The Problem with Counterfactuals is they can be Counterfactual

The problem with counterfactuals is they can be counterfactual

Counterfeit counterfactuals

Counterfactuals are doubtless very useful however if you posit x being different and everything else being the same that may well be logically inconsistent and impossible without observing making just that change you can't really know for sure or without having an already accurate causal model of the thing.

How can just that change be made is it even logically possible consider the logical formula

(A OR B) OR NOT A

What would it's value be if (A OR B) was false and NOT A was false?

This question is meaningless. You are positing at that point some different notion of the meaning of the symbols in an under specified manner not an alternative assignment of truth values to the variables.

In terminology borrowed from Judea Pearl in Causality: Models, Reasoning, and Inference it lacks any meaning in the do-calculus. Since the condition of doing the setting of those variables is false anything at all can follow.

More generally this means that while you can "locally" reason about necessary and or sufficient causes they may or may not actually exist any at all the causal patterns involved may preclude any local changes and mean only larger changes may make sense providing only a context in which those notions make little to no sense. Causality much like deep learning has an xor problem if you are considering the cause of the output to an xor where no one of the inputs can validly be changed in isolation (say round n-1 of sha256 a cryptographic hash function) then none of those can be considered either a necessary or sufficient causes for the output in a given at that makes any sense to me.

Further if only a subset of the counterfactuals could actually occur you run into two problems: - Your probabilities will have to be among only the possibilities that can actually occur with the counterfactual as specified which in itself can be un-intuitive - There may be a multitude of multivariant changes which are permissible which also have the corresponding counterfactual change made, but there will not be a fixed canonical weighting of which are closer

So does this mean that counterfactuals are bad questions, no only that they can be fraught.