Demystifier said:
OK, let me then look at it from a slightly different perspective. I think the Bell's theorem can now be concisely stated as follows: There is no single local Kolmogorov model compatible with all probabilistic predictions of QM.
So it seems that we have 3 options:
(i) abandoning locality (Bohmian mechanics, GRW, ...)
(ii) using many models of Kolmogorov axioms, one for each context (Bohr's complementarity, consistent histories, ...)
(iii) abandoning Kolmogorov axioms at some deeper level (??)
My question for you is this: Is there an explicit proposal of an interpretation in the category (iii)? (Perhaps many worlds?)
I think that's rather a question of opinion. I would, of course, choose (ii) but of course not Bohr's "complementarity" (because it's vague; I think the quantum theoretical formalism is much clearer by itself without additional philosophical lingo or consistent histories; just the minimal statistical interpretation). QT is indeed simply a theory to predict probabilities for the outcome of measurements ("observables") given a preparation of a system ("state").
For violating Bell's inequality you have to do different measurements which cannot be done on one system. In the example treated in Sakurai's textbook you prepare two spins 1/2 in the ##S=0## state
$$|\Psi \rangle=1/\sqrt{2}(|\hat{z} 1/2,\hat{z}-1/2 \rangle - |\hat{z} -1/2,\hat{z} 1/2 \rangle,$$
where ##|\hat{a} \pm 1/2 \rangle## denotes the eigenstate of the spin component in direction of the unit vector ##\hat{a}## (I use Sakurai's notation). Then you have to consider the probalities for the outcome of three measurements
$$P_1=P(\hat{a} 1/2,\hat{b} 1/2), \quad P_2=P(\hat{a} 1/2,\hat{c} 1/2), \quad P_3=P(\hat{b} 1/2,\hat{c} 1/2),$$
where the angles between the unit vectors are ##\theta_{ab}=2 \theta## and ##\theta_{ac}=\theta_{bc}=\theta##. Bell's inequality says
$$P_1 \leq P_2+P_3 \quad (\text{any local HV theory a la Bell}).$$
The QT probabilities are
$$P_1=\frac{1}{2} \sin^2(\theta_{ab}/2)$$
etc. and thus Bell's inequality would be
$$\sin^2 \theta \leq 2 \sin^2(\theta/2)?,$$
but
$$\sin^2 \theta-2 \sin^2(\theta/2)=2 \cos \theta \sin^2(\theta/2),$$
which is ##>0## for ##0<\theta<\pi/2##.
For the measurement of each of the ##P_j## (##j \in \{1,2,3\}##) you have to prepare an ensemble of systems in the state ##|\Psi \rangle \langle \Psi|## and perform the specific measurement for each case. There's no way to do all three measurements at one particle and use one big ensemble to get the probabilities needed to check Bell's inequality. You can only measure the spin component in one direction for each of the two particles but never spin components in different directions on one particle.
Just think how to do that measurement in the lab: You have to use a Stern-Gerlach apparatus with a magnetic field in the direction of the spin component you want to measure. This excludes the accurate determination of any other component of the spin in another direction. Note that this is true already for the classical picture of the description in terms of a classical magnetic dipole moment, which precesses rapidly around the direction given by the magnetic field, and thus only the component in diretion of the magnetic field is determined when defined as a time average of the rapidly precessing dipole (of course this is not the accurate quantum picture but it's close to make the SGE plausible!).