A quantum experiment with joint exogeneity violation

Xingjian Zhang; Yuhao Wang

REVIEW 4 major objections 3 minor 11 references

A quantum experiment with joint exogeneity violation

T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A Bell-type quantum experiment falsifies joint exogeneity: assuming potential outcomes exist yields Balke-Pearl bounds, and the observed correlations violate them while the true causal effect is zero.

desk verdict A thought experiment that credibly isolates joint exogeneity as the falsified assumption in a Bell-type quantum setup, but the proof is not yet reproducible and the title overclaims an experiment. read the letter →

arxiv 2507.22747 v1 pith:SQZBCGRC submitted 2025-07-30 quant-ph math.STphysics.hist-phstat.TH

Yuhao Wang , Xingjian Zhang This is my paper · ORCID

classification quant-phmath.STphysics.hist-phstat.TH

keywords potentialoutcomesjointexogeneityBalke-PearlboundsquantumcausalinferenceBellcorrelationsinstrumentalvariablesstratifiedexclusionrestrictionfatalism

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to show that joint exogeneity—the assumption that treatment assignment is statistically independent of the entire vector of potential outcomes—is not merely philosophically contestable but empirically false. Working in a Bell-type quantum network where a binary instrument Z controls which measurement is made on an entangled state, the authors argue that merely postulating potential outcomes forces a 'stratified exclusion restriction' on their joint distribution. Adding joint exogeneity then makes the average causal effect identifiable only up to the Balke-Pearl bounds. The measured quantum correlations fall outside those bounds: the true causal effect computed from the interventional protocol is 0, while the bounds demand a lower bound of about 0.134. The conclusion is that, in quantum regimes, any potential-outcome model accepting the paper's interventional semantics must give up joint exogeneity, undercutting the fatalist view that outcomes pre-exist the treatment assignment.

What carries the argument

The central object is the interventional distribution $Q_{xz}$ defined by a thought experiment: generate $(X,Z)$ as in the network, then reset them to fixed values $(x,z)$ before measuring $Y$. In quantum theory it satisfies $Q_{xz}(X=x',Y=y | Z=z') = tr[(M_x^{z'} \otimes N_y^x)\rho]$, which yields the stratified exclusion restriction. This restriction replaces the usual individual exclusion restriction, so the only added assumption beyond existence of potential outcomes is joint exogeneity. The argument's second piece is a linear program over the 64-dimensional conditional distribution of $(X, Y(0,0),...,Y(1,1))$ given $Z$, whose feasible region under joint exogeneity and the stratified exclusion restriction reproduces the Balke-Pearl bounds. Feeding the Bell-state probabilities from Eq. (13) into this program produces a lower bound of about 0.1339 for the average causal effect, whereas the interventional protocol gives 0.

What would settle it

A reader could settle the matter by building the photonic experiment described by Eqs. (12) and (13): if observed correlations match Eq. (13) and the interventional protocol yields a true effect of 0 while the linear-program bound is about 0.1339, the violation is real; if an alternative, independently motivated interventional semantics yields an effect inside the bounds, the claim is not.

Watch

Extended reading notes

Core claim

The discovery is that joint exogeneity can be violated in a realizable quantum experiment. In the network of Figure 2a, Z is a randomized instrument, X is a measurement whose setting depends on Z, and Y is a later measurement whose setting depends on X. Assuming potential outcomes $Y(x,z)$ exist, the reset protocol of Eq. (2) yields the stratified exclusion restriction $P(Y(x,1)=y | X=x', Z=z') = P(Y(x,0)=y | X=x', Z=z')$ for all $x,x',y,z'$. Under joint exogeneity, the average causal effect $E(Y(1,z)-Y(0,z))$ is bounded exactly by the Balke-Pearl bounds. Yet for the Bell state and the four Pauli-type measurements in Eq. (12), the observed distribution in Eq. (13) gives a Balke-Pearl lower bound around 0.1339 while the direct interventional computation gives a true effect of 0. The authors read this contradiction as falsifying joint exogeneity, with the stratified exclusion restriction derived rather than assumed.

Load-bearing premise

The argument stands or falls with the paper's choice to define the potential outcome distribution by the reset protocol of Eq. (2)—that setting X and Z in the quantum system means re-running the measurement with predetermined values—so if that semantics is not the right formalization, the contradiction with joint exogeneity does not follow.

