{"id":"28d3abba-8515-414c-b631-a79d5aef0086","arxiv_id":"2506.17020","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Linear min-entropy accumulation is claimed for time-ordered no-signalling adversaries in Bell tests with monogamous non-local games, while most bipartite pseudo-telepathy games are claimed not to certify randomness against such adversaries.","lead":"This paper studies how much certified randomness can be extracted from many runs of a Bell experiment when the adversary obeys only no-signalling, not quantum mechanics. It argues that randomness does accumulate linearly for certain Bell tests, contrary to earlier small-n numerics, and that many pseudo-telepathy games are useless for this purpose.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main theorem is false as stated: a two-branch hidden-variable TONS strategy has per-round chain value 17/18 yet gives Eve constant guessing probability, because the Azuma step only uses marginal value, not history-dependent conditional win rates.","rationale":"The reader's concerns about Lemma 1 and Lemma 4 are real, but the central claim fails earlier. The main theorem's proof uses Azuma-Hoeffding on the per-round indicator V_i, controlled only by the unconditional marginal Bell value omega*. In TONS, the device may be a hidden-variable mixture: one branch deterministic (value 8/9, Eve guesses perfectly), one branch perfect F-box (value 1, Eve guesses randomly). The mixture satisfies the per-round marginal constraint at omega*=17/18 yet gives Eve success probability at least 1/2. This is not a question of missing constants; it falsifies Theorem 1/3 as stated. The paper's security statement may be salvageable by conditioning on a passed Bell test with threshold t>8/9 and proving a correct concentration bound for history-dependent devices, or by strengthening the hypothesis to require each conditional round behavior given the past to have value omega*, but neither is in the manuscript. The reader's REJECT verdict remains appropriate, albeit for a different reason than the one identified as weakest.","tokens_in":22461,"tokens_out":24559,"duration_ms":266205,"concrete_test":"Verify the two-branch strategy for n=10 in the LP (4): H=0 branch uses a_i=b_i=0 and e_i=0; H=1 branch uses the F-box with a_i xor b_i = F(x_i,y_i), F=1 only at (0,2), and e_i random. Check: (i) per-round chain value is 17/18; (ii) TONS equations (Def. 1) hold; (iii) the objective equals 1/2 + 2^{-(n+1)}. All three follow directly from the definitions, and a standard LP solver will certify the point feasible with objective above e^{-Omega(n)}, contradicting Theorem 3. Separately, the abort probability for threshold 11n/12 is about 1/2, directly refuting Eq. (A15)'s Azuma bound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing problem is in Appendix A around Eq. (A15). The proof asserts that if each round's marginal Alice-Bob behavior has chain-game value omega*, then the number of winning rounds is at least (omega*-kappa)n with probability 1-2 exp(-n kappa^2/2) by Azuma-Hoeffding. But TONS (Def. 1, Eq. (2)) permits arbitrary correlations across rounds; only the unconditional round marginal is constrained, not E[V_i | past]. Thus Azuma's drift term is uncontrolled. This is not a technical nit: the theorem as stated is falsified by a valid TONS strategy. Let H in {0,1} be uniform. If H=0, use the deterministic strategy a_i=b_i=0, whose winning probability on the m=3 chain game is 8/9, and Eve outputs 0 always, so she guesses all n outputs with probability 1. If H=1, use the F-box P(a,b|x,y)=1/2 whenever a xor b =1 for (x,y)=(0,2) and a xor b=0 on the other five constrained pairs (automatic pairs arbitrary), which wins with probability 1 and has uniform marginals; Eve's success probability is 2^{-n}. The mixture has per-round marginal value (8/9+1)/2=17/18, below the quantum value (4+sqrt(3))/6, satisfies all TONS constraints, and gives Pg >= 1/2 + 2^{-(n+1)}. Hence Theorem 1/3's Pg <= e^{-Omega(n)} fails; the Azuma concentration step is the precise false premise.