{"id":"593b90de-c338-448b-9521-4c4fc6b03abe","arxiv_id":"1908.11123","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Under fair sampling, post-selected Bell-test statistics from lossy detectors exactly match the statistics of lossless detectors measuring a locally filtered quantum state, and small fair-sampling violations cause only small statistical deviations.","lead":"This paper formalizes what survives of device-independent quantum claims when detector losses are handled by post-selection under the fair sampling assumption. It shows that post-selected Bell-test data is exactly equivalent to ideal lossless measurements on a filtered quantum state, and it provides bounds for approximate fair sampling.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: Proposition 2 is sound, and the main limitation (unverifiable fair sampling) is explicitly acknowledged by the paper.","rationale":"The reader's verdict of ACCEPT with high confidence is consistent with my own reading. The paper's central claim is a conditional statement: under the fair sampling factorization, post-selected statistics are exactly reproduced by lossless measurements on a locally filtered state. I checked the main proof steps: the decomposition of a lossy device into filter and lossless measurement, the equivalence of the three formulations of fair sampling, the construction of the filtered state and ideal devices, and the approximate-fair-sampling TV bound. All appear mathematically sound. The unverifiability of the fair sampling assumption is a genuine epistemic limitation, but the paper does not overclaim: it repeatedly emphasizes that conclusions apply to the filtered state and that fair sampling cannot be certified from observed statistics alone. The Makarov attack in Section 5.1 further illustrates the need for the assumption to hold from the adversary's perspective, which is a proper caveat rather than a flaw. No red flags, unsupported leaps, or internal inconsistencies were found, so the verdict should remain unchanged.","tokens_in":32446,"tokens_out":35588,"duration_ms":356231,"concrete_test":"Independently re-derive the proof of Proposition 2 in the diagonal-filter gauge (Appendix B) and verify Eq. (17) numerically for a randomly generated device obeying Eq. (11); additionally, sample random operators satisfying the ε condition (39) and check numerically that the total-variation bound in Eq. (40) is respected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After checking the central argument, I find no internal inconsistency that would undermine the main claim. Proposition 2 follows from the filter factorization in Eq. (11): the proof in Appendix C correctly constructs the filtered state Ψ✓ and lossless devices M✓_k, and the probability identity (17) checks out. Proposition 1's equivalence between filter factorization, efficiency factorization (12), and POVM factorization (13) is also valid; the construction of a factorizing filter from a factorized M✓ is possible because only the success branch matters for the post-selected statistics, with failure branches contributing only to the no-click flag. The approximate result in Proposition 4 is supported by a direct calculation: the total-variation bound ε/(1−ε) follows from the stated operator-norm condition and the construction of the lossless device, with no hidden step that I can identify. The reader's weakest assumption—that fair sampling cannot be verified device-independently—is indeed the core limitation of the paper, but it is explicitly stated in Section 3 and the conclusions are carefully phrased as applying to the filtered state, not necessarily the original state. I therefore do not see a load-bearing concern that would warrant changing the verdict.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper analyses what can be concluded in device-independent protocols when Bell tests are performed with lossy detectors and the data are post-selected under a fair sampling assumption. The authors represent a lossy measurement as a filter followed by a lossless measurement, and define weak fair sampling as the factorization of the filter into classical and quantum parts (Eq. (11)). Their central result (Proposition 2, Eq. (17)) states that, under this assumption, the post-selected statistics of any Bell experiment are exactly reproduced by an ideal experiment in which lossless devices measure the locally filtered state