{"id":"55fef23a-309c-445d-ba88-49c6c035118c","arxiv_id":"2607.04035","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A pure robust self-test of the singlet and Paulis is obtained for non-unitary binary observables via regularization, yielding an explicit analytic O(√ε) distance bound without measurement dilation.","lead":"This paper derives a robust self-testing bound for the two-qubit singlet and Pauli measurements that works directly with non-projective (non-unitary) observables, without Naimark dilation. It shows that certifying real laboratory devices is quantitatively harder than standard projective analyses suggest.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Arithmetic errors in SM Steps 3–4 produce an unsupported (too-small) explicit constant C in Theorem 1.","rationale":"The pure-state hypothesis highlighted by the reader is standard and clearly flagged; it is not the weakest link for the strongest claim as written. The claim includes an explicit numerical C that the SM fails to justify because of elementary arithmetic slips in the norm bounds. The qualitative result (existence of some C√ε pure self-test that never dilates the measurements) survives, so the paper remains a solid contribution once the constant is corrected; hence CONDITIONAL rather than REJECT. The reader’s low correctness-risk assessment overlooked these calculation errors, producing only partial agreement.","tokens_in":18546,"tokens_out":586,"duration_ms":74519,"concrete_test":"Recompute the chain (A25)–(A85) from the SOS residual, replacing every occurrence of the erroneous 2^{9/4} by the triangle-inequality value 2^{7/4}, using ∥A−A_reg∥_∞⩽1, and collecting the final coefficient of √ε for Δ+f_unitaries. If that coefficient exceeds (179+53√2)/2^{3/4}, the constant stated in Theorem 1 is false.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 1 asserts a concrete robustness constant C=(179+53√2)/2^{3/4}≈151. The derivation of this number in the Supplemental Material contains multiple arithmetic inconsistencies: (A61) writes ∥(B̃_y−B̃_reg)|ψ⟩∥⩽2^{9/4}√ε while the triangle inequality on (A35)+(A36)+(A60) yields only 2^{7/4}√ε; the subsequent insertion into Δ (A62) then claims 2^{7/4}√ε, which matches neither the written (A61) nor the correct value √2·2^{3/4}+2^{7/4}=2^{5/4}(1+√2). Parallel coefficient errors appear in the anti-commutator expansions (A77)–(A84) (e.g., 7+√2 versus the correct 7+2√2). Re-evaluating the same chain of estimates with corrected exponents already produces a prefactor ≈248>151; even after tightening the loose operator-norm bound ∥A−A_reg∥_∞⩽2 to the sharp value 1 the prefactor remains ≈235>151. Consequently the specific numerical claim of Theorem 1 is not established by the given proof (while an O(√ε) statement with a larger explicit constant is recoverable by the same method).","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.5","summary":"The manuscript establishes a robust pure self-test of the two-qubit singlet and Pauli observables in the CHSH scenario without Naimark-dilating the local measurements. Binary Hermitian observables satisfying A_x^{2} ≤ I and B_y^{2} ≤ I act on a pure bipartite state; when ⟨B⟩ = 2√2 - ε the authors regularize them by a modified sign function, invoke McKague et al.’s unitary robust-self-testing theorem, and bound both the regularization error Δ and the anti-commutator residuals that enter that theorem. The resulting Euclidean distance is claimed to be at most C√ε with the explicit constant C = (179 + 53√2)/2^{3/4}. The proof is elementary (SOS residual of CHSH, spectral estimates, triangle/parallelogram/Cauchy–Schwarz inequalities) and is fully written out in the Supplemental Material.","tokens_in":18925,"tokens_out":852,"duration_ms":6374,"significance":"The work fills a genuine conceptual gap: almost all existing analytic robust self-tests assume projective (unitary) observables, thereby concealing the operational cost of realistic POVMs. By refusing measurement dilation while retaining state purification, the paper supplies the first fully device-independent analytic O(√ε) bound that quantifies that cost. The explicit constant, the clean separation into a unitary term and a POVM-deviation term Δ, and the concrete numerical illustration for photonic threshold detectors (ε ≈ 0.138) make the result immediately usable for assessing near-term experiments and for motivating non-projective numerical hierarchies. The pure-state restriction is an explicit modelling choice already standard in the literature; the technical contribution is therefore solid and timely.","major_comments":[{"comment":"Theorem 1 and SM Steps 3–4: the explicit numerical prefactor C = (179 + 53√2)/2^{3/4} is