research level. It is a conceptual framework only: non-executable, non-certifying, and implementation-agnostic. CEB v1 establishes reproducible conditions under which shared research artifacts may become validly executable or deliverable to external evaluators amid adversarial and supply-chain uncertainty. Word count: 193