Reproducible HPC software deployments, simulations, and workflows – a case study for far-field deep geological repository assessment

Abstract Reproducibility across diverse high-performance computing (HPC) environments remains a major challenge in computational science, particularly for complex, multi-physics simulation workflows. This study presents a comprehensive approach to achieving bit-for-bit result reproducibility in the context of the OpenGeoSys (OGS) simulation suite. Given the widespread use of OGS in environmental science applications such as safety assessments for radioactive waste disposal and the optimisation of geothermal energy systems our approach enhances the reliability, transparency, and acceptance of simulation results in these safety-critical domains, shown in a case study for far-field deep geological repository assessment. We use GNU Guix to define fully declarative, verifiable software environments and deploy them as portable Apptainer containers, enabling consistent execution across multiple HPC systems. Leveraging AiiDA for workflow automation and provenance tracking, we conduct simulations and complex simulation workflows on three heterogeneous clusters, confirming identical binary-level outputs. The results demonstrate that reproducible and portable software environments can offer a pathway toward long-term verifiability in scientific high-performance computing. We also show how full data provenance originating from software source code and model input data resulting in full simulation workflow result data can be achieved.