"Reproducibility and Auditability of Real- Time Control AI Systems Under Deterministic Execution Constraints"

"This paper presents a novel framework for execution-level reproducibility in artificial intelligence-enabled cyber-physical systems (CPS), with a specific focus on real-time control and industrial informatics applications. Unlike conventional notions of reproducibility that are limited to dataset-level or model-level consistency, the proposed approach redefines reproducibility as a deterministic system property, ensuring consistent outputs under identical operational conditions, including timing constraints and execution environments. The framework integrates bounded inference latency, execution-time predictability, and artifact-level traceability, thereby addressing critical challenges associated with deploying AI models in safety-critical and time-sensitive systems. A modular architecture is developed to support closed-loop control, combining machine learning inference with deterministic scheduling and verification mechanisms. The proposed methodology is validated through comprehensive simulations and implementation pipelines, supported by reproducible Python-based modules and system-level design artifacts. Furthermore, the work extends toward system-on-chip (SoC)-oriented deployment, demonstrating feasibility for embedded and edge environments such as smart grids and autonomous platforms. Experimental results highlight the effectiveness of the framework in maintaining consistent performance under varying operational conditions, while ensuring transparency and auditability. The proposed approach provides a foundational step toward certifiable, reliable, and deployment-ready AI systems for next-generation industrial applications."