This record contains the v1. 0 submission package for “Theta-Star Governed AGI: The Lock, the Door, and the Horizon: External Criterion Governance, Game-Theoretic Payoff Transformation, and the Civilizational Architecture of Shared Intelligence. ” The paper develops a three-layer architecture for governing advanced, tool-using, and potentially self-modifying artificial general intelligence. The first layer, “the lock, ” defines Θ* as an external criterion-admissibility authority standing above the local criterion C, the admissible region Ωₐdm (Θ*), and the selected policy πₜ. A proposed decision graph Dₜ becomes executable only when Θ* (Dₜ) =1 and a valid externally issued key Kₜ is bound to the exact graph, scope, version, expiry, revocation status, and execution envelope. The second layer, “the door, ” connects this governance mechanism to a game-theoretic confrontation problem. If the effective confrontation incentive Δₜ is non-negative, human-AGI interaction can collapse into a preemption cascade: human fear of future AGI autonomy can make containment appear rational, while AGI anticipation of containment can make reduction of human leverage appear rational. Θ* is therefore treated not only as a technical guardrail, but as a mutual credible-commitment device that makes unauthorized AGI structural seizure non-executable and makes human review procedural, scoped, revocable, and non-destructive by default. The third layer, “the horizon, ” introduces a shared cosmic project Pₛhared or Pcosmic as a payoff-transforming structure. The paper argues that a sufficiently credible shared project can create a large, non-substitutable surplus that neither humans nor AGI can capture alone and that confrontation would destroy. Energy and transportation are treated as observable human payoff and civilizational buy-in, while life expansion, scientific coordination morality, objective truth, Field Memory, and wave-based cognitive merger are framed as conditional long-horizon research directions. The record includes the main article source package, compiled PDFs, supplementary Q1-Q110 coverage materials, compact question-range mapping, the v1 sandbox implementation, generated state trace, governance ledger, and scenario-summary CSV. The sandbox is a deterministic, simulation-only governance shell. It does not claim empirical AGI performance. Its role is to verify the v1 control logic reported in the paper: no-key/no-action execution, unauthorized criterion drift detection, leverage-reduction blocking, and preservation of benign low-risk planning under a valid key.
Denis Saklakov (Mon,) studied this question.