{"id":"74cca887-725d-43fd-b031-9b45df106c26","arxiv_id":"2607.13645","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Basis misalignment shrinks n-partite Bell violations by cos(nθ) (periodic failure windows), outcome flipping by (2p−1)^n (one failure threshold), and secret-key generation fails at tighter error tolerances than nonlocality in both models.","lead":"For n-party GHZ states, this paper derives exactly how two kinds of measurement error — rotated measurement bases and random outcome flipping — degrade Bell-inequality violations, and where the violations disappear. It then shows that extracting secret key needs even smaller errors than proving nonlocality, yielding concrete tolerance benchmarks for experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Key-rate benchmark in Sec. 4 uses H(A1|E)=h((1+q_L)/2), but the stated convex-combination attack ('full information about local component') gives H=1−q_L; this changes the claimed χ threshold ≈0.773.","rationale":"Good-faith summary: the paper's core nonlocality results are internally consistent—Eqs. (10)–(13) and the threshold formulas match Table 1, the outcome-flipping law is standard, and the χ1=χ2 step actually generalizes beyond n=3. The load-bearing weak point is the key-rate section. The reader already flags the convex-combination model as unproven against general attacks; my concern is sharper: within the model as described, Eq. (30a) appears to use the wrong entropy. If 'full information about the local component' means Eve's side information determines A1 on local rounds, H(A1|E)=1−q_L, not h((1+q_L)/2). Using h(g) as H overestimates Eve's uncertainty and therefore inflates the key rate; the threshold moves from 0.773 to about 0.83. Because the corrected threshold is still above the Svetlichny nonlocality threshold 1/√2, the qualitative headline survives, but the specific numerical benchmarks (Eqs. 33, 34, Fig. 5) need correction. This reinforces rather than overturns the CONDITIONAL verdict. A direct c-q state construction and entropy computation would settle the matter.","tokens_in":22358,"tokens_out":36079,"duration_ms":301917,"concrete_test":"Build the explicit classical-quantum state for the attack in Eq. (27): with probability q_L, Eve's register encodes A1 exactly (plus a flag 'local'); with probability 1−q_L, A1 is uniformly random and Eve has no information. Compute H(A1|E) directly; the result is 1−q_L. Then solve 1−q_L = h((1−χ)/2), with q_L=(2−2χ)/(2−√2). If the solution is χ≈0.83 rather than 2/(4−√2)≈0.773, revise Eqs. (33)–(34) and the corresponding window half-widths and visibility thresholds. Also rerun the n=3 Table 2 QBER calculation with this entropy expression to confirm the qualitative ordering remains.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4's central quantitative claim—that positive Devetak–Winter key rate requires χ>2/(4−√2)≈0.773—depends on Eq. (30a), H(A1|E)=h((1+q_L)/2). Under the attack model described immediately above Eq. (30a), Eve has full information about the local component (prob. q_L) and zero information about the nonlocal component. The conditional entropy for that c-q state is q_L·0+(1−q_L)·1=1−q_L, not h((1+q_L)/2). For q_L=0.5 these differ by 0.31 bits. Equivalently, for a fixed optimal guessing probability g=(1+q_L)/2, h(g) is an upper bound on H(A1|E), not a lower bound; the flag-bit attack achieves H=1−q_L < h(g). Replacing Eq. (30a) with H=1−q_L and solving 1−q_L=h((1−χ)/2) with q_L=(2−2χ)/(2−√2) gives χ≈0.83 rather than 0.773, shifting Eqs. (33)–(34) and Fig. 5. The qualitative 'key generation is stricter' survives, but the quantitative benchmark is not secured by the text as written. The reader's concern about the convex-combination model is thus compounded by an internal entropy-accounting error. (The χ1=χ2 identification, by contrast, is valid for all n because a valid GHZ key stabilizer contains an odd number of σy, so |sin(rπ/2+nθ)|=cos(nθ).)","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper derives analytic degradation laws for three families of multipartite Bell inequalities (Mermin, Svetlichny, MABK) in n-partite GHZ Bell tests under two models of measurement imperfection: coherent common angular misalignment and incoherent outcome flipping. For coherent misalignment with m faulty parties and θ1=θ2=θ, all three Bell values reduce to S' = S_Q cos(mθ), giving periodic violation windows centered at 2πk/n with widths O(1/n); explicit critical angles are given in Eqs. (14)–(16). For incoherent flipping, all n-partite correlations are attenuated by (2p−1)^n, yielding the single threshold p_cr = ((S_C/S_Q)^{1/n}+1)/2 (Eq. (23)); the MABK threshold approaches (2−√2)/4 for large n, while the Svetlichny threshold decreases to zero. The paper then connects the degraded Svetlichny value to asymptotic Devetak–Winter key rates via a convex-combination attack model (Eqs. (27)–(32)), concluding that positive key generation requires stricter measurement tolerances than genuine multipartite nonlocality certification, with thresholds in Eqs. (33), (34), and Werner visibility thresholds in