Almost all aspects of scientific research and communication are now supported by software systems. Even though most of these systems allow the user to specify interaction preferences or even employ user modeling techniques, every system is an island with this respect. In particular, different systems cannot share user models or predict in the absence of prior interactions.We use ideas from the theory of Communities of Practice to arrive at declarative
models that predict prior knowledge, preferences, and learning paths of scientists. Our models build on collections of semantically marked up artifacts (CoPfolios) that inscribe scientific practice. Our vision is to provide a toolkit that encapsulates CoPfolios allowing scientific applications to outsource and share user data.