Editorial extensions

If this is right

In any experiment where quantum correlations are non-negligible, invoking joint exogeneity together with the existence of potential outcomes can produce bounds that contradict the true effect, so the assumption cannot be treated as a free background condition.
Prior causal conclusions that rely on joint exogeneity in classical settings are not overturned, because the violation appears precisely where a classical description fails.
The result pins down the average causal effect in the Bell configuration as zero, while the Balke-Pearl lower bound from the observed distribution is about 0.1339, so the observed correlations are the empirical witness of the violation.
The authors' reading shifts the philosophical default: potential outcomes should be thought of as generated at or after treatment assignment, with a joint distribution that may depend on the experimenter's choice, rather than as pre-existing attributes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A testable extension: a photonic realization of Eqs. (12)-(13) would let an experimenter directly verify the interventional effect of 0; observing the predicted violation in a real device would make the falsification of joint exogeneity an experimental result rather than a calculation.
If the violation is taken at face value, then future causal analyses involving quantum devices, such as entangled communication networks, should either test joint exogeneity or replace potential outcomes with quantum-compatible interventions; the paper leaves the latter framework unspecified.
The stratified exclusion restriction is derived from the reset protocol, so the same Bell correlations could be reinterpreted as a violation of that restriction if one adopts a different semantics for 'setting X and Z'; the argument thus locates the contradiction specifically at the choice of interventional meaning.
By showing realism fails without any appeal to locality, the result dissolves the realism-locality dilemma in potential-outcome modeling: keeping locality no longer protects realism.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Take this as a preliminary but interesting contribution. It is a thought experiment, not an experiment: no lab data, just a concrete Bell state and measurements. The novel step is to define potential outcomes via a reset protocol that yields a 'stratified exclusion restriction,' then show that adding joint exogeneity forces Balke-Pearl bounds which the quantum correlations violate. This cleanly isolates joint exogeneity as the assumption that fails, and it sharpens the realism-locality debate in the way the authors claim. The philosophical discussion is honest and well-connected to Dawid, Gill, and Robins et al.

The main soft spots are presentation and verification. Proposition 1 is the load-bearing step, but the LP in Eq. (11) is asserted to be analyzable without showing the solution, dual, or vertex. The reported bound 0.1339 is just stated. The notation in Eq. (11) is also garbled—summation indices don't match the subscripts—so a reader cannot verify the target. I don't think the stress-test's specific claim that the objective fixes y01=1 is right; the intended target is clearly P(Y(1,0)=1)-P(Y(0,0)=1). But even so, the whole derivation should be made checkable, e.g., with code or a formal LP solution. The 'true ACE=0' is also outsourced to Chaves et al.; a direct calculation from Eq. (2) would make the paper self-contained.

The bigger conceptual issue is the reset protocol in Eq. (2). It is a modeling choice, not a forced consequence of quantum mechanics. The paper acknowledges this implicitly but the abstract and title overclaim: calling it a 'quantum experiment' and saying joint exogeneity is falsified without qualifying that this is within their operational definition. With that qualification, the reductio is valid and the result is genuinely new in framing—Chaves et al. used a stronger individual exclusion restriction, and here the stratified version suffices.

The paper is not ready as is, but the core idea deserves serious referee time. I would ask for a completed proof of Proposition 1 (or a verifiable computational certificate), a direct computation of ACE=0, a clean rewrite of Eq. (11), and a softer title/abstract. For the right reader—quantum causal inference, potential outcomes, Bell inequalities—this is worth engaging with.

Referee Report

4 major / 3 minor

Summary. The manuscript claims to exhibit the first randomized experiment violating joint exogeneity. It defines potential outcomes Y(x,z) through a reset protocol on a bipartite quantum system, derives the stratified exclusion restriction, and argues that assuming joint exogeneity together with the existence of these potential outcomes yields the Balke-Pearl bounds on the average causal effect. It then specifies a Bell state and measurement settings whose observed conditional distribution violates those bounds while the true ACE is claimed to be 0, and concludes that joint exogeneity is falsified in quantum regimes. The paper also discusses implications for the fatalism view of potential outcomes and for the realism-locality debate.

