{"id":"11878d8e-fa0d-46ed-842b-73cefffd3f97","arxiv_id":"2501.11064","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A fine-tuned model with backward conditional probabilities can reproduce EPR and GHZ correlations while preserving Statistical Independence and forbidding superluminal signaling, because the correlations are inserted into the hidden-variable distribution by hand.","lead":"This paper shows that a statistical model with backward influences from measurement outcomes to a hidden variable can reproduce quantum correlations while still forbidding faster-than-light signaling. It offers a new way to think about which assumption must be given up when explaining quantum entanglement.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Missing preparation variable: with P(λ1)=1 the backward-arrow model cannot satisfy Eq. (10), and the GHZ analog forces P(λ0)≤1/2; the central EPR-reproduction claim therefore applies only to a mixture, not to a prepared state.","rationale":"The reader's conditional verdict is appropriate: the paper honestly flags its limitations, and the consistency proof for the ensemble-of-states construction is arithmetically sound. My concern sharpens the reader's weakest assumption into a concrete normalization failure. The central claim in the abstract is that such models 'account for EPR/Bell correlations and, analogously, the GHZ predictions'; as written, the model only accounts for correlations after conditioning on λ_i within a uniform mixture over the four Bell states, and it fails for a fixed prepared state unless an additional preparation variable P is introduced. The GHZ section is even less secure because Eq. (16) with a single λ0 cannot be normalized for a pure GHZ preparation. Since the author explicitly acknowledges the need to distinguish P from λ in Section VI, the right outcome is a conditional acceptance contingent on that extension, not a rejection: the core idea is clearly presented and the ensemble-level math checks out, but the physical scenario of a fixed prepared entangled state is not yet modeled. This does not change the reader's verdict, so I recommend UNCHANGED.","tokens_in":9003,"tokens_out":9296,"duration_ms":97485,"concrete_test":"Set P(λ1)=1 and P(λ_i≠1)=0 in Eqs. (9)-(10), then evaluate P(λ1|a1,a2,α1,α2) for the four outcome pairs of the Bell state |ψ1>. If the constraint P(λ1|a1,a2,α1,α2)=1 for every pair is incompatible with any normalizing constant N in Eq. (10), the model cannot represent a fixed prepared state without adding a preparation variable P. Separately, insert P(λ0)=1 into Eq. (16) and compute the conditional for a GHZ-allowed triple; if it exceeds 1, the GHZ model is not a well-defined probability distribution for a pure GHZ preparation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The construction's λ is not a prepared state but a posterior label over an equiprobable ensemble of all four Bell states. From Eq. (9) and Eq. (10) with N=1, P(λ_i|α1,α2)=1/4 for each i, and conditioning on λ_i yields P_i(a1,a2|α). This is a valid consistency exercise, but it does not model a run in which the experimenter prepares |ψ1>: there P(λ1)=1, so P(λ1|a1,a2,α) must be 1 for every possible outcome; Eq. (10) would require N P1(a1,a2|α)=1 for all a1,a2, impossible because P1 is not constant. The paper's Section VI concedes that a preparation variable P must be separated from λ, but without it the advertised reproduction of EPR correlations holds only for a uniform mixture over λ_i, not for a fixed prepared state. The GHZ extension is starker: with only λ0, Eq. (16) gives P(λ0|allowed)=8P(λ0)(1/4)=2P(λ0), so consistency requires P(λ0)≤1/2; a pure GHZ preparation with P(λ0)=1 makes the conditional exceed 1. The model is therefore internally inconsistent for the fixed-state scenarios it aims to describe, and the lack of a preparation variable is not merely a philosophical worry but a normalization failure.