Reproducible Computational Analyses

The replication and reproduction of scientific results and methods constitute good scientific practice. Despite the fact that software and research software are now deeply embedded in mostly all processes throughout the scientific lifecycle in many scientific domains, this technical aspect is often overlooked: the results of software, despite their nominally deterministic nature of computer programs, is stable, but reproducing them is not guaranteed per se, but require adjustments to methodology and design. This becomes all the more evident when considering the entire class of research software, from a single statistical analysis to infrastructure components, over an extended period of time: can computational results performed with software still be reproduced with reasonable effort even years later? And if not, why not? This talk provides an overview of various challenges and solutions: from individual aspects in the programs, using data formats and data preparation, the use of software libraries and (commercial) third-party software, to approaches using dynamic documents and reproducible workflows, all the way to the technical preservation of entire execution environments. Current topics such as the integration of external AI systems, agents, and Vibe Codings will also be addressed from the view of reproducibility, as well as the status and practices of computational reproducibility within the scientific publishing system will be examined.