Significance. The central idea is original and thought-provoking: it connects a foundational assumption in causal inference to quantum violations of instrumental inequalities and does so with a concrete, implementable quantum configuration and no fitted free parameters. If the derivation is made fully valid, the paper would provide the first concrete counterexample to joint exogeneity, lending empirical support to Dawid's critique and sharpening the realism-locality discussion in potential-outcome modelling. The significance is currently tempered by the fact that the load-bearing linear program is not correctly or completely transcribed, the proof of the key proposition is only asserted, and the scope of the conclusion depends on a nonstandard reset-protocol definition of potential outcomes.

major comments (4)

[Section 3, Eq. (11)] The linear program as printed cannot be checked and does not match the stated target. In the objective, the first sum is written over x,y10,y01,y11 for q_{x,y00,y01,1,y11|0}, which leaves y00 unsummed and attempts to sum over y10 although that index is fixed to 1; the correct first term for the ACE in Eq. (6) is sum over x,y00,y01,y11 of q_{x,y00,y01,1,y11|0}. In addition, the two displayed stratified-exclusion constraints only enforce equality for the cases (x=0,y=0) and (x=1,y=1); the cases (x=0,y=1) and (x=1,y=0) are missing. Finally, the non-negativity constraint omits the index y11 and is written as q_{x,y00,y01,y10|z}. Without a corrected LP, the claimed equivalence to the Balke-Pearl bound and the numerical lower bound 0.1339 are unverifiable.
[Section 3, Proposition 1] The proof of Proposition 1 is not provided. The sentence 'This linear programming can be handled analytically, which yields the same bounds as the so-called Balke-Pearl bound' is an assertion rather than a derivation; no dual solution, vertex enumeration, or reproducible code is supplied. Since the contradiction in Section 2 depends entirely on the numerical lower bound 0.1339, the authors should provide either a complete analytical derivation of the Balke-Pearl equivalence or a reproducible computation and an optimality certificate for the LP.
[Section 2, paragraph after Eq. (13)] The statement that 'the true value of ACE in this case is 0' is delegated to Chaves et al. without reproduction, and the definition of ACE used there may not coincide with the reset-protocol definition in this paper. The paper can compute this quantity directly from its own definitions: for rho = |Phi-><Phi-|, the reduced state on Y is the maximally mixed state I/2, so P(Y(x,z)=1) = 1/2 for all x,z and the ACE is 0. This computation should be included to make the contradiction self-contained and to fix the target quantity.
[Abstract and Section 4] The conclusion that the experiment 'falsifies' joint exogeneity is only as strong as the modeling choice that the reset protocol in Eq. (2) correctly defines the potential outcomes Y(x,z) in a quantum system. A reader who does not accept this counterfactual interpretation could instead take the contradiction as showing that the potential outcome framework, or the stratified exclusion restriction derived from Eq. (2), is inapplicable to this quantum experiment. The authors should state the theorem explicitly as conditional on the reset-protocol definition and discuss this limitation, rather than claiming unconditional falsification of joint exogeneity.

minor comments (3)

[Section 3, Eq. (4)] In the consistency equation, the third term is printed as X(1-Z)Y(0,1); it should be X(1-Z)Y(1,0).
[Section 3, Eq. (6)] The phrase 'where the last inequality is directly from Assumption 1' should read 'last equality', since the displayed relation is an equality.
[Throughout] There are several typographical errors, including 'Prelimiary' in the header, 'practial' in the abstract, 'attidude' in the introduction, and 'distribtuion' in Section 2; these should be corrected in a final version.

Circularity Check

1 steps flagged · score 2.0 of 10

No serious circularity; one definitional step: the stratified exclusion restriction is built into the reset-protocol definition of potential outcomes, so the joint-exogeneity violation is conditional on that modeling choice.

self definitional [Section 2, Eqs. (2)-(3)]
"Write Qxz as the distribution of random variables X, Y, Z under this interventional experiment, then by definition, Qxz(Y = y | X = x′, Z= z′) = P(Y (x, z) = y | X = x′, Z= z′). Moreover, according to standard results in quantum theory, Qxz(X = x′, Y= y | Z = z′) = tr[(M z′ x′ ⊗ N x y )ρ], where the right hand side does not depend on z."