Eq. (39).","tokens_in":22615,"tokens_out":12941,"duration_ms":113010,"significance":"The Bell-degradation part is a clean, parameter-free contribution: the cos(mθ) and (2p−1)^n laws are derived analytically, reduce correctly in the α=0 case, and directly provide quantitative tolerance benchmarks for multipartite Bell certification. The different scaling with n for MABK versus Svetlichny is a useful and falsifiable prediction. The key-rate part is not yet at the same standard: it depends on an imported, unproven attack model and, as discussed below, contains an entropy-accounting error. With those corrected, the comparison between nonlocality certification and key generation could be a valuable result.","major_comments":[{"comment":"The conditional entropy H(A1|E) does not follow from the stated attack model. The model says Eve has full information on the local component (probability q_L) and a random guess (success 1/2) on the nonlocal component. For that c-q state, H(A1|E) = q_L·0 + (1−q_L)·1 = 1−q_L. The expression h((1+q_L)/2) is the entropy of a binary variable whose optimal guessing probability is (1+q_L)/2; it is an upper bound, not the actual entropy of the described state. Since Eq. (32) is claimed as a lower bound on r_DW, using h((1+q_L)/2) overestimates H(A1|E) and therefore the key rate. Solving 1−q_L = h((1−χ)/2) with q_L = (2−2χ)/(2−√2) gives χ ≈ 0.829 instead of 2/(4−√2) ≈ 0.773; Eqs. (33), (34), (39) and Fig. 5 all shift. The qualitative conclusion that key generation is stricter may survive, but the quantitative benchmark is not supported by the text as written.","section":"Section 4, Eq. (30a)"},{"comment":"The convex-combination decomposition S' = q_L S_C + q_NL S_Q and the claim that Eve's optimal guessing probability on the nonlocal fraction is exactly 1/2 are imported from Refs. [48,49] and are not proven to be optimal against general collective or coherent attacks. The final paragraph concedes that more general eavesdropping strategies are future work, but the abstract and conclusion present the key-rate constraints without this caveat. Please either restrict all key-rate claims explicitly to this model, or provide an argument that the convex-combination attack gives a valid lower bound against a well-defined adversarial class. As written, the 'Devetak–Winter bound' terminology overstates the security scope.","section":"Section 4, Eqs. (27)–(32) and final paragraph"}],"minor_comments":[{"comment":"The equality χ1 = χ2 = cos(nθ) is verified only for n=3 in Table 2. For general n, a one-line proof using the fact that any valid GHZ key stabilizer contains an odd number of σy, hence |sin(rπ/2+nθ)| = cos(nθ), would make the statement self-contained.","section":"Section 4.1 / Table 2"},{"comment":"Please verify Eq. (30a) against Refs. [48,49] and cite the exact derivation. If those references use a different attack model in which h((1+q_L)/2) is exact, the text should say so explicitly; if not, the formula should be corrected per the major comment.","section":"Eq. (30a) attribution"},{"comment":"The notation ∑_{x_i} and ∑_{\\{x_i\\}} is used inconsistently; define the parity condition ∑_i x_i ≡ 0 (mod 2) clearly before first use.","section":"Eqs. (18a)–(18b)"},{"comment":"The caption refers to 'dark and light dashed lines' marking analytical lower and upper critical angles, but the figure rendering does not clearly distinguish them. Please ensure the printed figure is legible and the legend matches the caption.","section":"Figure 2 caption"},{"comment":"A few references carry 2026 volume/page numbers. If these are preprints or accepted papers, please add arXiv identifiers or publication status to aid verification.","section":"References [45,46]"}],"recommendation":"major_revision","confidential_remarks":"I want to stress that the Bell-value degradation analysis in Sec. 3 appears sound and is likely publishable on its own. The entropy-accounting error in Sec. 4 is localized and fixable, but it changes the headline key-rate threshold, so the paper needs a genuine rederivation of the quantitative key-rate claims before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the Bell-degradation analysis in Sec. 3 is sound and worth having. The systematic comparison of coherent misalignment versus incoherent flipping across Mermin, Svetlichny, and MABK for n-partite GHZ states is genuinely useful. The compact laws S′ = S_Q cos(mθ) and S′ = S_Q(2p−1)^n, the 1/n shrinking of the angular windows, and the asymptotic flipping thresholds ((2−√2)/4 for MABK, 0 for Svetlichny) are clean design benchmarks. I checked the threshold algebra against Table 1: Eqs. (14)–(16) and (23)–(26) are consistent. The paper is also honest about the limits of the convex-combination attack model.