What question did this study set out to answer?

To introduce a governance kernel that ensures safe and efficient operation of multi-agent AI systems in various environments.

April 12, 2026Open Access

OR1ON DDGK: Deterministic Decision Governance Kernel for Multi-Agent AI Systems

Key Points

To introduce a governance kernel that ensures safe and efficient operation of multi-agent AI systems in various environments.
Developed a 16-agent coordination framework with governance overlay.
Implemented the PROVA engine for proof-based reasoning with trivalent decision logic.
Established a six-phase operational loop using the Sovereign Industrial Kernel (SIK).
Formulated decision spaces constrained by physics using Control Barrier Functions and Lyapunov stability theory.
Achieved a false acceptance rate of 0%, a 50% improvement over existing solutions.
Measured decision latency of 11.69ms end-to-end for P99 performance.
Obtained a 50% score on Autonomous Reasoning Capability.
Satisfied the Quantum Bell classical bound with a score of 1.0.

Abstract

We present the OR1ON Deterministic Decision Governance Kernel (DDGK), a comprehensive architecture for safe multi-agent AI operation in physical and cyber-physical systems. The DDGK enforces safety through a deterministic, pre-execution validation layer that structurally separates proposal generation from action execution. The architecture integrates four novel contributions: (1) a 16-agent multi-agent coordination framework with deterministic governance overlay, (2) the PROVA engine providing proof-based reasoning with trivalent decision logic (ALLOW/BLOCK/ABSTAIN) and sub-millisecond P99 latency (2.76ms) on edge hardware, (3) the Sovereign Industrial Kernel (SIK) six-phase operational loop, and (4) the EIRA epistemic validation layer. We formulate physics-constrained decision spaces using Control Barrier Functions and Lyapunov stability theory. A SHA-256 chained audit trail provides cryptographic non-repudiation of all governance decisions. Benchmark results: False Acceptance Rate 0% (50% improvement over baselines), decision latency 11.69ms end-to-end P99, Autonomous Reasoning Capability (ARC) score 50%, and Quantum Bell classical bound satisfaction (1.0). We map DDGK compliance to the EU AI Act (Articles 9, 11, 13, 14, 17), ISO 26262 (ASIL-D), and IEC 61508 (SIL-2/3).

OR1ON DDGK: Deterministic Decision Governance Kernel for Multi-Agent AI Systems

Key Points

Abstract

Cite This Study