Causal Friendliness Paradox

• The Causal Friendliness Paradox is defined by foundational no-go results where classical causal logic fails to predict quantum operational outcomes.
• It integrates detailed analyses of the Neyman–Fisher and Wigner’s Friend setups, deriving inequalities that defy traditional interpretation.
• The paradox reveals that equivalent null hypotheses can yield divergent test results, challenging both experimental design and causal assumptions.

The Causal Friendliness Paradox designates a suite of foundational no-go results in causal inference and quantum foundations, focused on the impossibility of reconciling classical causal reasoning with certain quantum phenomena under natural metaphysical and statistical assumptions. The term has two distinct but related cores: first, within randomization-based classical inference (notably the Neyman–Fisher paradox), and second, in the context of Wigner’s Friend-type quantum thought experiments culminating in the Local Friendliness and Causal Friendliness no-go theorems. Both strands confront the notion that logical implications between null hypotheses (or metaphysical assumptions) entail corresponding implications at the level of empirical statistical tests or inequalities. The paradox illustrates that, even when logical containment of hypotheses or metaphysical axioms holds, quantum theory or even classical randomization can yield counterintuitive separations in operational predictions.

1. Classical Origins: Neyman–Fisher Paradox in Causal Inference

The classical form of the causal friendliness paradox was first articulated by Ding in the context of randomization-based causal inference, where two prominent null hypotheses are considered:

• Neyman’s weak null: $H_0^{\rm Neyman}: \tau = 0$, i.e., the average causal effect is zero.
• Fisher’s sharp null: $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$, i.e., every unit has zero effect.

Logically, $H_0^{\rm Fisher} \implies H_0^{\rm Neyman}$. However, their associated tests—the Neyman difference-in-means $p$-value and the Fisher randomization $p$-value—do not respect this implication operationally. Specifically, Neyman’s test often has greater power than Fisher’s, sometimes leading to empirical cases where one rejects $H_0^{\rm Neyman}$ but not $H_0^{\rm Fisher}$. Ding provides detailed asymptotic expansions:

$$\widehat V_{\rm Fisher} - \widehat V_{\rm Neyman} = (N_0^{-1} - N_1^{-1})(S_1^2 - S_0^2) + \frac{\tau^2}{N} + o_p(1/N)$$

implying a more diffuse null for Fisher’s test when $\tau \ne 0$. This separation persists under constant treatment effect, stratified experiments, matched pairs, and $2^K$ factorials. The paradox signals that logical containment between hypotheses does not entail inclusion of the corresponding rejection regions in finite-sample or asymptotic analysis (Ding, 2014).

2. Quantum Formulation: Wigner’s Friend, Local Friendliness, and the LF No-Go Theorem

The quantum version arises in extensions of the Wigner’s Friend scenario, culminating in the Local Friendliness (LF) and Causal Friendliness (CF) paradoxes. Central is the assertion that classical notions of causality—supervening on axioms such as Absoluteness of Observed Events (AOE), Local Agency (LA), and related causal DAG constraints—fail to account for quantum predictions in friend-type settings.

The minimal LF scenario involves three parties (Charlie, Alice, Bob) and leads to the derivation of LF inequalities, such as:

$H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$0

Quantum mechanics predicts violation of the bound ($H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$1 with perfect tracking), thus forcing the abandonment of at least one LF assumption (Yīng et al., 2023).

3. Formal Structure in Timelike and Spacelike Scenarios

The Causal Friendliness Paradox is formulated in both spacelike and timelike settings:

• Spacelike (LF): Local Agency, AOE, and Tracking yield LF inequalities that are analytically and empirically violated by quantum strategies.
• Timelike (CF): The spacelike “Locality” axiom is replaced by Axiological Time Symmetry (ATS) and No Retrocausality (NRC). Screening via Pseudo Events (SPE) is also introduced. The result is a causal Bell–CHSH-type inequality on the observed correlations between truly-observed events $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$2:

$H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$3

Quantum protocols reach the Tsirelson bound $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$4, in contradiction (Mukherjee et al., 30 Oct 2025). Even weakening AOE to an operational mediation property suffices for the paradox to hold.