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses the accumulation of device-independent quantum randomness against time-ordered no-signalling (TONS) adversaries. It claims that for any non-local game satisfying a monogamy-of-nonlocality property, Eve's guessing probability decays exponentially in the number of rounds n, and it gives a detailed constant computation for the three-input chained Bell game (Theorems 1 and 3, Appendix A). It further claims that most bipartite pseudotelepathy games do not certify randomness against no-signalling adversaries, via an explicit attack construction (Theorems 2 and 6, Appendix B). Finally, it derives an analytical expression for the min-entropy of the three-input chained inequality against quantum and no-signalling adversaries (Appendix C). The main proof strategy is to upper-bound the guessing probability by the value of a related tripartite guessing game, apply no-signalling parallel-repetition theorems, and use a concentration argument on the number of winning rounds.","tokens_in":22834,"tokens_out":5584,"duration_ms":60749,"significance":"If the accumulation theorem were correct, it would resolve an open question about linear min-entropy accumulation in the TONS scenario and would support the practical use of a single pair of devices for device-independent randomness generation against no-signalling adversaries. The analytical min-entropy comparison for the three-input chained inequality in Appendix C is a potentially useful result in its own right, as analytical bounds for this inequality have been lacking. However, the central accumulation theorem is false as stated, and the pseudotelepathy attack construction in Appendix B is not a valid normalized no-signalling strategy. The paper therefore does not currently provide a reliable proof of its advertised claims, although the Appendix C derivation and the clear formulation of the TONS model are valuable elements that could be reused in a corrected version.","major_comments":[{"comment":"The Azuma-Hoeffding concentration step is invalid for TONS. The TONS constraints in Definition 1, Eq. (2), constrain only the unconditional marginal of each round; they impose no bound on E[V_i | past] for the indicator V_i of a chain-game win in round i. Therefore the drift of the associated martingale is uncontrolled and Eq. (A15) does not follow. This is not a technical gap: the claim of Theorem 3 is false as stated. Consider the following valid TONS strategy for the m=3 chain game. Let H be uniform on {0,1}. If H=0, use the deterministic strategy a_i=b_i=0 for all i, which wins on 8 of the 9 input pairs and hence has per-round value 8/9. If H=1, use per-round F-box behaviors with P(a,b|x,y)=1/2 for a⊕b=1 at (x,y)=(0,2) and a⊕b=0 on the other five constrained pairs, which win with probability 1 and have uniform marginals. Each branch is TONS, so the mixture is TONS, and the per-round Alice-Bob marginal has chain-game value 17/18, which is below the quantum value (4+√3)/6 and hence inside the range stated in Eq. (A17). Yet Eve can simply output the all-zero string: her success probability is (1/2)·1 + (1/2)·2^{-n} = 1/2 + 2^{-(n+1)}, contradicting the asserted P_g(x*,ω*) ≤ e^{-Ω(n)}.","section":"Appendix A, Eq. (A15)"},{"comment":"The claimed ε-almost-no-signalling value of the guessing game, (8+10ε)/9 for ε ∈ [0,1/10], is asserted to follow from the stated dual LP via strong duality, but no dual feasible solution, no certificate, and no reproducible code are provided. This value is load-bearing: it determines ω_NS(G_g)=8/9, the constant α(G_g) in Lemma 2, the admissible range of ω* in Eq. (A17), and the exponential rate in Eq. (A22). Without a checkable dual solution or an explicit sensitivity argument, the reader cannot verify the claimed constants.","section":"Appendix A, Lemma 1, Eqs. (A5)-(A6)"},{"comment":"The theorems are stated for every non-local game with the monogamy-of-nonlocality property, but the proof in Appendix A explicitly fixes the m=3 chained game and only asserts without argument that 'the statement applies generally' to all such games. No reduction from an arbitrary monogamy game to the chained game is given, and no definition of the relevant monogamy property is supplied that would make the reduction checkable. The advertised generality of Theorem 1 is therefore unsupported.","section":"Theorems 1 and 3, Appendix A"},{"comment":"The bipartite behavior constructed in Lemma 4 is not a normalized probability distribution. Eq. (B8) sets P(f)(u|w,x,y)=1 for every u∈x with ⟨u|w⟩≠0 and 0 otherwise. For a fixed basis x and fixed w∈y, the number of vectors u∈x with ⟨u|w⟩≠0 can be greater than 1, so the sum over