not supported by the written chain of estimates. Equation (A61) claims ∥(B̃_y - B̃_reg)|ψ⟩∥ ≤ 2^{9/4}√ε, yet the triangle inequality applied to (A35)+(A36)+(A60) yields only 2^{7/4}√ε; the subsequent insertion into Δ (A62) then asserts 2^{7/4}√ε, which matches neither the written (A61) nor the correct arithmetic √2·2^{3/4} + 2^{7/4} = 2^{5/4}(1+√2). Parallel coefficient mismatches appear in the anti-commutator bounds (A77)–(A84) (e.g., 7+√2 versus the correct 7+2√2). Re-evaluating the same inequalities already produces a prefactor ≈ 248; even after tightening the loose operator-norm bound ∥A - A_reg∥_∞ ≤ 2 to the sharp value 1 the constant remains ≈ 235. The O(√ε) scaling itself is recoverable, but the concrete numerical claim of Theorem 1 must be corrected (or the intermediate inequalities tightened","section":null}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":"The arithmetic slip is load-bearing for the only quantitative claim of the paper, yet it is purely mechanical and can be fixed by a careful re-computation of the constants; once that is done the manuscript is essentially ready. I see no deeper conceptual flaw."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The real contribution is a clean regularization (modified sign function) that lets them feed non-unitary binary observables into McKague et al. without Naimark dilation, while still recovering an analytic O(√ε) robustness bound for the singlet and Paulis. That is new relative to the usual projective proofs and to Baptista et al.’s lifting theorems, which pay an extra square-root and drop to O(ε^{1/4}). The SOS residual is used correctly to control both the regularization error Δ and the anti-commutators, the pure-state modeling choice is stated up front, and the whole argument is elementary and self-contained.\n\nThe soft spot is the concrete constant. Theorem 1 claims C = (179 + 53√2)/2^{3/4} ≈ 151. The Supplemental Material chain that produces it has inconsistent exponents: the triangle inequality on the tilted residual gives 2^{7/4}√ε, yet (A61) writes 2^{9/4}; the subsequent insertion into Δ and the anti-commutator expansions (A77–A84) then under-count coefficients (7+√2 instead of something closer to 7+2√2). Re-running the same estimates already yields a prefactor around 235–248. So the O(√ε) statement is fine and the method works; the specific numerical claim does not hold as written. That is a fixable arithmetic gap, not a conceptual hole.\n\nCitation pattern is appropriate (McKague, Bamps-Pironio, Kaniewski, Baptista, Vivoli). No free parameters or circular normalizations. The pure-state restriction is essential for the vector-norm steps and is standard DI practice.\n\nThis is for people who actually write or use robust self-testing bounds and care about what happens when the lab POVMs are not projective. It deserves a serious referee; the constant just needs to be recomputed and the theorem statement adjusted. I would engage with the corrected version.","headline":"Solid idea and complete elementary proof that non-projective CHSH self-testing keeps O(√ε) scaling, but the explicit constant C in Theorem 1 is arithmetically unsupported and too small.","tokens_in":19494,"tokens_out":555,"would_cite":true,"duration_ms":5276,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A robust analytic self-test for the two-qubit singlet and Pauli observables works without forcing measurements to be projective.","keywords":["device-independent self-testing","CHSH inequality","non-projective measurements","robust self-testing","singlet state","Pauli observables","sum-of-squares","Naimark dilation"],"falsifier":"Compute the analytic bound for a laboratory CHSH value near 2.69 (ε ≈ 0.138) and check whether the resulting distance exceeds 2 (the maximum possible Euclidean distance); if the distance remains informative or if a tighter numerical method without Naimark recovers a smaller distance, the claimed severity of non-projectivity is refuted.","tokens_in":19454,"feed_emoji":"⚛️","tokens_out":971,"duration_ms":8924,"temperature":0.7,"pith_summary":"Device-independent self-testing certifies that an experiment is close to an ideal quantum state and set of measurements using only the observed input-output statistics. Most existing proofs treat the local measurements as projective by embedding them in a larger space; that step hides the fact that real laboratory devices produce non-unitary observables. This paper shows that the embedding can be avoided. By keeping the shared state pure (the usual model of an untrusted source) while leaving the measurements undilated, the authors regularize the physical operators with a modified sign function, split the total error