The stratified exclusion restriction (3) is presented as derived from the existence of potential outcomes, but it is actually an identity fixed by the definition of P(Y(x,z)|...) via Eq. (2): the right-hand side of Eq. (2) is independent of z. Hence Eq. (3) holds by construction for every potential outcome defined through the reset protocol. The Balke-Pearl bounds in Proposition 1 are therefore derived under a constraint that is definitionally true, and the resulting violation of joint exogeneity is conditional on this specific reset-protocol formalization, not on existence of potential outcomes alone. The paper is transparent that Eq.

full rationale

The paper is a reductio ad absurdum, not an estimation exercise: the Bell state, measurement operators, and observed pXY|Z are fixed in advance, and no parameter is fitted to the LP output, so the reported bound (about 0.1339) is not a fitted input disguised as a prediction. The main caveat is that the 'stratified exclusion restriction' (3) is not a consequence of mere existence of potential outcomes; it is installed by the reset-protocol definition of P(Y(x,z)|...) in Eq. (2), whose right-hand side is independent of z. Thus the claimed falsification of joint exogeneity is conditional on that definitional choice; the paper is transparent about this, but the wording 'by merely assuming the existence of potential outcomes' overstates the derivation. Proposition 1's proof is only sketched ('This linear programming can be handled analytically'), and the LP target in Eq. (11) is printed in a way that does not match Eq. (6), as the first sum appears to fix y10=1 while also summing over y10; both are reproducibility and correctness gaps, not circularity. The true ACE=0 is imported from Chaves et al. (2018), an external result, and the only self-citation (Zhang and Wang 2024) is a forward-looking pointer with no load-bearing role. Overall, no step reduces the paper's central claim to its own input by construction; score 2 reflects the one self-definitional constraint.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No constants are fitted to data; the Bell state and measurements are chosen from Chaves et al. The central claim depends on the existence of potential outcomes and on the specific reset protocol; neither is independently evidenced outside the paper. No new physical entities are introduced.

assumptions (3)

domain assumption Potential outcomes {Y(x,z)} exist as jointly distributed random variables for x,z in {0,1}.
Used throughout Sections 2 and 3; required to even state joint exogeneity and to define ACE. The paper argues that this assumption plus the thought experiment yields the contradiction.
ad hoc to paper The 'reset' thought experiment correctly defines the interventional distribution via Eq. (2): Qxz(X=x',Y=y|Z=z') = tr[(M^{z'}_{x'}⊗N^x_y)ρ].
This is the bridge between quantum mechanics and potential outcomes. If this interventional protocol is not accepted, the stratified exclusion restriction (Eq. 3) and the subsequent LP bounds do not follow.
domain assumption Standard quantum measurement theory (Born rule, tensor product structure, completeness of measurements) applies to the network in Fig. 2a.
Used to compute Eq. (1) and Eq. (2), and to compute the probabilities in Eq. (13).

how reviews work

0 comments

Cite this review

Pith. "Pith review of A quantum experiment with joint exogeneity violation." pith.science (2026). https://pith.science/paper/SQZBCGRC

@misc{pith2026250722747,
  author       = {Pith},
  title        = {Pith review of: A quantum experiment with joint exogeneity violation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SQZBCGRC}},
  note         = {Machine review of arXiv:2507.22747}
}

read the original abstract

In randomized experiments, the assumption of potential outcomes is usually accompanied by the \emph{joint exogeneity} assumption. Although joint exogeneity has faced criticism as a counterfactual assumption since its proposal, no evidence has yet demonstrated its violation in randomized experiments. In this paper, we reveal such a violation in a quantum experiment, thereby falsifying this assumption, at least in regimes where classical physics cannot provide a complete description. We further discuss its implications for potential outcome modelling, from both practial and philosophical perspectives.

Figures

Figures reproduced from arXiv: 2507.22747 by the authors.

**Figure 2.** (a) A directed acyclic graph representing the quantum communication network considered [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 9 canonical work pages

[1]

Bounds on treatment effects from studies with imperfect compliance

Alexander Balke and Judea Pearl. Bounds on treatment effects from studies with imperfect compliance. Journal of the American statistical Association, 92 0 (439): 0 1171--1176, 1997

work page 1997
[2]

Quantum violation of an instrumental test

Rafael Chaves, Gonzalo Carvacho, Iris Agresti, Valerio Di Giulio, Leandro Aolita, Sandro Giacomini, and Fabio Sciarrino. Quantum violation of an instrumental test. Nature Physics, 14 0 (3): 0 291--296, 2018

work page 2018
[3]

Causal inference without counterfactuals

A Philip Dawid. Causal inference without counterfactuals. Journal of the American statistical Association, 95 0 (450): 0 407--424, 2000

work page 2000
[4]