4. Causal-Graphical and Marginal Approaches

The paradox is rigorously anchored in the structure of causal models:

• Causal DAGs and $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$5-separation: LF and CF assumptions correspond to compositional separation rules in the relevant DAG, constraining the set of compatible joint distributions. The monogamy relations and polytope constraints derived from the marginal problem for overlapping distributions $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$6 or $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$7 yield the LF inequalities (Yīng et al., 2023).
• No Fine-Tuning (NFT): Any causal model (classical or nonclassical) is “faithful” if every conditional independence in the observed probabilities arises from a $H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$8-separation in the graph. The paradox demonstrates that, even with access to generalized probabilistic theories (GPTs) and cyclic causal models (i.e., allowing for cycles in causality), violation of LF/CF inequalities cannot be replicated without explicit fine-tuning, i.e., conspiratorial parameter choices not mandated by the graphical structure.

5. Implications for Interpretations and Causal Inference

The Causal Friendliness Paradox delineates sharp boundaries for models of causality:

• Stronger than Bell: The LF and CF no-go theorems strictly strengthen Bell’s theorem, applying even when allowing nonclassical, nonlocal, and cyclic causal models, provided they obey basic metaphysical and statistical constraints.
• Relativity of Events: To resolve the paradox, one must forfeit at least one foundational assumption. Denying AOE leads toward a “relativity of events,” where measurement records exist only relative to particular observers, in keeping with relational quantum mechanics, QBism, or certain Everettian (many-worlds) interpretations (Cavalcanti et al., 2021).
• Limits of Classical Causal Inference: In randomization-based contexts, the paradox stresses the non-equivalence of logical and operational containment and guides practical inference methodology choice.

A summary of the key conceptual and operational distinctions entailed by the paradox is presented below:

Domain	Logical Null Implies...	Empirical Rejection May...
Neyman–Fisher	$H_0^{\rm Fisher}: \tau_i = 0 \ \forall i$9	...see $H_0^{\rm Fisher} \implies H_0^{\rm Neyman}$0 without $H_0^{\rm Fisher} \implies H_0^{\rm Neyman}$1
Wigner/LF/CF	LF/CF assumptions $H_0^{\rm Fisher} \implies H_0^{\rm Neyman}$2 LF/CF inequalities	...see quantum protocols violate LF/CF bounds

6. Possible Resolutions and Generalizations

Two principal avenues for reconciling the paradox with quantum and causal modeling are proposed:

• Non-compositional causal structures: Abandoning compositionality in separation criteria allows marginal independences not to factor through pairwise relations, but this undermines the explanatory power of standard DAG methodology (Yīng et al., 2023).
• Relational/event-indexed causality: Developing frameworks where events are not absolute, but observer- or context-dependent, would generalize relativity to the structure of events themselves, demanding a radical departure from classical causal inference and potentially providing new axiomatic bases for quantum information and foundations (Cavalcanti et al., 2021, Mukherjee et al., 30 Oct 2025).

A plausible implication is that even minimal operational forms of event absoluteness—such as the existence of certain marginal probabilities and mediation relations—are sufficient to derive no-go constraints, highlighting the robustness of the paradox to assumptions weakening.

7. Broader Impact and Open Problems

The Causal Friendliness Paradox fundamentally alters the landscape of both quantum foundations and causal inference. For quantum foundations, it demonstrates that no extension of classical causal models (including those inspired by quantum causal structure or cyclicity) can accommodate quantum violations under standard metaphysical axioms. For causal inference, it reveals limitations of relying on logical implications between nulls for operational testing. Open questions include the formalization of relational event manifolds, the construction of mediation-preserving quantum causal frameworks, and the search for empirical tests distinguishing between absolute and relational event formalisms (Cavalcanti et al., 2021Mukherjee et al., 30 Oct 2025Yīng et al., 2023).