Data governance scholarship has been dominated by a structuralist tradition that treats governance as a design problem: specify the correct structures, assign accountabilities, and appropriate behaviours will follow. Recent empirical research challenges this assumption, consistently finding that human and cultural factors are stronger predictors of governance outcomes than structural arrangements. This paper proposes an interpretivist definition of data stewardship grounded in Bevir's decentred governance theory, Weick's organisational sensemaking, Löfstedt's risk governance, and Flyvbjerg's phronetic analysis. The definition is constructed in three cumulative clauses: stewardship as the exercise of situated agency oriented toward governance objectives; the scope of that agency as bounded by interpretive space shaped by structural constraints and power relations; and decisions as produced through active interpretation drawing on professional traditions, inherited webs of belief, and locally accumulated knowledge. The definition is compared systematically against nine practitioner sources and against the academic literature. No scientific source offers a theoretically developed definitional account of data stewardship; the proposed definition addresses a gap explicitly identified in the literature as a research priority.