It is common to use a reduced density matrix to describe the physics of an arbitrarily defined subsystem. In this talk, I'll argue that in many physical applications this description fails to capture the fact that the boundary of the subsystem (the ``entangling surface'') is not a real surface. I'll review motivation for a description of subsystem physics in terms of an ensemble of density matrices that capture all physical configurations of the boundary degrees of freedom -- reflecting the non-physics of the entangling surface -- and explore implications for the thermal double decomposition of the vacuum.