As science becomes increasingly data-intensive, the requirements for comprehensive Research Data Management (RDM) grow. This often overwhelms scientists, requiring more workload and training. The failure to conduct effective RDM leads to producing research artefacts that cannot be reproduced or reused. Past research placed high value on supporting data science workers, but focused mainly on data production, collection, processing, and sensemaking. In order to understand practices and needs of data science workers in relation to documentation, preservation, sharing, and reuse, we conducted a cross-domain study with 15 scientists and data managers from diverse scientific domains. We identified five core concepts which describe requirements, drivers, and boundaries in the development of commitment for RDM, essential for generating reproducible research artefacts: Practice, Adoption, Barriers, Education, and Impact. Based on those concepts, we introduce a stage-based model of personal RDM commitment evolution. The model can be used to drive the design of future systems that support a transition to open science. We discuss infrastructure, policies, and motivations involved at the stages and transitions in the model. Our work supports designers in understanding the constraints and challenges involved in designing for reproducibility in an age of data-driven science.