A Python Toolkit for Standardized Job Shop Scheduling Benchmark Instances

The Job Shop Scheduling Problem (JSP) is a core challenge in operations research, often used to benchmark optimization and machine learning algorithms. Although many benchmark instances are publicly available, their formats typically require pre- processing before integration into Python-based research workflows. A Python module named jsp-instance-utils has been developed to provide ready-to-use JSP benchmark instances in a standardized format. Benchmark instances were obtained from the pub- lic repository jobshop.jjvh.nl. Each instance was parsed and converted into a NumPy, the de facto standard in Python scientific computing. The instances were then consolidated into a Python module to simplify direct access without needing textfile parsing. The dataset and software are published following the FAIR principles. Persistent iden- tifiers and metadata allow users to locate the dataset across platforms such as GitHub, GitLab, PyPi, and Zenodo. Access is unrestricted, and no registration or proprietary tools are required. The complete data acquisition and conversion code is documented, allowing users to reproduce or extend the dataset using the same procedures. The module also includes a baseline solver using Google OR-Tools Constraint Program- ming Solver. The solver provides a benchmark solver against more sophisticated Al- gorithms allowing comparisons without requiring additional software installations. Util- ity functions for generating custom instances beyond the provided dataset are also included. Documentation is provided through Read the Docs and covers installation, function usage, and practical examples, including integrating benchmark instances into reinforcement learning pipelines for JSP. The software is implemented in Python and re- leased as open source. The provided dataset and software reduce technical overhead in JSP research and promote reproducibility. Publishing according to FAIR principles ensures that the data can be found, accessed, used in different systems, and reused in future research contexts. The module assists optimization and learning-based schedul- ing research by aligning benchmark availability with modern Python workflows.