CoT+ (Chain-of-Thought Extended): A reproducible multi-reasoning prompting architecture for deterministic PLC code generation

Generating production-grade Programmable Logic Controller (PLC) code using Large Language Models (LLMs) presents a critical challenge: ensuring the determinism and safety required for industrial environments. Current baseline prompting methods—such as Chain-of-Thought (CoT), Program-of-Thought (PoT), Chain-of-Code (CoC), and Contrastive CoT (CCoT)—often yield fragmented results that lack the structural integrity needed for safety-critical IEC 61131-3 synthesis. This paper introduces CoT + (Chain-of-Thought Extended), a unified prompting architecture that integrates CoT, PoT, CoC, and CCoT into a deterministic, multi-stage pipeline for IEC 61131-3-oriented PLC synthesis. The novelty of this research lies in its rigorous three-layer validation framework, which comprises Human-in-the-Loop (HITL), LLM-in-the-Loop (LITL), and Bilingual Evaluation Understudy (BLEU) analyses. To provide additional evidence for the robustness of CoT + , we employ an Inter-Rater Reliability (IRR) assessment using Cohen’s κ for pairwise agreement and Krippendorff’s α for multi-rater agreement. This reliability-qualified statistical approach ensures that performance gains are not only observable across lexical and semantic layers but are also robust to evaluator substitution. Our results demonstrate that CoT + achieves significantly higher safety compliance and cross-validator stability than existing baselines. This research contributes: (i) a deployment-oriented prompting template for industrial PLC synthesis and (ii) a fully reproducible, IRR-qualified validation methodology designed for external verification in safety-critical domains.