JMIR Publications

ABSTRACT

Prevailing ethical oversight of data science health research concentrates on privacy, consent, bias, and fairness. These concerns are necessary but insufficient, because they presuppose that the targets, proxies, labels, classifications, ontologies, and population descriptors already truthfully represent the populations and phenomena they claim to describe. That prior question is often left unexamined. In this Viewpoint, we introduce representational veracity (RV) as a companion concept to the Continuity Trap (CT) framework, grounded in a persistence-based account of representational adequacy developed in the parent framework manuscript. The Continuity Trap names the governance error in which a salient signal of continuity in one domain - technical, administrative, or infrastructural - is treated as sufficient evidence that ethical continuity has been preserved across all domains. In contrast, RV asks whether stored data continue to accurately and responsibly represent the persons, communities, and phenomena from which they were derived — at the point of use, not merely the point of collection. We draw on scholarship in quantification, classification, measurement, critical data studies, algorithmic fairness, and health artificial-intelligence (AI) governance to show that a model may be accurate, reproducible, and formally fair while resting on a representation that is too thin, unstable, or normatively misdirected for the proposed use. Our analysis proceeds through the four-domain representational-veracity architecture (material provenance, informational descriptors, normative authorization, relational community), examines four recurrent failure modes (proxy substitution, category misassignment, label-generation error, descriptor sedimentation), and applies these to a polygenic risk score (PRS) use case. Our argument is not that existing governance mechanisms are wrong but that they need an additional upstream layer. Investigators should be required to justify not only whether their models perform, but also whether their representations are truthful enough for the clinical, public health, or policy claims at hand.