Consider an experiment of two sources (X, Y) of polarization entangled photons, which are anti correlated, and 4 polarization detectors (arranged from left to right as A, B, C, D), where A, B are set up to measure the photons from X. C, D measure photons from Y. Now imagine that detectors B and C are on a lazy susan gear contraption, where the experimenter can turn a crank, and B and C will switch positions, so now X photons go to A & C, and Y photons go to B & D.

Finally, imagine a demon with access to the hidden variable information that perfectly predicts each detector measurement.

Now, regardless of whether the detectors are ordered ABCD or ACBD, the measurement results from left to right can only be one of the following: HVVH, HVHV, VHHV, VHVH where H = horizontal, V = vertical. This is due to the prepared Bell states.

We will now focus on the case where i) the detectors were originally arranged as ABCD, ii) the photons are emitted at t=0, iii) the experimenter turns the crank at t=1, changing the order to ACBD, and iv) the measurement at t=2 reveals the result HVHV.

Finally, we ask the demon to tell us the status of the hidden variables at t=0 and t=1 in this postselected subset of runs. As i the TI, the hidden variables here are assigned to the detectors, not the photons.

In a nonlocal deterministic interpretation, the demon will say that at t=0 the hidden variables were arbitrary, but that by t=2 they were nonlocally steered into the right correlations. For example, it could be that at t=0, the hidden variables predicted HVHV, but then the crank rotation at t=1 turned this into HHVV. Since HHVV is not a possible outcome, then between t=1 and t=2, the nonlocal beables had to steer the B and C local beables, so the prediction returned back to HVHV.

In a superdeterministic interpretation, the demon will instead say that at t=0, the hidden variable state was always, in every run, HHVV, and it was in fact the crank rotation itself that set it to the acceptable HVHV. There is no nonlocal beable to nudge or steer or correct the hidden variables here, so they had to be fine tuned from the start, as if they already knew the future experimental procedure.

In short, superdeterminism means the hidden variables are fine tuned to always *exploit the crank* in order to deliver correct quantum correlations, seemingly knowing in advance that the crank will be turned, or being astoundingly lucky. In contrast, nonlocality means hidden variables exist to *fix the situation when the crank spoils* the correct quantum correlations.

Retrocausal approaches are sort of a compromise of these two approaches.