into a standard unitary term and an explicit non-unitarity term, and obtain a fully analytic distance bound that scales as the square root of the CHSH deficit. The bound makes concrete that certifying genuine non-projective implementations is quantitatively more demanding than projective models suggest, especially for photonic experiments whose detection efficiency already limits the CHSH value far below the Tsirelson bound.","feed_headline":"Self-test of the singlet works without projective measurements","feed_subtitle":"Analytic O(√ε) bound shows real POVM devices cost far more to certify than projective models claim","key_machinery":"Regularization of the physical non-projective observables by a modified sign function that maps non-negative eigenvalues to +1 and negative eigenvalues to −1, producing exactly unitary operators to which McKague’s robust-singlet theorem can be applied; the triangle inequality then isolates an extra deviation term Δ that is controlled by the sum-of-squares residual of the CHSH operator.","core_discovery":"If a pure bipartite state and binary Hermitian observables with A^{2} ⩽ I and B^{2} ⩽ I achieve CHSH expectation 2√2 − ε, then there exist local isometries such that the Euclidean distance between the physical (regularized-tilted) operators applied to the state and the ideal Pauli operators on the singlet is at most C√ε, with the explicit constant C = (179 + 53√2)/2^{3/4}.","pith_inferences":["The large prefactor C ≈ 150 means that even modest experimental noise already saturates the trivial distance bound of 2, so practical certification will almost certainly need tighter constants or adaptive extraction methods.","Once non-projective numerical self-testing tools exist, the analytic O(√ε) scaling derived here can serve as a sanity-check benchmark for those hierarchies.","The pure-state restriction, while standard for untrusted sources, leaves open whether a fully mixed-state version of the same bound can be recovered by a different regularization."],"forward_implications":["Standard analytic and SDP self-testing bounds that assume projectivity systematically understate the distance of real POVM implementations from the ideal singlet and Paulis.","Photonic Bell experiments limited by threshold detectors to CHSH ≈ 2.69 require either much higher visibility or entirely new non-projective numerical hierarchies before meaningful device-independent certification is possible.","Security proofs for device-independent QKD and randomness that import projective self-testing statements must be re-examined once measurement non-unitarity is quantified.","The same regularization-plus-SOS strategy can be attempted for multipartite or higher-dimensional Bell inequalities where non-projective measurements are even more common."],"fun_headline_variants":["Singlet self-test holds without projective measurement assumption","Robust pure self-test of two-qubit singlet for non-unitary observables","O(√ε) bound certifies singlet without Naimark dilation of POVMs","Self-testing Pauli observables on singlet skips unitary assumptions","Real POVM devices need stronger bounds than projective models claim"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The shared state is required to be pure so that residual non-unitarity can be turned into concrete vector-norm bounds on the operators; without purity those estimates fail.","fun_headline_variants_meta":{"raw":{"variants":["Singlet self-test holds without projective measurement assumption","Robust pure self-test of two-qubit singlet for non-unitary observables","O(√ε) bound certifies singlet without Naimark dilation of POVMs","Self-testing Pauli observables on singlet skips unitary assumptions","Real POVM devices need stronger bounds than projective models claim"]},"model":"grok-4.5","effort":"low","cost_usd":0.005422,"raw_usage":{"total_tokens":1409,"prompt_tokens":662,"num_sources_used":0,"completion_tokens":91,"cost_in_usd_ticks":54220000,"prompt_tokens_details":{"text_tokens":662,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":656,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":662,"tokens_out":91,"duration_ms":5095,"temperature":1.0,"reasoning_tokens":656,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-11T22:08:47.063003+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Compute the analytic bound for a laboratory CHSH value near 2.69 (ε ≈ 0.138) and check whether the resulting distance exceeds 2 (the maximum possible Euclidean distance); if the distance remains informative or if a tighter numerical method without Naimark recovers a smaller distance, the claimed severity of non-projectivity is refuted.","supporting_citations":[],"review_version":1}