Empirical Software Engineering increasingly relies on controlled experimentation, yet current reproducibility practices focus primarily on artifact availability rather than on how experimental knowledge is represented and reused at scale. As a result, many experiments remain reproducible in principle but difficult to interpret, compare, and integrate across contexts and time. This position paper argues that this gap constitutes a structural challenge for the future of Software Engineering and proposes metadata-driven experimentation as an evolution of existing reproducibility practices, treating experimental metadata and conceptual structure as first-class research artifacts. By enabling structured, machine-interpretable representations of experimental intent, design, context, and analysis, metadata complements narrative reporting and executable artifacts, supporting cumulative knowledge building and scalable empirical research in the coming decade.

@inproceedings{paperbot12651,
  title = {Toward Metadata-Driven Experimentation in Software Engineering: A Vision for Reproducible, Interoperable, and Transparent Research},
  author = {Filipe A. Santana and André F. R. Cordeiro and Edson OliveiraJr},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3793657.3793894}
}