Total: 1
Reproducibility has been an increasingly important focus within the Security Community over the past decade. While showing great promise for increasing the quantity and quality of available artifacts, reproducibility alone only addresses some of the challenges to establishing experimental validity in scientific research and is not enough to move forward our discipline. Instead, replicability is required to test the bounds of a hypothesis and ultimately show consistent evidence to a scientific theory. Although there are clear benefits to replicability, it remains imprecisely defined, and a formal framework to reason about and conduct replicability experiments is lacking. In this work, we systematize over 30 years of research and recommendations on the topics of reproducibility, replicability, and validity, and argue that their definitions have had limited practical application within Computer Security. We address these issues by providing a framework for reasoning about replicability, known as the Tree of Validity (ToV). We evaluate an attack and a defense to demonstrate how the ToV can be applied to threat modeling and experimental environments. Further, we show two papers with Distinguished Artifact Awards and demonstrate that true reproducibility is often unattainable; however, meaningful comparisons are still attainable by replicability. We expand our analysis of two recent SoK papers, themselves replicability studies, and demonstrate how these papers recreate multiple paths through their respective ToVs. In so doing, we are the first to provide a practical framework of replicability with broad applications for, and beyond, the Security research community.