POLICY-GROUNDED AI DECISION SUPPORT FOR INDUSTRIAL MODERNIZATION: ARCHITECTURE AND EVIDENCE FROM SMART-FACTORY REGISTRIES IN SHANDONG, CHINA
Abstract
Industrial policy increasingly operates in data-rich environments, yet AI-in-manufacturing research largely focuses on firm-level outcomes rather than accountable decision support for public agencies. This study develops a policy-grounded, governance-aware digital decision support architecture for industrial modernization and evaluates a bounded public-data proof of concept using smart-factory registries in Shandong Province, China. Eight national and provincial policy documents yielded a purposive, non-exhaustive corpus of 55 architecture-relevant operational clauses. Two authors independently coded the frozen units; for the primary policy-instrument code, observed agreement was 0.927 and Cohen's κ was 0.917. Four official releases covering 2023–2026 yielded 775 records, conservatively resolved into 607 enterprise entities. Models were selected on the 2023 records (n = 100) and evaluated on a 2024 chronological holdout (n = 100) excluded from fitting and hyperparameter selection, with zero enterprise overlap. Character-level TF–IDF with logistic regression and a linear support vector machine achieved identical performance (accuracy = 0.780; macro-F1 = 0.778), whereas XGBoost achieved 0.510 and 0.371. The architecture separates implemented registry and diagnostic functions from design-only causal simulation and monitoring under provenance, explainability, auditability, privacy, and human-approval requirements. It provides a partially validated alignment of tasks, methods, and governance; no policy effects, official relabeling of 2025–2026 records, or government deployment are claimed.