In this provocation, I investigate machine learning (ML) developers' accounts, which highlight situated, ongoing improvisational work practices that strive to appropriately match datasets, algorithms/modelling techniques, and domain questions. In conceptualizing ML developers' work as practices of care, I provide a case to more closely examine relationships which emerge between local innovations and global regimes of scientific formalization and standardization within sociotechnical systems (in this case, for example, between applied ML projects and scholarly algorithm development). In discussing these local/global relations, this provocation also brings two concepts from the ML field itself to bear (concept drift and transfer learning), productively challenging how we conceptualize everyday technical work ? and the configurative practices of care it includes.