We develop a framework for partial identification of causal effects under violations of key assumptions when combining heterogeneous data sources.