Knowledge-Guided Synthesis of No-Code Dispatching Rules with Large Language Models: Structural Invariants, Execution, and Feedback for Smart-Factory Material Control

Abstract

Reproduction package (version 4) for the article "Knowledge-Guided Synthesis of No-Code Dispatching Rules with Large Language Models: Structural Invariants, Execution, and Feedback for Smart-Factory Material Control". It contains the synthesis pipeline (prompt builder, tool schema, invariant validator, lint checks), the evaluation harness, all natural-language inputs (standard, adversarial, persona-simulated, English), the rule catalogs, the retrieval corpus and embeddings, every raw model output and per-call result, the rater sheets and consensus ground truth, the rule-flow interpreter and simulators, the three layout graphs, the per-seed KPI results, and the statistics and table scripts. Proprietary production traces of the reference fab are not included (confidentiality agreement); all reported results reproduce from the deposited layout graphs and fixed random seeds. See README.md for the file map and the commands that regenerate every table. Source code is licensed under the MIT License; data and documentation under CC BY 4.0.