Reproducibility is essential to the credibility, validation, reuse, and cumulative progress of computer and computational science. However, in computing, published results often depend not only on source code and data, but also on software stacks, execution workflows, hardware platforms, middleware, runtime systems, infrastructure configurations, and institutional environments. Grounded in our experience with reproducibility initiatives in SC, ICPP, and TPDS, this paper argues that wide-spread adoption of reproducibility practices is constrained by three structural deficits: the lack of accepted processes, standards, system support for artifact preparation and evaluation, the lack of shared or reproducible infrastructure for executing and assessing computational artifacts, and the lack of adequate incentives for authors, reviewers, chairs, editors, venues, institutions, and funding agencies. We analyze how these deficits affect artifact preparation, review, preservation, and reuse, and we propose recommendations centered on standardized reproducibility processes, shared infrastructure models, and better recognition of reproducibility work. We also explore agentic software systems as emerging mechanisms that can facilitate reproducibility by supporting artifact inspection, environment reconstruction, workflow execution, output comparison, and evidence generation. We argue that such systems can reduce mechanical burden and improve consistency only if they are embedded in auditable, bounded, and human-supervised workflows. Overall, the paper argues that reproducibility in computing should be treated as a structured scholarly process supported by infrastructure and incentives, rather than as an informal expectation for sharing code and data.
Tolosana‐Calasanz et al. (2026) studied this question.