Psi_check, which is obtained from the actual state by local probabilistic operations. They then define strong fair sampling (Section 4), for which Psi_check = Psi; discuss cryptographic consequences and an explicit attack (Section 5); prove an approximate version (Proposition 4) with a total-variation bound epsilon/(1-epsilon); prove robustness to state-preparation imperfections (Proposition 5); and apply the formalism to polarization analysers (Sections 6 and 8) and to state-dependent fair sampling (Section 9).","tokens_in":32696,"tokens_out":8844,"duration_ms":85757,"significance":"If the results hold — and the proofs in the appendices are sound — the paper provides a clean and useful clarification of an assumption that is ubiquitous in experimental Bell tests. The contribution is not merely terminological: Proposition 2 gives an explicit, falsifiable identity, Proposition 4 gives a quantitative robustness guarantee with a constructive proof, and the optical examples show how to certify the needed POVM element M_check in practice. The authors also honestly state the main limitation, namely that fair sampling cannot be verified from the observed statistics alone (Section 3), and they carefully phrase the conclusions as applying to the filtered state rather than to the original state. The equivalence to the earlier definition of Berry et al. is disclosed, and the new claims do not rely on circular reasoning or fitted parameters.","major_comments":[],"minor_comments":[{"comment":"The \"lossless\" POVM elements sum to the projector Pi_check rather than to the identity; since the relevant state rho_check has support in Pi_check this is harmless, but the completion to a full POVM on the orthogonal complement should be stated explicitly.","section":"Section 2.2 and Appendix D, Eq. (78)"},{"comment":"The sentence introducing the symbols defines F_{C,check} twice; the second occurrence should refer to F_{Q,check}.","section":"Appendix C, after Eq. (67)"},{"comment":"The no-click operators are typeset with superscripts in a way that obscures the intended expression R^{\\hat N_theta}(1+delta)^{\\hat N_theta}R^{\\hat N_{theta_perp}}; please fix the typesetting.","section":"Section 8.1, Eqs. (42) and (44)"},{"comment":"The inequalities in this equation contain garbled subscripts and superscripts (for example, the terms involving R^{n-1} and R_2) that should be cleaned up; the final result is correct.","section":"Appendix G, Eq. (109)"},{"comment":"The computed approximation error epsilon = (1-eta)delta/eta is used in Proposition 4, which requires epsilon < 1; please state the corresponding restriction delta < eta/(1-eta) explicitly.","section":"Sections 8.2 and 8.3"},{"comment":"The notation F_{Q,check} is used both for a Kraus operator (e.g., Eq. (16)) and for the corresponding CP map (e.g., Eq. (20)); defining this distinction once would avoid confusion.","section":"Sections 2.2, 3, and 4"}],"recommendation":"accept","confidential_remarks":"The manuscript is the published Quantum version; this report concerns the arXiv v2. I found no reason to doubt the central claims. The relation to Berry et al. is properly credited, and the main limitation is stated in the paper itself. The paper is a good match for a quantum-information journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is worth your time: it gives a clean, general answer to what survives of device-independent certification when you post-select under fair sampling. The short version is that post-selected statistics are exactly those of an ideal lossless experiment measuring a locally filtered state Ψ✓, so DI claims (Bell correlation, self-testing, randomness) apply to Ψ✓, not to the original Ψ. That's the right way to frame the detection loophole: fair sampling doesn't let you ignore loss; it tells you what you are actually certifying.\n\nThe main theorem (Prop 2) is solid. The filter formalism is intuitive, the proof in Appendix C checks out, and the equivalence with Berry et al.'s efficiency factorization (Prop 1) is explicitly acknowledged. The genuinely new pieces are the approximate fair sampling bound (Prop 4), the state-preparation-error bound (Prop 5), and the concrete polarization-analyser analysis. Prop 4's ε/(1−ε) total-variation bound is practical and constructive—they even give the explicit lossless device. The paper is honest about the central limitation: fair sampling is not verifiable from the observed statistics alone, and the local hidden-variable models that fake Bell violations violate the factorization. That's not a flaw in the paper; it's the nature of the assumption, and they say so clearly.