u∈x of P(f)(u,w|x,y) equals k·f(w) for k≥1 rather than f(w), and normalization fails. The statement that 'there must exist some u′' such that the conditional is 1 only shows k≥1, not k=1. Consequently the no-signalling attack in Theorem 6, including the perfect-guessing conclusion of Eq. (B4), is not established.","section":"Appendix B, Lemma 4"}],"minor_comments":[{"comment":"The text states that the minimum input probability for the tripartite guessing game is π_min = 1/33, but the game has 3^3 = 27 equally likely input triples, so the correct value appears to be 1/27. Since the constant µ in Eq. (A11) depends on π_min, this typo should be corrected and the constants re-derived.","section":"Appendix A, Eq. (A11)"},{"comment":"In the second displayed line of Eq. (A4), the right-hand side sums over a, but for the marginal of Bob's output the sum should be over b. As written, the condition is not the intended almost-no-signalling inequality for Bob's marginal.","section":"Definition 2, Eq. (A4)"},{"comment":"The axes and curves in Figure 2 are not described fully in the text. Please state explicitly in the caption that panel (a) plots P_g(A0|E) versus w_Q, panel (b) plots H_min versus w, and which NPA level is used for the numerical upper-bound curve.","section":"Figure 2"}],"recommendation":"reject","confidential_remarks":"The counterexample described in major comment 1 is a genuine falsification of Theorem 3 as stated, and the normalization failure in Lemma 4 is also real. The Appendix C analytical work appears potentially salvageable, but the central claims of the paper are not currently correct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Ravi,\n\nI read this paper carefully. The short version: the central claim does not hold as written. The proof of Theorem 1/3 uses Azuma-Hoeffding on the number of winning rounds, but TONS (Def. 1, Eq. (4)) only fixes the round-wise input-conditional marginal. It does not fix E[V_i | past outputs], which is what Azuma needs. This is not a technical nit. Here is a concrete TONS strategy: pick a hidden bit H uniformly. If H=0, both players output 0 every round; the 3-input chain win probability is 8/9. If H=1, use the F-box with P(a,b|x,y)=1/2 for the constrained winning correlations; it wins with probability 1 and has uniform marginals. The unconditional per-round chain value is 17/18, below the quantum value, and all TONS constraints are satisfied. Eve, who knows H, guesses all n outputs with probability 1 in the H=0 branch, so Pg >= 1/2 + 2^{-(n+1)}. That contradicts the exponential decay claimed in Theorem 3. The flaw is specifically around Eq. (A15): the drift should be conditional on the past, and it isn't.\n\nThere are other load-bearing gaps. Theorem 1 is stated for all monogamy games but the proof is only for the 3-input chain; that would be a generality gap if the chain proof were sound. Lemma 1's almost-NS value (8+10ε)/9 is asserted from an LP dual that is not shown; not verifiable without a certificate or code. Lemma 4 in the pseudo-telepathy appendix constructs a conditional distribution P(u|w,x,y) that is 1 for every u non-orthogonal to w; for a basis x with several such u, the sum exceeds 1, so the object is not normalized and is not a no-signalling behavior. That kills Theorem 2/6 as written.\n\nWhat is good: the paper tackles an important open question and the idea of using parallel-repetition plus concentration is the right instinct. The analytical chained-game guessing probability in Appendix C is a useful formula, though its optimality rests on a level-2 NPA match rather than a proof. The graph-theoretic attack template for PT games is conceptually interesting and might be repairable. The citation pattern seems fine; the self-references are relevant.\n\nWho should read it: people working on DI randomness against NS adversaries will want to see the counterexample; it clarifies the gap between 'round marginal value' and 'conditional drift.' But the paper as a whole should not be the basis of further results.