Perhaps one should say even that the physical correlates are never physical entities (of course, depending on what you mean by “entities”). What we perceive in the environment is not an entity but a property of one or more entities. What are (real, basic, indivisible) entities; it is problem of ontology and theoretical physics, but in our normal physics based everyday model of reality every entity is divisible and consists of other smaller entities. Thus, a perceived variable can a property of an atom, a molecule, a separate object, an amount of liquid, a heap of grains, a boxful of bolts and nuts, etc. or any combination of these. But these entities are never perceived as such but only their properties. However, if we would want to control (e-control) the perceived property we must cause physical effects to those entities the properties of which we are perceiving. If it is a (combined) property of a complex entity then we must, depending on the case, affect either one, some or all of the participating sub-entities. (In computer simulations the value of a correlate variable can adjusted straightly just by changing the calculations but not so in real life.)
According to this (Powers): CP = controlled perception (inside controller); CV = controlled (physical) variable (in the external environment of the perceivers). But then (Powers):
Here is an apparent error: Observer’s perceptions are NOT observables - not even to the observer herself without a fantastic neurological apparatus. We do not observe our perceptions but those physical properties which are available to our senses. We have our perceptions but we do not perceive them.
So, for the description of TCV we need at least three variables: one is a theoretical variable inside the controller, the controlled perception (CP); another is an observable physical property which I would like to call CEV (Corresponding Environmental Variable) rather than qi which refers to the value (quantity) of that input variable; and the third is the perception inside the observer/tester which she infers and decides to be the “same” (=similar, ideally identical) perception than that which the controller is controlling. This third could be called Controlled Variable (CV) because it is the result of the Test for the Controlled Variable. I think this is a valid inference “[a]s far as the observer is concerned” as Powers says.
The critical question here is - in addition to the methodological questions of TCV - whether the input functions of both the controller and observer/tester really are similar or identical. With low-level perceptions and those which have been important in the evolutionary history it is a viable assumption. But with higher level (and continuously developing) cultural perceptions (like just honesty and democracy) it is very susceptible!
Best
Eetu