Richard,
I'd like to address several of your points:
a.) The essential triviality of Bell's inequality, based on binary outcomes.
b.) The metaphysical implications: realism, locality, logic (or conspiracy).
c.) The meaning of your fifth case, "that QM is wrong?"
To avoid a very long comment, I will limit this comment to a.).
I certainly agree that (the inequality part) of Bell's theorem is, mathematically speaking, a complete triviality. And, as you note, my local model of spin in a non-constant field shows that quantum correlations cannot be matched when the outputs are constrained. As Tim Maudlin repeats above in many different ways, "the theorem is about binary outputs."
The question (to a physicist) is what is the relevance of the theorem about binary outputs to physics? The implication seems to be that QM predicts ±1 and that the QM correlation agrees with experiment. Is that true?
My interpretation of Bell's model is that he applies the wrong quantum mechanical map, Pauli's equation, which is applicable only to a constant magnetic field. The correct QM map would include the deflection energy in the Hamiltonian and would produce a split continuous spectrum of outputs (as observed in Stern-Gerlach), not a binary output. My local model does produce this continuum, and the values are correlated as both QM and experiment imply.
The first objection to this might be simply that "the binary model of QM works!" But is that a consequence, or is in an obvious coincidence?
The question is whether the 'binary' (±1) nature of Bell's (mis-)interpretation of Pauli has anything to do with the correlation? I believe it does not.
Where, in the QM singlet-based expectation value (see eqn (1) in my essay) does the binary nature exhibit itself? One might claim, and even believe, that the sigma-dot-a and sigma-dot-b must be ±1, but the same correlation is obtained from measurements yielding the X cos(θ) values that my model produces.
Discussion of Bell's theorem seems to assume a quantum mechanical 'calculation' based on actual measurement values (assumed identically equal to eigenvalues.) But that is not how the calculation is performed. Instead, the formal QM expression is written down [see the singlet eqn (1)] and the formalism assumes that the correct eigenvalues are being measured. Then, the usual approach to calculating the expectation value [see Peres, page 162 or JJ Sakurai, page 165] is based on an identity that is essentially a geometric relation, independent of the values of a or b or sigma-dot-a, etc.
What is often forgotten is that (per MJW Hall) Bell's theorem includes the physical requirement of perfect anti-correlation when a = b. This implies normalization, since the -1 correlation must obtain, whatever the actual energy eigenvalues. [In fact the 'numbers' are E = ±hw/2, not ±1. I.e., normalization is built-in.]
I think that it is difficult, on a physical basis, to prove that the expectation value -a.b derives from the binary nature of the Stern-Gerlach measurement. And, having briefly looked at the links you provided above, it is also difficult for me to relate the rules:
'+1' stands for "one or more detector clicks",
'-1' stands for "no detector clicks",
to any underlying physics, whatever the statistical significance.
But, as I have noted, I have yet to perform an equivalent analysis of photon physics [equivalent to my SG-analysis]. On the other hand, Bell's derivation, description, and explanation of his model is primarily in terms of Stern-Gerlach, so I don't think SG can be just 'written-off' as some seem inclined to do.
Thanks again for your observations,
Edwin Eugene Klingman