Research Data Management in Stochastic Optimization: A Case Study on Heuristic Search and Retrospective Curation

Abstract

The complexity of stochastic optimization and large-scale benchmarking has elevated Research Data Management (RDM) from a peripheral task to a core methodological requirement. While the FAIR principles, that ensures data is Findable, Accessible, Interoperable, and Reusable, provide high-level guidance, their practical implementation in doctoral-level computational research remains insufficiently documented. In fact, existing guidelines are often difficult to apply as they do not immediately adapt to the diverse and specific scenarios of high-dimensional research. This paper presents a case study on the RDM lifecycle of a heuristic framework, documenting the transition from non-standardized experimental logs to FAIR-compliant datasets in the CORA.RDR repository. We identify critical challenges such as data heterogeneity, volume, and the technical debt generated by stochastic variability across thousands of independent runs. In response, we propose a lightweight, practical framework based on hierarchical structures, self-documenting naming conventions, and multi-level metadata schemas. The contribution of this work lies in bridging the gap between theoretical RDM and the non-linear reality of doctoral research. By providing actionable lessons on retrospective curation and interpretable transparency, we offer a roadmap for researchers to ensure the long-term reusability and integrity of complex computational artifacts.