\n\nRecommendation: reject, but invite a resubmission. The authors need to either strengthen the TONS assumption to include a conditional drift condition (which would change the model), or restrict the theorem to a class where Azuma applies. Lemma 4 needs a real construction. Appendix C could be split into a separate, defensible paper.","headline":"The main accumulation theorem is false as stated: TONS permits a two-branch strategy with per-round value 17/18 but constant guessing probability, because the Azuma step relies on a conditional drift the constraints never provide.","tokens_in":23300,"tokens_out":8803,"would_cite":false,"duration_ms":93497,"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":"Raw Bell-test randomness does accumulate linearly against time-ordered no-signalling adversaries, while many pseudotelepathy games certify none.","keywords":["device-independent randomness","time-ordered no-signalling","no-signalling adversary","chained Bell inequality","min-entropy accumulation","monogamy of nonlocality","pseudotelepathy games","guessing probability"],"falsifier":"Take the dual linear program in Eqs. (A5)-(A6) and solve it to certified precision for $\\epsilon=0,0.05,0.1$; if any optimum exceeds $(8+10\\epsilon)/9$, Lemma 1 is false and the exponential bound in the TONS proof does not follow. Separately, instantiate the attack construction in the pseudotelepathy section on the Magic Square game and check non-negativity, normalisation, and every no-signalling condition; a violation would refute the claimed universality of that attack.","tokens_in":22223,"feed_emoji":"🎲","tokens_out":19415,"duration_ms":169553,"temperature":0.7,"pith_summary":"The paper's central claim is that raw randomness from a Bell test does accumulate linearly against the most general causality-only eavesdropper: a no-signalling adversary in the time-ordered no-signalling (TONS) setting, where past rounds may influence future behaviour but no signal passes between separated devices. Earlier numerics for up to five rounds suggested that the guessing probability might not decay exponentially; the paper argues those were finite-size effects and proves an exponential bound $e^{-\\Omega(n)}$ for large $n$ for every game with monogamy of nonlocality, working out explicit rates for the three-input chained Bell inequality. The second claim is that this property is not universal: for bipartite pseudotelepathy games won by maximally entangled states of dimension $d\\ge 3$ and satisfying a graph-theoretic assumption, even maximal violation leaves Alice's output perfectly guessable by a no-signalling adversary. If both claims hold, protocol designers can use chained or CHSH-type tests for causality-based security but must avoid pseudotelepathy games. A side result gives exact analytical min-entropy formulas for the chained Bell test against both quantum and no-signalling adversaries, useful for rate estimation.","feed_headline":"Randomness accumulates linearly against no-signalling eavesdroppers","feed_subtitle":"Sequential Bell tests make guessing decay exponentially; pseudotelepathy games certify none.","key_machinery":"The load-bearing object is the tripartite guessing game $G_g$ built from a bipartite non-local game $G$: in each round Alice, Bob, and Eve receive inputs $x,y,z$, and win if Alice and Bob satisfy $G$'s winning condition while Eve's output matches Alice's whenever $z=x$. Monogamy of nonlocality is the property that near-maximal winning probability in $G$ forces Alice's output to be essentially uncorrelated with any third party; it manifests here as a gap between $\\omega_{NS}(G)$ and the no-signalling value of $G_g$ (equal to $8/9$ for the three-input chained game). The proof leverages two quantitative tools: the $\\epsilon$-almost-no-signalling stability of the guessing-game value (Lemma 1, value $(8+10\\epsilon)/9$) and the $t$-out-of-$n$ parallel-repetition concentration theorem for no-signalling games, which converts a constant win probability below one into exponential decay over $n$ independent rounds. Standard martingale inequalities and a generalized tail bound connect the game value to the protocol's abort condition and to the subset of rounds in which Eve's input equals Alice's. For the negative result on pseudotelepathy games, the machinery is different: weak Kochen-Specker sets (vector sets with no consistent binary outcome assignment on every basis), the fractional stable set polytope of the orthogonality graph (the set of non-negative vertex assignments obeying $u_i+u_j\\le 1$ on every edge; its vertices are known to be $\\{0,1/2,1\\}$-valued), and a construction turning such assignments into no-signalling bipartite behaviours