Arguing machine learning assurance for certification

Aviation certification traditionally relies on standards shaped by expert consensus on best practices. However, for emerging technologies like machine learning (ML), establishing these practices is difficult, particularly when in-flight testing is not feasible before certification, which can slow innovation. Argument-based certification offers a promising solution to bridge this gap by providing a structured and transparent framework for discussions between applicants and certification authorities. It allows authorities to compare proposed compliance methods across similar technologies from different applicants, helping to identify best practices and assess risks associated with new innovations. In this paper, we apply the Overarching Properties organizing principle to define and assess the means of compliance (MoC) for three publicly available ML aviation applications. This paper demonstrates how OPs can be used to support the demonstration of compliance. Although we observed some commonalities in the three arguments, the specificities of the technology and its application highlight the differences in the strategies used to demonstrate assurance.