[Strilanc]

Not a physicist, but in my mind the basic obstacle to superdeterminism is that in the initial state at t=0 you have O(S) degrees of freedom but throughout some chunk of spacetime there will be O(S * T) Bell violations (where 'S' is "amount of space" and 'T' is "amount of time"). The degrees of freedom you get is fixed, but the constraints grow with time; the system is asymptotically overconstrained. So at first glance you'd guess "no superdeterministic solution exists".

If a solution did exist, it would probably be because of some convenient symmetry or law w.r.t. how the Bell violations played out. But you can use the behavior of arbitrary computer programs to trigger Bell tests, and computer programs are not a well behaved sort of thing. So that makes it seem unlikely for a set of O(S*T) Bell violations out in the world to follow a well behaved pattern that could be compressed into O(S) bits. Like, suppose I decide to dovetail through all computer programs, running a Bell test every time one of the programs halts. This would appear to force the initial state to encode information about solutions to the Halting problem, without the benefit of encoding it into a process that executes over time. But the Halting sequence is algorithmically random; incompressible...