that, averaged over Eve's choices, allow perfect guessing.","core_discovery":"The discovery is that the asymptotic behaviour of randomness accumulation in the TONS scenario is controlled by whether the non-local game exhibits monogamy of nonlocality. For such games, with the argument made explicit for the $m=3$ chained Bell inequality, Eve's guessing probability satisfies $P_g(x^*,\\omega^*)\\le e^{-\\Omega(n)}$ whenever each round's Alice-Bob marginal achieves value $\\omega^*$; the proof constructs the associated tripartite guessing game, computes its no-signalling value as $8/9$, shows that this value is stable under $\\epsilon$-almost-no-signalling perturbations with slope $10/9$ on $\\epsilon\\in[0,1/10]$, and then applies $t$-out-of-$n$ parallel-repetition concentration. For bipartite pseudotelepathy games the opposite holds: using local assignments valued in $\\{0,1/2,1\\}$ from the fractional stable set polytope, the paper constructs no-signalling tripartite strategies that win the game with probability one while Eve guesses Alice's output perfectly, so the single-round guessing probability equals $1$. Finally, for the chained Bell expression the paper derives a tight guessing-probability curve against quantum adversaries and the linear formula $P_g=2-w_{NS}/4$ against no-signalling adversaries, displaying the quantitative gap between the two.","pith_inferences":["If Lemma 1 is supplied with an explicit dual certificate, the same proof template would give ready exponential rates for CHSH and any other monogamous game, enabling a rate comparison among candidate Bell tests.","The small-n anomaly reported for $n\\le 5$ likely reflects large constants in the exponential bound rather than absence of accumulation; practical protocols may need sufficiently many rounds before the linear behaviour is visible.","The normalisation issue in the pseudotelepathy construction suggests the universal attack may need a repaired argument, although the Magic Square case is independently grounded in earlier work.","The quantum-versus-no-signalling min-entropy gap for the chained test could serve as a finite-size benchmark for experimental demonstrations of causality-based security."],"forward_implications":["For chained-Bell or CHSH-style tests run in sequence, the raw output string carries at least linearly many bits of min-entropy against a no-signalling adversary, so entropy accumulation is linear rather than stuck at a constant.","The analytical formulas give protocol designers closed-form rates: $H_{\\min}=-\\log_2(2-w_{NS}/4)$ against a no-signalling adversary for the three-input chained test, and the corresponding closed form against a quantum adversary.","Bipartite pseudotelepathy games satisfying the stated assumptions cannot be used for device-independent randomness against NS adversaries, because maximal violation gives single-round guessing probability $1$.","In those pseudotelepathy scenarios, the non-local face of the no-signalling polytope reached by quantum correlations has dimension at least $d-1$, so quantum behaviours sit on positive-dimensional faces rather than vertices.","The three-input chained Bell inequality is certified suitable for both quantum and no-signalling security, supporting its use in device-independent randomness and key distribution protocols."],"supporting_citations":[{"why":"Defines the ABNS and TONS scenarios and supplies the small-n numerics for noisy PR boxes that the linear-accumulation result extends.","marker":"[1]"},{"why":"Defines the chained Bell inequality for which the paper computes explicit guessing-probability rates.","marker":"[19]"},{"why":"Introduces the guessing-game formalism used to define the tripartite game G_g at the core of the proof.","marker":"[4]"},{"why":"Provides the t-out-of-n parallel-repetition concentration theorem for no-signalling game value, giving the exponential decay.","marker":"[2]"},{"why":"Supplies the perturbation bound showing that epsilon-almost-no-signalling changes the game value by at most a linear amount.","marker":"[3]"},{"why":"Gives the generalized tail bound used to restrict to the rounds in which Eve's input matches Alice's.","marker":"[58]"},{"why":"Establishes the correspondence between pseudotelepathy games and weak Kochen-Specker sets that underpins Theorem 2.","marker":"[13]"},{"why":"Shows the Magic Square game has