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a hidden-variable model with backward-in-time conditional probabilities that preserves Statistical Independence (SI) as a fine-tuning condition, with the aim of reproducing the EPR/Bell and GHZ correlations while forbidding superluminal signalling. The central construction is Eq. (9), a factorization in which the hidden variable λ acts as a collider depending on settings and outcomes; Eq. (10), which sets the posterior P(λ_i|a1,a2,α1,α2) proportional to the target quantum probability; and Eq. (11), which inverts this to recover the quantum correlations when conditioning on λ_i. The paper also discusses why the model violates Local Causality but satisfies no-signaling, and gives a GHZ analog. The formal algebra is transparent and internally consistent as a consistency exercise, but the advertised physical account is undermined by the absence of a preparation variable: the construction fails for a fixed prepared state and, in the GHZ case, produces an impossible conditional probability.","tokens_in":9304,"tokens_out":9703,"duration_ms":92579,"significance":"If the construction worked as advertised, it would be a notable conceptual contribution: a backward-arrow, SI-preserving model that reproduces Bell-state and GHZ correlations while avoiding superluminal signaling. The equations are simple enough to be checked by hand, and the paper is explicit about several limitations, including the Tsirelson bound and the need for a preparation variable. However, the claimed significance is currently overstated: because the hidden-variable posterior is stipulated to be proportional to the desired quantum probability, the reproduction is built into the model, and the lack of a preparation variable causes concrete normalization failures for prepared states. The paper is therefore best read as a formal consistency exercise rather than a fully specified account of EPR correlations.","major_comments":[{"comment":"The claimed reproduction of the EPR correlations holds only for an equiprobable mixture over the four λ_i, not for a run in which a definite Bell state is prepared. If P(λ_1)=1, then Bayes's theorem gives P(λ_1|a1,a2,α1,α2)=1 for every possible outcome, whereas Eq. (10) with N=4P(λ_1)=4 sets this posterior to 4P_1(a1,a2|α1,α2)=1±a1a2 cos(α1−α2), which takes values between 0 and 2. The author's own Section VI concedes that a separate preparation variable P must be introduced and that this is needed for non-maximally entangled states; until that is done, the central claim of accounting for a prepared EPR pair is not established.","section":"Section IV, Eqs. (10)–(11)"},{"comment":"The GHZ extension suffers from a normalization failure when the GHZ state is the prepared state. With P(a_i|α_i)=1/2, the coefficient in Eq. (16) is N=8P(λ_0). For any GHZ-allowed triple, P_GHZ=1/4, so Eq. (16) gives P(λ_0|a1,a2,a3,α1,α2,α3)=2P(λ_0). If λ_0 is the only hidden-variable value and the state is prepared, P(λ_0)=1 and this conditional probability equals 2, which is impossible. The model is therefore internally inconsistent for the fixed-state GHZ scenario; consistency requires P(λ_0)≤1/2, i.e. an ensemble interpretation rather than a single prepared state.","section":"Section V, Eq. (16)"},{"comment":"The key move is a stipulation, not a derivation: the hidden-variable posterior P(λ_i|a1,a2,α1,α2) is set equal to N times the target quantum probability P_i(a1,a2|α1,α2), and Eq. (11) then recovers that same probability by Bayesian inversion. This is a valid consistency check, but it means the model's ability to reproduce the EPR correlations is built into its definition rather than following from the backward-arrow structure plus SI. The paper should state this more cautiously; as it stands, the abstract's claim that the model 'accounts for' EPR correlations overstates the logical status of the construction.","section":"Section IV, Eq. (10)"}],"minor_comments":[{"comment":"The argument list in P|ψ⟩GHZ (a1,a2,a2|α1,α2,α3) contains a2 twice; the third outcome should be a3.","section":"Section V, Eq. (15)"},{"comment":"The reference list contains two entries both numbered [9] (Evans 2015 and Evans 2018); the citations should be renumbered consistently.","section":"References"},{"comment":"There are typographical errors, e.g. 