The point that Bell tried to point out in his “Against ‘Measurement’” article is that when you say “we start including atomic scale things we might as well just include everything up to and including the cat,” but you have to place the line somewhere, sometimes called the “Heisenberg cut,” and where you place the line has empirically different implications, so wherever you choose to draw the line must necessarily constitute a different theory.

Deutsch also published a paper “Quantum theory as a universal physical theory” where he proves that drawing a line at all must constitute a different theory from quantum mechanics because it will necessarily make different empirical predictions than orthodox quantum theory.

A simple analogy is, let’s say, I claim the vial counts as an observer. The file is simple enough that I might be able to fully model it in quantum mechanics. A complete quantum mechanical model would consist of a quantum state in Hilbert space that can only evolve through physical interactions that are all described by unitary operators, and all unitary operators are reversible. So there is no possible interaction between the atom and the vial that could possibly lead to a non-reversible “collapse.”

Hence, if I genuinely had a complete model of the vial and could isolate it, I could subject it to an interaction with the cesium atom, and orthodox quantum mechanics would describe this using reversible unitary operators. If you claim it is an observer that causes a collapse, then the interaction would not be reversible. So I could then follow it up with an interaction corresponding to the Hermitian transpose of the operator describing the first interaction, which is should reverse it.

Orthodox quantum theory would predict that the reversal should succeed while your theory with observer-vials would not, and so it would ultimately predict a different statistical distribution if I tried to measure it after that interaction.