Who governs government data? A bounded data stewardship role for national statistical offices

National statistical offices (NSOs) increasingly operate within data ecosystems in which administrative data remain under the authority of sectoral institutions. This creates pressure for NSOs to contribute to government-wide data governance without assuming responsibility for all government data. This conceptual and practitioner-oriented paper proposes bounded data stewardship as a model for allocating horizontal stewardship functions to an NSO while preserving sectoral authority and operational accountability. It develops four criteria for allocation: cross-government value, dependence on semantic or methodological coordination, alignment with NSO competence, and the ability to exercise the function without transferring substantive sectoral responsibility. Comparative arrangements in Lithuania, New Zealand, Canada, Switzerland and Germany illustrate alternative distributions of stewardship and statistical authority. The paper relates the model to public-sector interoperability frameworks and to SDMX and GSIM, identifying metadata, classifications, reference data and quality methodology as candidate functions rather than universal NSO responsibilities. Practitioner evidence from statistical modernization in the Kyrgyz Republic shows that an expert-reviewed specification for a statistical metadata repository did not proceed to development because operational responsibility, maintenance processes and the readiness of the initial metadata content remained insufficiently defined. Bounded stewardship combines distributed responsibility with horizontal coordination and requires sufficient institutional definition before technological formalization.