'meausrement' in Section II and 'simultaneouslky' in the concluding paragraph.","section":"Various"},{"comment":"The text says 'fix N so that P(λ1)/[P(a1|α1)P(a2|α2)] = N'; it would be clearer to state explicitly that SI and the normalization of Eq. (10) force P(λ_i)=1/4 and hence N=1 for the Bell-state case, so the reader can see the consistency conditions.","section":"Section IV, Eq. (11)"},{"comment":"The notation P1;2;3;4 is ambiguous: it should be clear that the subscript ranges over the four Bell states and that the corresponding λ_i are equipped with equal priors.","section":"Section IV, Eq. (10)"}],"recommendation":"major_revision","confidential_remarks":"For the editor: The paper is clearly written and the formal algebra is easy to verify, but the advertised central result is weaker than claimed. The missing preparation variable is not a cosmetic issue; Eq. (16) produces a conditional probability greater than 1 for a prepared GHZ state. If the author can reformulate the model with an explicit preparation variable and show that the renormalized construction still works, the paper could become a valuable conceptual contribution. In its present form, the main result is a consistency exercise for an ensemble of hidden-variable labels rather than an account of a single prepared entangled state."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is worth a serious look if you work on Bell’s theorem and retrocausality. What is new is the specific combination: a collider DAG with backward arrows from settings and outcomes into λ, with Statistical Independence imposed as a fine-tuning condition rather than violated. The Section IV algebra is internally consistent. Equation (9) is a valid factorization, Eq. (10) with N=1 satisfies normalization and SI, and Eq. (11) correctly recovers the quantum probabilities. The no-signaling check is also correct: with P(a_i|α_i)=1/2, changing α2 leaves a1’s distribution unchanged after conditioning on λ. This is a genuine, if modest, addition to the literature, and it is clearly positioned relative to Wood and Spekkens and Price and Wharton.\n\nThe soft spot is the one the author half-concedes in Section VI. The model treats λ as a posterior label over a uniform ensemble of all four Bell states, not as a prepared state. If you actually prepare |ψ1>, so P(λ1)=1, then Eq. (10) sets P(λ1|a1,a2,α) = N P1(a1,a2|α), which cannot equal 1 for all outcomes. The GHZ case is worse: Eq. (16) forces P(λ0|allowed)=2P(λ0), so with P(λ0)=1 the conditional probability exceeds 1. The construction therefore reproduces EPR correlations only for a mixture, not for a fixed prepared state. The author’s suggested separate preparation variable P is the right fix, but as it stands the advertised claim that the model accounts for EPR correlations is overstated.\n\nThis does not sink the paper. It is an honest, clearly written consistency proof for a fine-tuned retrocausal probability model, and the author flags the main limitation himself. What is missing is a derivation rather than a stipulation: Eq. (10) essentially inserts the target quantum probabilities by hand, so the “reproduction” is by construction. The paper is a useful starting point for thinking about what a retrocausal single-world model would need, not a working model.\n\nI would give it a conditional acceptance if I were refereeing: the technical core is sound, the novelty is real, and the limitations are acknowledged. It deserves a serious referee, and should be cited in discussions of measurement independence and retrocausality.","headline":"A clean retrocausal construction that reproduces Bell correlations only as a mixture, not for a fixed prepared state, but still worth a serious referee.","tokens_in":756,"tokens_out":1102,"would_cite":true,"duration_ms":41077,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs probability models with backward-in-time conditional probabilities that preserve Statistical Independence and forbid superluminal signalling while reproducing Bell-state and GHZ correlations.","keywords":["Bell's theorem","EPR correlations","Statistical Independence","Local Causality","backward-in-time conditional probabilities","no superluminal signalling","GHZ correlations","hidden variables"],"falsifier":"In a run prepared in the GHZ state $\\lambda_0$, the model assigns zero probability to any outcome triple with $a_1 a_2 a_3 \\neq (-1)^{\\alpha_1+\\alpha_2+\\alpha_3}$; a single observed occurrence of such a triple would falsify the model.","tokens_in":8724,"feed_emoji":"⚛️","tokens_out":9384,"duration_ms":95218,"temperature":0.7,"pith_summary":"Bell's theorem