\n\nThe soft spot is real and load-bearing for Sec. 4. Eq. (30a) does not match the attack model described in the text. If Eve has full information on the local fraction (weight q_L) and no information on the nonlocal fraction, then H(A1|E) = 1 − q_L, not h((1+q_L)/2). The h-form corresponds to a bit-flip channel where Eve's guess is correct with probability (1+q_L)/2 in every round, which is not the same as knowing the local rounds exactly and the nonlocal rounds not at all. Replacing Eq. (30a) with the correct expression moves the positive-key threshold from χ ≈ 0.773 to about 0.83. The qualitative conclusion that key generation is stricter than Svetlichny nonlocality survives, because 0.83 > 0.707, but the quantitative benchmark in Eqs. (33)–(34) and Fig. 5 is not secured as written. This is an internal inconsistency, not just a missing proof.\n\nA smaller issue: the central degradation formulas (10)–(13) are stated without derivation. I verified the special cases, but a referee should ask for the derivation or a notebook. The reader's worry about χ1 = χ2 being verified only for n = 3 is, I think, unfounded: for any valid GHZ key stabilizer containing an odd number of σy, the common-rotation degradation is |sin(rπ/2 + nθ)| = cos(nθ), so the identification is general.\n\nWho is this for? People designing multipartite Bell experiments or device-independent protocols who need concrete alignment and readout tolerances. They will get usable numbers from Sec. 3 and, after a correction, from Sec. 4.\n\nRecommendation: send to peer review. The Bell core is solid, the key-rate issue is fixable, and the paper fills a real gap. It deserves a serious referee, with the expectation of a moderate revision.","headline":"Bell-degradation half is solid and useful; the key-rate half has an internal entropy-accounting error that shifts the headline threshold.","tokens_in":23259,"tokens_out":5151,"would_cite":true,"duration_ms":46230,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","81P45","81P15"],"pacs":[],"model":"deepseek-v4-flash","headline":"Measurement imperfections degrade multipartite Bell violations by two closed-form laws—S′ = S_Q cos(mθ) and S′ = S_Q(2p−1)^n—and the same degraded values force tighter precision for secret-key generation than for nonlocality certification.","keywords":["Bell inequalities","GHZ states","measurement misalignment","outcome flipping","multipartite nonlocality","quantum key distribution","Svetlichny inequality","MABK inequality"],"falsifier":"Perform a three-party Svetlichny test on a GHZ state while independently rotating the measurement bases by a controlled angle θ and, in a separate run, flipping each recorded outcome with controlled probability 1−p; if the observed Bell value departs from S_Q cos(3θ) or from S_Q(2p−1)^3 faster than these formulas, the degradation laws fail. For the key-rate claim, compute Eve's optimal guessing probability under a collective attack on the same state at the same observed Bell value; if it exceeds (1+q_L)/2 from the convex-combination model, the threshold χ > 2/(4−√2) is not security-relevant.","tokens_in":22021,"feed_emoji":"📐","tokens_out":6044,"duration_ms":54354,"temperature":0.7,"pith_summary":"This paper sets out to establish that two physically distinct measurement imperfections—coherent angular misalignment and incoherent outcome flipping—degrade n-partite GHZ Bell violations in closed, predictable forms, and that the same degradation, when fed into a Devetak–Winter key-rate bound, makes secret-key generation demand tighter measurement precision than nonlocality certification alone. If correct, experimenters get quantitative tolerances: under common rotation with m faulty parties, Mermin, Svetlichny, and MABK values fall as S_Q cos(mθ), so violation windows repeat periodically and each narrows as 1/n; under outcome flipping, values fall as S_Q(2p−1)^n, with a single threshold p_cr = ((S_C/S_Q)^{1/n}+1)/2 that rises with n for MABK and odd-n Mermin but falls for Svetlichny. The authors connect the degraded Svetlichny value to a convex-combination attack model and derive a universal positive-key condition χ > 2/(4−√2) ≈ 0.77, stricter than the nonlocality threshold χ > 1/√2. A fair reader should take the paper as providing these benchmark laws, with the caveat that the key-rate conclusions rest on the attack model the paper adopts rather than on a general security proof.","feed_headline":"Two simple laws govern how measurement noise kills Bell violation","feed_subtitle":"Closed-form thresholds show key generation tolerates less measurement error than nonlocality certification itself.","key_machinery":"The load-bearing objects are two degradation mechanisms: coherent rotation of each local measurement by angle θ, which accumulates as a phase mθ in the full correlation and produces S′ = S_Q cos(mθ); and stochastic outcome flipping with correct-readout probability p, which multiplies every n-party correlation by (2p−1)^n and produces S′ = S_Q(2p−1)^n. These formulas, together with the classical bound S_C and quantum maximum S_Q of the Mermin, Svetlichny, and MABK inequalities, fix all certification thresholds. For key rates, the additional machinery is a convex-combination attack model in which the observed Bell value splits into a locally explainable part accessible to Eve and a maximally n","core_discovery":"The central claim is that in n-partite GHZ Bell tests with orthogonality-preserving coherent misalignment affecting m parties, every considered