\n\nSoft spots, in proportion. The bound in Prop 4 is not tight, and they admit it; for low detector efficiencies ε/(1−ε) can get large, so the 'robustness' guarantee is meaningful only for small deviations. The cryptographic section (Sec 5) is more of a threat-model discussion than a formal security proof—it argues composability, but I'd want a fuller proof if this were a crypto paper. The non-i.i.d. extension is sketched rather than proven in detail; the definition is plausible but the treatment is lighter than the i.i.d. case. None of this undercuts the central argument.\n\nWho it's for: anyone running or interpreting Bell tests with post-selection—experimentalists in quantum optics, DI-QKD/QRNG people, and theorists who want to know what fair sampling actually buys. It deserves a serious referee. I'd accept it at a good journal, and I'd cite it in my own work when discussing detection loopholes.\n\nRecommendation: send to peer review. My verdict: accept, with minor revisions at most; the authors could tighten the non-i.i.d. section and add a note about the regimes where the ε/(1−ε) bound is loose.","headline":"A rigorous and useful framework for what post-selected Bell data actually certifies: the locally filtered state, not the original—worth a serious referee.","tokens_in":33183,"tokens_out":3042,"would_cite":true,"duration_ms":27912,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","81P15"],"pacs":["03.65.Ud","03.67.Dd"],"model":"deepseek-v4-flash","headline":"Under the fair sampling assumption, post-selected Bell statistics are exactly those of an ideal experiment measuring a locally filtered state \\(\\Psi_{\\checkmark}\\), so device-independent certifications target \\(\\Psi_{\\checkmark}\\) rather…","keywords":["fair sampling","post-selection","detection loophole","device-independent certification","Bell inequalities","filtered quantum state","total variation distance","quantum key distribution"],"falsifier":"Measure, for a single lossy detector, the detection efficiency \\(E(x,\\rho)\\) for two settings \\(x\\) and two probe states \\(\\rho\\); if \\(E(x,\\rho_1)/E(y,\\rho_1)\\neq E(x,\\rho_2)/E(y,\\rho_2)\\), the factorization \\(E(x,\\rho)=E_C(x)E_Q(\\rho)\\) fails and the paper's Proposition 2 cannot apply. Alternatively, exhibit a local hidden-variable model with state-dependent detection efficiency whose post-selected data reach the CHSH algebraic bound, which would show that post-selected statistics alone cannot certify the claims the paper makes under fair sampling.","tokens_in":32306,"feed_emoji":"🔬","tokens_out":5677,"duration_ms":52121,"temperature":0.7,"pith_summary":"The paper asks what remains of device-independent certification when Bell experiments rely on post-selected data under the fair sampling assumption. It shows that if a lossy detector's decision to click factorizes into an independent classical part and an independent quantum part, then the post-selected statistics are exactly reproduced by an ideal lossless experiment measuring a locally filtered state \\(\\Psi_{\\checkmark}\\). Trusted conclusions about Bell violation, self-testing, randomness, and key distribution therefore apply to \\(\\Psi_{\\checkmark}\\), and not automatically to the original state \\(\\Psi\\). Under strong fair sampling, where the quantum filter is proportional to the identity, the conclusions do apply to \\(\\Psi\\). The paper also proves that small deviations from fair sampling shift the post-selected distribution by at most \\(\\epsilon/(1-\\epsilon)\\) in total variation distance.","feed_headline":"Post-selected Bell data certify a filtered state, not your state","feed_subtitle":"Fair sampling makes lossy-detector results match ideal measurements on a locally filtered state; only strong fairness keeps the original.","key_machinery":"The load-bearing object is the filter: a probabilistic preprocessing stage placed before an ideal detector, which either accepts (\\(\\checkmark\\)) or rejects (\\(\\varnothing\\)) each round. Fair sampling is defined by demanding that the filter factorizes as \\(F=\\wedge(F_C\\otimes F_Q)\\), meaning the acceptance probability splits into a classical part depending only on the setting \\(x\\) and a quantum part depending only