says no model with both Local Causality and Statistical Independence can reproduce the correlations of entangled quantum states. This paper tries to show that the way out is to keep Statistical Independence and drop Local Causality, using backward-in-time conditional probabilities as the mechanism. It constructs explicit probability models of this kind that reproduce the Bell-state correlations and, by the same recipe with three measurement wings, the GHZ predictions. The payoff, if the construction is accepted, is that a single-world realist account can match quantum correlations and still forbid superluminal signalling.","feed_headline":"Backward time arrows reproduce EPR correlations without signaling","feed_subtitle":"A hidden-variable model keeps Statistical Independence but drops Local Causality, still matching quantum predictions.","key_machinery":"The carrying object is a collider hidden variable $\\lambda$ in a causal diagram: all arrows point into $\\lambda$ from the measurement settings and outcomes. Equation (9) is the defining factorization, and Eq. (10) fixes the backward conditional $P(\\lambda_i\\mid a_1,a_2,\\alpha_1,\\alpha_2)$ to be proportional to the target quantum probability. Statistical Independence then functions as a fine-tuning condition: it demands that, after summing over outcomes, $\\lambda$'s marginal distribution does not depend on $\\alpha_1,\\alpha_2$, which prevents distant settings from influencing local outcomes once $\\lambda$ is conditioned on. Conditioning on a specific $\\lambda_i$ recovers the quantum joint probabilities by Bayes' rule. For the GHZ case, Eq. (16) plays the same role, with $\\lambda_0$ marking the GHZ-allowed outcome triples.","core_discovery":"The paper's central claim is that a hidden variable $\\lambda$ can serve as a collider receiving arrows from both measurement settings and outcomes, with the joint distribution factorizing as $P(a_1,a_2,\\lambda\\mid\\alpha_1,\\alpha_2)=P(a_1\\mid\\alpha_1)P(a_2\\mid\\alpha_2)P(\\lambda\\mid a_1,a_2,\\alpha_1,\\alpha_2)$. Statistical Independence is imposed by requiring the summed marginal $P(\\lambda\\mid\\alpha_1,\\alpha_2)$ to be independent of the settings; this fine-tuning also cancels the signaling paths that would otherwise run through the collider. Choosing $P(\\lambda_i\\mid a_1,a_2,\\alpha_1,\\alpha_2)=N P_i(a_1,a_2\\mid\\alpha_1,\\alpha_2)$, with $P_i$ the quantum probability for Bell state $\\lambda_i$, and conditioning on $\\lambda_i$, yields exactly the quantum correlations. The same construction with a third wing reproduces the GHZ correlations by making $\\lambda=\\lambda_0$ exactly when the outcome triple satisfies the GHZ rule. Thus the model violates Local Causality rather than Statistical Independence and is explicitly non-signalling.","pith_inferences":["I infer that the explanatory weight is carried entirely by Eq. (10): because $\\lambda$'s conditional distribution is set equal to the quantum probability, the construction shows which assumption must fall, namely Local Causality, rather than why the correlations have the quantum form.","I infer that the same method would reproduce any no-signaling distribution, so additional physical constraints on $\\lambda$'s dynamics are needed if the approach is to explain why Nature respects the Tsirelson bound.","I infer that extending the model to non-maximally entangled states requires separating a forward-going preparation variable from the backward-collider $\\lambda$, converting the pure backward-arrow diagram into a hybrid two-time structure."],"forward_implications":["Bell's theorem is bypassed by dropping Local Causality while keeping Statistical Independence.","No superluminal signalling is possible: for each prepared state $\\lambda_i$, each local outcome distribution is $1/2$ regardless of the distant setting.","The GHZ correlations are obtained without attributing pre-existing values to all possible spin components.","The model is necessarily fine-tuned, violating Faithfulness, in line with known results about Bell-compatible models without signaling.","Because it can also reproduce non-quantum no-signaling correlations such as Popescu-Rohrlich boxes, the construction gives no account of the Tsirelson bound."],"supporting_citations":[{"why":"Establishes