Active Learning for Interactive Prompt Clarification in Safety-Critical Programmable Logic Controller Code Generation

This document provides the companion reproducibility and audit package for the manuscript Active Learning for Interactive Prompt Clarification in Safety-Critical Programmable Logic Controller Code Generation. It documents the complete implementation and evaluation of Interactive Prompt Clarification with Active Learning (IPC+AL), its fixed-protocol baseline IPC, and the reference prompting methods RaR, APE, AP, ReAct, and MTPR. The package covers 25 industrial PLC use cases spanning simple, medium, and high complexity and provides task-specific prompts for generating IEC~61131-3 programs in Structured Text (ST/SCL) and Instruction List (IL/STL). Nine representative use cases—three from each complexity level—are examined in detail to compare IPC+AL with IPC using three LLM-based validators and a fixed rubric covering correctness, readability, safety, modularity, overall quality, and acceptance-gate attainment. Component-level ablations further examine the contributions of value-of-information ranking, uncertainty and safety-risk governance, clarification budget, and fail-closed emission. The package also formalizes LLM-in-the-Loop validation, Human-in-the-Loop adjudication, and inter-rater reliability using Cohen's $\kappa$ and Krippendorff's $\alpha$. Overall, it enables independent regeneration, verification, and extension of the reported experiments while transparently identifying where IPC+AL improves, preserves, or reduces performance across models, PLC languages, and industrial tasks.