Existing literature on bibliographical metadata presents, among others, 'accuracy', the correct description of a resource, and 'consistency', the uniformity of mappings between its features and metadata fields, as important criteria for the evaluation of metadata quality. My presentation aims to show that these criteria may sometimes be in tension, and that digital humanists treating metadata as research data may have good reasons to prefer consistency to accuracy. The case study I present deals with problems of authorship attributions in early modern dissertations.