the Bell inequality that the model must evade.","marker":"[1]"},{"why":"Provides the CHSH form of the inequality used to define EPR correlation violations.","marker":"[3]"},{"why":"Shows that conditioning on a collider induces correlations between otherwise independent variables.","marker":"[7]"},{"why":"Introduces the GHZ state and the GHZ contradiction that the model reproduces.","marker":"[13]"},{"why":"Presents the GHZ correlations in the form used for the three-setting construction.","marker":"[18]"},{"why":"Supplies the Popescu-Rohrlich box as an example of non-quantum no-signaling correlations the model can also reproduce.","marker":"[21]"},{"why":"Suggests that Bell correlations can be recovered by conditioning on a collider variable.","marker":"[23]"},{"why":"Extends the collider and postselection scenario to delayed-choice entanglement swapping.","marker":"[24]"},{"why":"Proves that reproducing EPR correlations without signaling requires Faithfulness violation, motivating fine-tuning via Statistical Independence.","marker":"[27]"}],"fun_headline_variants":["Fine-tuned collider reproduces EPR without signaling","Backward probabilities match quantum correlations, no superluminal","Statistical Independence kept, Local Causality dropped for EPR","GHZ also reproduced by fine-tuned backward hidden variables"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a hidden variable $\\lambda$ can stand for the prepared quantum state while all of its conditional probabilities depend on later settings and outcomes, so stipulating $P(\\lambda_i\\mid a_1,a_2,\\alpha_1,\\alpha_2) = N P_i(a_1,a_2\\mid\\alpha_1,\\alpha_2)$ is a legitimate model-building move rather than a restatement of the correlations one wants to explain.","fun_headline_variants_meta":{"raw":{"variants":["Fine-tuned collider reproduces EPR without signaling","Backward probabilities match quantum correlations, no superluminal","Statistical Independence kept, Local Causality dropped for EPR","GHZ also reproduced by fine-tuned backward hidden variables"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000127,"raw_usage":{"total_tokens":1073,"prompt_tokens":859,"completion_tokens":214,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":475,"completion_tokens_details":{"reasoning_tokens":147}},"tokens_in":475,"tokens_out":214,"duration_ms":3123,"temperature":1.0,"reasoning_tokens":147,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:40:08.074541+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a run prepared in the GHZ state $\\lambda_0$, the model assigns zero probability to any outcome triple with $a_1 a_2 a_3 \\neq (-1)^{\\alpha_1+\\alpha_2+\\alpha_3}$; a single observed occurrence of such a triple would falsify the model.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the Bell inequality that the model must evade."},{"cited_title":"F., Horne, M","cited_arxiv_id":null,"evidence_quote":"Provides the CHSH form of the inequality used to define EPR correlation violations."},{"cited_title":"and Winship, C","cited_arxiv_id":null,"evidence_quote":"Shows that conditioning on a collider induces correlations between otherwise independent variables."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the GHZ state and the GHZ contradiction that the model reproduces."},{"cited_title":"M., Horne, M","cited_arxiv_id":null,"evidence_quote":"Presents the GHZ correlations in the form used for the three-setting construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Popescu-Rohrlich box as an example of non-quantum no-signaling correlations the model can also reproduce."},{"cited_title":"Genovese, M","cited_arxiv_id":null,"evidence_quote":"Suggests that Bell correlations can be recovered by conditioning on a collider variable."},{"cited_title":"[2009]: Causality, 2nd ed., Cambridge Univer- sity Press","cited_arxiv_id":null,"evidence_quote":"Extends the collider and postselection scenario to delayed-choice entanglement swapping."},{"cited_title":"[1996]: Time’s Arrow and Archimedes’ Point: New Directions for the Physics of Time , Oxford Univer- sity Press","cited_arxiv_id":null,"evidence_quote":"Proves that reproducing EPR correlations without signaling requires Faithfulness violation, motivating fine-tuning via Statistical Independence."}],"review_version":1}