From the paper: "causally closed levels can be efficiently controlled from just macroscopic interventions, without needing to intervene on microscopic condition

From the paper: "causally closed levels can be efficiently controlled from just macroscopic interventions, without needing to intervene on microscopic conditions." I think that's stating things causally rather than predictively but two sides of the same coin. I agree though. One issue the paper doesn't seem to underscore or delineate are the roles of measurement or intervention costs (in an information or utility theoretic sense) at different levels of emergence and how those affect decisions to work at one level or another.