Statistics, causality and bell’s theorem

Richard D Gill. Statistics, causality and bell’s theorem. Statistical Science, 29 0 (4): 0 512--528, 2014

work page 2014
[5]

Richard Guo

F. Richard Guo. Likelihood Analysis of Causal Models. PhD thesis, University of Washington, Seattle, WA, USA, 2021. URL https://digital.lib.washington.edu/researchworks/items/f933f3b1-951d-49c6-890d-471674f27679. PhD thesis

work page 2021
[6]

Causal inference in statistics, social, and biomedical sciences

Guido W Imbens and Donald B Rubin. Causal inference in statistics, social, and biomedical sciences. Cambridge university press, 2015

2015
[7]

Micius quantum experiments in space

Chao-Yang Lu, Yuan Cao, Cheng-Zhi Peng, and Jian-Wei Pan. Micius quantum experiments in space. Reviews of Modern Physics, 94 0 (3): 0 035001, 2022

work page 2022
[8]

A proof of bell's inequality in quantum mechanics using causal interactions

James M Robins, Tyler J VanderWeele, and Richard D Gill. A proof of bell's inequality in quantum mechanics using causal interactions. Scandinavian Journal of Statistics, 42 0 (2): 0 329--335, 2015

work page 2015

Show all 11 references

[9]

Causal inference using potential outcomes: Design, modeling, decisions

Donald B Rubin. Causal inference using potential outcomes: Design, modeling, decisions. Journal of the American statistical Association, 100 0 (469): 0 322--331, 2005

2005
[10]

Partial identification of the average treatment effect using instrumental variables: review of methods for binary instruments, treatments, and outcomes

Sonja A Swanson, Miguel A Hern \'a n, Matthew Miller, James M Robins, and Thomas S Richardson. Partial identification of the average treatment effect using instrumental variables: review of methods for binary instruments, treatments, and outcomes. Journal of the American Stati...

2018
[11]

On the physics of nested markov models: a generalized probabilistic theory perspective

Xingjian Zhang and Yuhao Wang. On the physics of nested markov models: a generalized probabilistic theory perspective. arXiv preprint arXiv:2411.11614, 2024

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Bounds on treatment effects from studies with imperfect compliance

Alexander Balke and Judea Pearl. Bounds on treatment effects from studies with imperfect compliance. Journal of the American statistical Association, 92 0 (439): 0 1171--1176, 1997

work page 1997

[2] [2]

Quantum violation of an instrumental test

Rafael Chaves, Gonzalo Carvacho, Iris Agresti, Valerio Di Giulio, Leandro Aolita, Sandro Giacomini, and Fabio Sciarrino. Quantum violation of an instrumental test. Nature Physics, 14 0 (3): 0 291--296, 2018

work page 2018

[3] [3]

Causal inference without counterfactuals

A Philip Dawid. Causal inference without counterfactuals. Journal of the American statistical Association, 95 0 (450): 0 407--424, 2000

work page 2000

[4] [4]

Statistics, causality and bell’s theorem

Richard D Gill. Statistics, causality and bell’s theorem. Statistical Science, 29 0 (4): 0 512--528, 2014

work page 2014

[5] [5]

Richard Guo

F. Richard Guo. Likelihood Analysis of Causal Models. PhD thesis, University of Washington, Seattle, WA, USA, 2021. URL https://digital.lib.washington.edu/researchworks/items/f933f3b1-951d-49c6-890d-471674f27679. PhD thesis

work page 2021

[6] [6]

Causal inference in statistics, social, and biomedical sciences

Guido W Imbens and Donald B Rubin. Causal inference in statistics, social, and biomedical sciences. Cambridge university press, 2015

2015

[7] [7]

Micius quantum experiments in space

Chao-Yang Lu, Yuan Cao, Cheng-Zhi Peng, and Jian-Wei Pan. Micius quantum experiments in space. Reviews of Modern Physics, 94 0 (3): 0 035001, 2022

work page 2022

[8] [8]

A proof of bell's inequality in quantum mechanics using causal interactions

James M Robins, Tyler J VanderWeele, and Richard D Gill. A proof of bell's inequality in quantum mechanics using causal interactions. Scandinavian Journal of Statistics, 42 0 (2): 0 329--335, 2015

work page 2015