Contemporary healthcare systems generate vast volumes of data, with algorithmic interrogation promising disease prediction, improved diagnoses, and optimised treatment. Despite significant investment, biases in data used for algorithmic interrogation persist, leading to inequities in health outcomes. Scale alone cannot address these biases. Rather, considerations of the contextual dimensions of data need to be reflected upon. Nevertheless, calls for more data to ‘iron out’ such issues are common. Drawing on qualitative interviews with UK-based health data researchers, we use Lucy Suchman's concept of configuration to explore how sociotechnical imaginaries of ‘big data’, which lead to calls for more data, are sustained, operationalised and enacted in everyday research practice. Specifically, we identify three interconnected processes that sustain these imaginaries: (1) risk-oriented narratives that organise research around calculable futures; (2) decontextualising translation processes that align data with algorithmic requirements and (3) a persistent gap between algorithmic capacity and data availability. We conceptualise this third mechanism as a productive gap, as it continually renews commitments to scale by attributing limitations to insufficient data. We argue this gap represents a critical juncture for reconfiguration, revealing where assumptions about decontextualisation might be challenged to create space for more situated approaches to health data research.