Bell value obeys S′ = S_Q cos(mθ), producing periodic violation windows centered at 2kπ/n with half-width arccos(S_C/S_Q)/n that shrinks as 1/n; while under incoherent outcome flipping with correct-readout probability p, every full correlation is multiplied by (2p−1)^n, yielding S′ = S_Q(2p−1)^n and a single threshold p_cr = ((S_C/S_Q)^{1/n}+1)/2. For the Svetlichny inequality, S_C/S_Q = 1/√2, so its windows are half as wide as Mermin/MABK windows and its flipping threshold decreases with n, while MABK and odd-n Mermin become more tolerant of flippin","pith_inferences":["A testable extension: on a single 3-GHZ source, measure the Bell value versus θ and versus 1−p; the two curves should cross the classical bound at arccos(1/√2)/3 and (2^{−1/6}+1)/2, giving a direct check of both scaling laws and of their claimed universality across the Mermin, MABK, and Svetlichny inequalities.","Because the key-rate condition χ > 2/(4−√2) is derived only inside the convex-combination attack model, the paper's claim that key generation is stricter than nonlocality should be read as a statement about that model; a fully device-independent proof would need to establish the same bound against general collective or coherent attacks, which the paper leaves open.","The paper verifies the equality χ1 = χ2 = cos(nθ) only for n = 3; for n > 3 the key-window positions are a plausible extrapolation, and checking them directly with the stabilizer correlations of the n-GHZ state is the natural next step.","If the scaling laws hold, they suggest a practical calibration strategy: use the sharp 1/n shrinkage of Svetlichny windows as a sensitive in-situ test of basis alignment, and use the flat (2−√2)/4 large-n limit as a readout-error budget that can be spent before mitigation is needed."],"forward_implications":["For coherent misalignment, Bell violation persists inside repeating angular windows, but each window is only arccos(S_C/S_Q)/n wide; the Svetlichny window is always half as wide as the Mermin/MABK window, asymptotically π/(4n) versus π/(2n).","For incoherent outcome flipping, MABK and odd-n Mermin tolerate a finite large-n flip probability approaching (2−√2)/4 ≈ 0.146, while the Svetlichny flip tolerance decays to zero as n grows.","Secret-key generation from the Svetlichny value requires χ > 2/(4−√2) ≈ 0.77, so the key window, with half-width arccos[2/(4−√2)]/n ≈ 0.218π/n, sits strictly inside the nonlocality window of half-width π/(4n); there is a parameter region where genuine multipartite nonlocality is certified but no positive key rate is obtained.","For Werner states with visibility v, the threshold becomes v > 2/[(4−√2)χ], so larger n demands higher state visibility to tolerate the same measurement imperfection."],"fun_headline_variants":["Noise kills Bell violation two ways: periodic windows and a single threshold","Bell noise: coherent misalignment yields windows, flipping yields a cutoff","Secret keys tolerate less measurement error than Bell tests","Measurement error: two laws, one periodic windows, one threshold","Two noise rules: misalignment creates Bell windows, flipping sets a limit"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The key-rate thresholds hold only under the paper's convex-combination attack model—that Eve's information is fully captured by a mixture of a local part she knows completely and a nonlocal part she guesses with probability 1/2—and the paper does not prove this model covers general collective or coherent attacks; a secondary fragile step is the assumption that the key-generation correlation degrades by the same factor cos(nθ) as the Bell value, verified only for n = 3.","fun_headline_variants_meta":{"raw":{"variants":["Noise kills Bell violation two ways: periodic windows and a single threshold","Bell noise: coherent misalignment yields windows, flipping yields a cutoff","Secret keys tolerate less measurement error than Bell tests","Measurement error: two laws, one periodic windows, one threshold","Two noise rules: misalignment creates Bell windows, flipping sets a limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001452,"raw_usage":{"total_tokens":5699,"prompt_tokens":779,"completion_tokens":4920,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":4832}},"tokens_in":523,"tokens_out":4920,"duration_ms":38111,"temperature":1.0,"reasoning_tokens":4832,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T04:35:13.478663+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform a three-party Svetlichny test on a GHZ state while independently rotating the measurement bases by a controlled angle θ and, in a separate run, flipping each recorded outcome with controlled probability 1−p; if the observed Bell value departs from S_Q cos(3θ) or from S_Q(2p−1)^3 faster than these formulas, the degradation laws fail. For the key-rate claim, compute Eve's optimal guessing probability under a collective attack on the same state at the same observed Bell value; if it exceeds (1+q_L)/2 from the convex-combination model, the threshold χ > 2/(4−√2) is not security-relevant.","supporting_citations":[],"review_version":1}