bound randomness against NS adversaries, the phenomenon Theorem 2 generalises.","marker":"[16]"},{"why":"Together with [16], shows strong monogamy failures in bipartite no-signalling theories that motivate the PT-game attack.","marker":"[26]"},{"why":"Supplies the tight quantum strategy and self-testing results for the chained Bell inequality used to set the admissible violation range.","marker":"[59]"}],"fun_headline_variants":["Linear randomness accumulation holds for large-n Bell tests","No-signalling entropy scales linearly in sequential Bell tests","Chained Bell test: entropy grows linearly against no-signalling","Pseudotelepathy games let Eve guess perfectly in Bell tests","Large-n Bell tests restore linear entropy against no-signalling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof rests on the single asserted value in Lemma 1: the almost-no-signalling value of the tripartite guessing game is $(8+10\\epsilon)/9$ for $\\epsilon\\in[0,1/10]$, asserted to follow from a linear program but shown without a dual certificate; if the true optimum is any larger, the exponential rate and the admissible Bell-value range no longer follow.","fun_headline_variants_meta":{"raw":{"variants":["Linear randomness accumulation holds for large-n Bell tests","No-signalling entropy scales linearly in sequential Bell tests","Chained Bell test: entropy grows linearly against no-signalling","Pseudotelepathy games let Eve guess perfectly in Bell tests","Large-n Bell tests restore linear entropy against no-signalling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000217,"raw_usage":{"total_tokens":1466,"prompt_tokens":1006,"completion_tokens":460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":622,"completion_tokens_details":{"reasoning_tokens":380}},"tokens_in":622,"tokens_out":460,"duration_ms":4775,"temperature":1.0,"reasoning_tokens":380,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:16:50.811789+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the dual linear program in Eqs. (A5)-(A6) and solve it to certified precision for $\\epsilon=0,0.05,0.1$; if any optimum exceeds $(8+10\\epsilon)/9$, Lemma 1 is false and the exponential bound in the TONS proof does not follow. Separately, instantiate the attack construction in the pseudotelepathy section on the Magic Square game and check non-negativity, normalisation, and every no-signalling condition; a violation would refute the claimed universality of that attack.","supporting_citations":[{"cited_title":"Bourdoncle, S","cited_arxiv_id":null,"evidence_quote":"Defines the ABNS and TONS scenarios and supplies the small-n numerics for noisy PR boxes that the linear-accumulation result extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the chained Bell inequality for which the paper computes explicit guessing-probability rates."},{"cited_title":"Parallel DIQKD from parallel repetition","cited_arxiv_id":"1703.08508","evidence_quote":"Introduces the guessing-game formalism used to define the tripartite game G_g at the core of the proof."},{"cited_title":"Buhrman, S","cited_arxiv_id":null,"evidence_quote":"Provides the t-out-of-n parallel-repetition concentration theorem for no-signalling game value, giving the exponential decay."},{"cited_title":"Holenstein","cited_arxiv_id":null,"evidence_quote":"Supplies the perturbation bound showing that epsilon-almost-no-signalling changes the game value by at most a linear amount."},{"cited_title":"Panconesi and A","cited_arxiv_id":null,"evidence_quote":"Gives the generalized tail bound used to restrict to the rounds in which Eve's input matches Alice's."},{"cited_title":"Renner and S","cited_arxiv_id":null,"evidence_quote":"Establishes the correspondence between pseudotelepathy games and weak Kochen-Specker sets that underpins Theorem 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows the Magic Square game has bound randomness against NS adversaries, the phenomenon Theorem 2 generalises."},{"cited_title":"Ramanathan and P","cited_arxiv_id":null,"evidence_quote":"Together with [16], shows strong monogamy failures in bipartite no-signalling theories that motivate the PT-game attack."},{"cited_title":"Supic, R","cited_arxiv_id":null,"evidence_quote":"Supplies the tight quantum strategy and self-testing results for the chained Bell inequality used to set the admissible violation range."}],"review_version":2}