HITL provides an expert, safety-focused audit of LLM-generated PLC code in native engineering environments (Siemens TIA Portal, Beckhoff TwinCAT). It serves both as the acceptance gate for deployment and the reference signal used to calibrate automated validators. For each use case, the review verifies four aspects: compilation for syntactic correctness and build integrity; simulation/runtime behaviour covering sequence logic, timers, and interlocks; safety and regulatory compliance—including fail-safe modes, E-stop dominance, and watchdogs; and maintainability assessed from structure, naming, comments, and modularity. The representative outcomes for simple, medium, and complex scenarios are summarised in the attached table; the complete matrix spanning all 25 industrial cases is available in the supplementary material (and Zenodo archive). This protocol yields auditable ground truth, captures vendor-specific nuances that automated metrics may miss, and provides high-precision labels for calibrating LITL thresholds and prioritising risk-critical fixes.

HITL provides an expert, safety-focused audit of LLM-generated PLC code in native engineering environments (Siemens TIA Portal, Beckhoff TwinCAT). It serves both as the acceptance gate for deployment and the reference signal used to calibrate automated validators. For each use case, the review verifies four aspects: compilation for syntactic correctness and build integrity; simulation/runtime behaviour covering sequence logic, timers, and interlocks; safety and regulatory compliance—including fail-safe modes, E-stop dominance, and watchdogs; and maintainability assessed from structure, naming, comments, and modularity. The representative outcomes for simple, medium, and complex scenarios are summarised in the attached table; the complete matrix spanning all 25 industrial cases is available in the supplementary material (and Zenodo archive). This protocol yields auditable ground truth, captures vendor-specific nuances that automated metrics may miss, and provides high-precision labels for calibrating LITL thresholds and prioritising risk-critical fixes.

This study employs a benchmark that couples 21 prompting techniques with 25 real-world PLC use cases, yielding 525 prompt–task evaluations across simple, medium, and complex scenarios. The prompting suite spans foundational methods (zero-/few-shot), reasoning-focused prompts (CoT, PoT, CoC, SCoT), utility and optimization prompts (documentation, OP, RaR, APE), verification and role prompts (CoVe, CoN, ReAct, AP, RBP), and vendor-/symbol-aware variants (RAG, ITP, CoS, CoTP, CCoT, MTPR). The industrial cases range from conveyor and tank-level control to CNC tool change, AGV docking, and robot pick–place, providing broad coverage of sequencing, safety interlocks, and multi-axis coordination. Each instance includes a problem description, an expected IEC~61131-3 output (ST/IL), and a human-verified ground truth, enabling consistent application of fixed prompts and fair comparison across vendors. This design supports reproducible ablations on prompting strategy, facilitates cross-task generalization studies, and allows researchers to select subsets aligned with their evaluation goals.

This study employs a benchmark that couples 21 prompting techniques with 25 real-world PLC use cases, yielding 525 prompt–task evaluations across simple, medium, and complex scenarios. The prompting suite spans foundational methods (zero-/few-shot), reasoning-focused prompts (CoT, PoT, CoC, SCoT), utility and optimization prompts (documentation, OP, RaR, APE), verification and role prompts (CoVe, CoN, ReAct, AP, RBP), and vendor-/symbol-aware variants (RAG, ITP, CoS, CoTP, CCoT, MTPR). The industrial cases range from conveyor and tank-level control to CNC tool change, AGV docking, and robot pick–place, providing broad coverage of sequencing, safety interlocks, and multi-axis coordination. Each instance includes a problem description, an expected IEC~61131-3 output (ST/IL), and a human-verified ground truth, enabling consistent application of fixed prompts and fair comparison across vendors. This design supports reproducible ablations on prompting strategy, facilitates cross-task generalization studies, and allows researchers to select subsets aligned with their evaluation goals.

This record contains the dataset (QuRE.csv) and replication package (QuRE-analysis-main.zip) associated with the paper "Description and Comparative Analysis of QuRE: A New Industrial Requirements Quality Dataset."The QuRE (Quality in Requirements) dataset consists of 2,111 industrial requirements, annotated through a real-world quality review process conducted over several years. The dataset has been used extensively in an industrial context and is now released to the research community to support empirical studies on requirements quality.This version (1.0.1) corrects a previous issue in which the file 1_QuRE_description.ipynb was empty.The accompanying replication package includes descriptive statistics, comparative analyses with existing and synthetic requirements datasets, and additional material to facilitate reuse and validation.

This record contains the dataset (QuRE.csv) and replication package (QuRE-analysis-main.zip) associated with the paper "Description and Comparative Analysis of QuRE: A New Industrial Requirements Quality Dataset."The QuRE (Quality in Requirements) dataset consists of 2,111 industrial requirements, annotated through a real-world quality review process conducted over several years. The dataset has been used extensively in an industrial context and is now released to the research community to support empirical studies on requirements quality.This version (1.0.1) corrects a previous issue in which the file 1_QuRE_description.ipynb was empty.The accompanying replication package includes descriptive statistics, comparative analyses with existing and synthetic requirements datasets, and additional material to facilitate reuse and validation.

Additional file 1.

Additional file 2.

Additional file 2.

Additional file 1.