on the input state \\(\\rho\\); equivalently, the detection efficiency factorizes as \\(E(x,\\rho)=E_C(x)E_Q(\\rho)\\). This factorization is what allows the rejected rounds to be discarded and the surviving statistics to be re-interpreted as ideal measurements on the locally filtered state \\(\\Psi_{\\checkmark}\\). The argument works because the quantum branch of the filter commutes through the Bell experiment, so post-selection becomes a heralded state preparation.","core_discovery":"The central discovery is that fair sampling, formalized as a factorization of the filter or of the detection efficiency \\(E(x,\\rho)=E_C(x)E_Q(\\rho)\\), turns post-selection into a benign local operation. For any Bell experiment with lossy detectors satisfying this condition, the post-selected distribution equals the distribution obtained from lossless detectors acting on the normalized filtered state \\(\\Psi_{\\checkmark}\\), which is produced from \\(\\Psi\\) by local probabilistic maps (Proposition 2). Consequently a Bell violation in the post-selected data certifies that \\(\\Psi_{\\checkmark}\\) is Bell-correlated, implying that \\(\\Psi\\) itself possesses hidden nonlocality, and any self-testing or cryptographic statement drawn from the data refers to \\(\\Psi_{\\checkmark}\\). Strong fair sampling, in which the quantum filter is proportional to the identity, makes \\(\\Psi_{\\checkmark}=\\Psi\\) and restores the usual device-independent meaning. Under approximate fair sampling the post-selected and ideal distributions are within \\(\\epsilon/(1-\\epsilon)\\) in total variation distance (Proposition 4).","pith_inferences":["One consequence the authors leave implicit is that post-selected Bell tests are effectively entanglement-distillation procedures: the certified object is the state after local filtering, so the amount of certified randomness or key should be quoted together with the filter success probability.","The paper's recommendation to publish the operator \\(M_{\\checkmark}\\) (or bounds on it) rather than the full POVM suggests a practical calibration standard for post-selected DI experiments, a protocol step that goes beyond the theorems proved here.","A quantitative prediction of Section 8 is that deliberately detuning the two detectors of a polarization analyser by a relative amount \\(\\delta\\) should move the post-selected distribution by no more than \\(\\delta(1-\\eta)/\\eta\\) in total variation distance; an experimental check of this bound would be a direct test of the approximate-fair-sampling framework."],"forward_implications":["A Bell violation obtained from post-selected data under fair sampling proves that the filtered state \\(\\Psi_{\\checkmark}\\) is Bell-correlated, not that the original prepared state is.","Self-testing, certified randomness, and DI-QKD protocols run on post-selected data certify properties of \\(\\Psi_{\\checkmark}\\); a separate argument is needed to relate them to \\(\\Psi\\).","If strong fair sampling holds, the post-selected statistics coincide with ideal lossless statistics on the original state \\(\\Psi\\), so all usual device-independent conclusions are restored unchanged.","Small deviations from exact fair sampling are controlled: the total variation distance between post-selected and ideal filtered-state distributions is at most \\(\\epsilon/(1-\\epsilon)\\), and Bell-operator expectation values shift by at most \\(2\\epsilon_{\\rm tot}\\) in units of the algebraic bound.","State-dependent fair sampling extends the result to devices that fail fair sampling globally but satisfy the factorization on the actual experimental state, reproducing the same filtered-state conclusion."],"supporting_citations":[{"why":"Supplies the efficiency-factorization definition of fair sampling that the paper adopts and proves equivalent to its filter definition.","marker":"[1]"},{"why":"Gives the classic data-rejection local model showing how post-selection can fake a Bell violation, motivating the need for the assumption.","marker":"[5]"},{"why":"Establishes the relationship between Bell violation and detection efficiency, a benchmark for the class of models that violate fair sampling.","marker":"[7]"},{"why":"Demonstrates experimentally that Bell violations can be faked without fair sampling, justifying the assumption in cryptographic settings.","marker":"[9]"},{"why":"Provides the device-independent QKD security proof that the paper