How do Bell's inequalities actually constrain what hidden variables could predict?

# How Bell's Inequalities Constrain Hidden Variables

Bell's inequalities are mathematical statements about what correlations are possible if nature behaves according to two simple rules: locality (what happens here doesn't instantly affect what happens far away) and realism (properties exist before measurement). The genius of Bell's work is that he showed these two assumptions together imply specific upper bounds on how correlated distant measurements can be. When experiments violate those bounds, at least one assumption must fail.

The Core Constraint

Here's the intuition: imagine two distant labs measuring entangled particles. If hidden variables (unknown properties carried by the particles) determine all outcomes before measurement, and if no signal travels between labs, then the correlations you measure are limited [AG-2024.06-2038]. Specifically, the hidden variables must assign definite results to *all possible measurement choices at both labs, simultaneously*—even though each lab chooses its measurement setting independently and only learns one outcome.

This is the constraint: a local hidden variable model must produce a joint probability distribution over all possible outcomes across all possible measurement settings [AG-2024.06-2038]. That distribution must satisfy no-signaling: changing Alice's measurement choice cannot alter Bob's measurement statistics, and vice versa [AG-2024.06-2038].

The Bell-CHSH inequality (a common form) then bounds the strength of correlations a model respecting these constraints can produce. Quantum mechanics violates this bound [AG-2026.03-1579], implying either locality or realism (or both) fails in nature.

The Subtlety: What Bell's Inequalities *Actually* Rule Out

However, the preprints in this archive highlight an important and somewhat controversial point: the constraints are tighter than they initially appear, and loopholes exist in how we derive them.

The most careful statement is this: Bell's theorem assumes that the probability density governing hidden variables satisfies a particular mathematical form (factorability) [AG-2026.01-032]. If you allow hidden variables with a *richer* probability structure—one permitted by Lebesgue's decomposition theorem—you can construct models that violate the Bell-CHSH inequality while still respecting locality, realism, and measurement independence [AG-2026.01-032]. So Bell's inequalities don't rule out *all* hidden variable theories; they rule out a specific class.

More provocatively, some authors argue that Bell's derivation itself smuggles in an unjustified assumption: that certain sufficient conditions for local realism are actually *necessary* ones [AG-2024.03-1501]. Under this reading, even perfectly loophole-free violations wouldn't definitively rule out local realism.

Why This Matters

The practical upshot: Bell inequalities are powerful, but they constrain hidden variables only *given specific assumptions about how probability and causality work* [AG-2026.01-032]. Experiments can violate the inequalities and still be compatible with hidden variables if those variables have unusual mathematical structure. The inequalities are better understood as ruling out a family of theories (those with factorizable probability distributions, measurement independence, and no retrocausality) rather than all possible hidden variable theories.

The violation does, however, force a genuine choice: either abandon locality, abandon the idea that properties are definite before measurement, or accept that measurement choices and distant properties are conspiratorially correlated [AG-2024.06-2038].