extends to post-selected data on a filtered state.","marker":"[21]"},{"why":"Provides the randomness-expansion protocol whose guarantees carry over to post-selected statistics under fair sampling.","marker":"[26]"},{"why":"Shows that local filters can reveal hidden nonlocality, which the paper uses to interpret a violation of \\(\\Psi_{\\checkmark}\\) as hidden nonlocality in \\(\\Psi\\).","marker":"[39]"},{"why":"Models a detector attack where fair sampling must hold in the adversary's POVM description for security, a cautionary example used in Section 5.","marker":"[46]"}],"fun_headline_variants":["Fair sampling turns Bell tests into certification of a filtered state","Post-selected Bell data certify a filter, not your original state","Device-independent? Only for the filtered state under fair sampling","Fair sampling: your Bell violation may certify a different state","Bell tests under fair sampling: what you certify is filtered"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole result rests on the hypothesis that a detector's acceptance probability factorizes into a part depending only on the setting and a part depending only on the quantum state, \\(E(x,\\rho)=E_C(x)E_Q(\\rho)\\), a property that cannot be verified from the observed statistics alone.","fun_headline_variants_meta":{"raw":{"variants":["Fair sampling turns Bell tests into certification of a filtered state","Post-selected Bell data certify a filter, not your original state","Device-independent? Only for the filtered state under fair sampling","Fair sampling: your Bell violation may certify a different state","Bell tests under fair sampling: what you certify is filtered"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000385,"raw_usage":{"total_tokens":2077,"prompt_tokens":1026,"completion_tokens":1051,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":642,"completion_tokens_details":{"reasoning_tokens":968}},"tokens_in":642,"tokens_out":1051,"duration_ms":9232,"temperature":1.0,"reasoning_tokens":968,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:23:58.175212+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure, for a single lossy detector, the detection efficiency \\(E(x,\\rho)\\) for two settings \\(x\\) and two probe states \\(\\rho\\); if \\(E(x,\\rho_1)/E(y,\\rho_1)\\neq E(x,\\rho_2)/E(y,\\rho_2)\\), the factorization \\(E(x,\\rho)=E_C(x)E_Q(\\rho)\\) fails and the paper's Proposition 2 cannot apply. Alternatively, exhibit a local hidden-variable model with state-dependent detection efficiency whose post-selected data reach the CHSH algebraic bound, which would show that post-selected statistics alone cannot certify the claims the paper makes under fair sampling.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the efficiency-factorization definition of fair sampling that the paper adopts and proves equivalent to its filter definition."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the classic data-rejection local model showing how post-selection can fake a Bell violation, motivating the need for the assumption."},{"cited_title":"Massar, and S","cited_arxiv_id":null,"evidence_quote":"Establishes the relationship between Bell violation and detection efficiency, a benchmark for the class of models that violate fair sampling."},{"cited_title":"Gerhardt, Q","cited_arxiv_id":null,"evidence_quote":"Demonstrates experimentally that Bell violations can be faked without fair sampling, justifying the assumption in cryptographic settings."},{"cited_title":"Pironio, A","cited_arxiv_id":null,"evidence_quote":"Provides the device-independent QKD security proof that the paper extends to post-selected data on a filtered state."},{"cited_title":"Colbeck, and A","cited_arxiv_id":null,"evidence_quote":"Provides the randomness-expansion protocol whose guarantees carry over to post-selected statistics under fair sampling."},{"cited_title":"Gisin, Hidden quantum nonlocality revealed by local filters , Physics Letters A 210(3), 151- 156 (1996)","cited_arxiv_id":null,"evidence_quote":"Shows that local filters can reveal hidden nonlocality, which the paper uses to interpret a violation of \\(\\Psi_{\\checkmark}\\) as hidden nonlocality in \\(\\Psi\\)."},{"cited_title":"Lydersen, C","cited_arxiv_id":null,"evidence_quote":"Models a detector attack where fair sampling must hold in the adversary's POVM description for security, a cautionary example used in Section 5."}],"review_version":1}