{"id":"7748ee2f-59d7-492c-a0b4-2f52a08990d5","arxiv_id":"2512.04913","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"STT-UEP transmits classical shadow data with unequal error protection, enabling a decoder to estimate M local Pauli observables using O(9^w log(M)/epsilon^2) state copies, with basis bits coded more strongly than outcome bits.","lead":"The paper presents a protocol for sending classical measurement data about a quantum state over a noisy channel so a receiver can estimate many properties of the state efficiently. It protects the bits that describe measurement bases more strongly than the bits that describe outcomes, because a wrong basis is catastrophic while a wrong outcome is just noise.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The unbiasedness and Hoeffding bound in Prop. 1 both rely on exact knowledge of p_err and on independence of post-decoding bit errors; the paper only asserts 'approximately' with interleaving, so the central guarantee is not established for the LDPC-coded channel used in the simulations.","rationale":"The reader's weakest_assumption matches my own: the proof of Proposition 1 depends on exact, independent post-decoding error statistics. I reviewed the appendix carefully. Equations (22)–(24) derive the unbiasedness of \\hat o_m using E[∏_{j∈S_m} \\hat b_{i,j}] = (1−2p)^{|S_m|}⟨O_m⟩, which holds only when the residual error bits are independent. Equation (25) applies Hoeffding's inequality to a sum of N bounded terms, also requiring independence across copies. The paper's appeal to interleaving [11] does not establish independence of the LDPC decoder's output errors; interleaving mitigates bursty channel errors before decoding, but residual errors after decoding can still be correlated. Moreover, the protocol gives no procedure for the decoder to know p_err; using a mismatched value introduces a bias that grows with 3^w/(1−2p)^w. This is an internal gap between the stated assumptions and the guarantee claimed for the simulated system, not a disagreement with external consensus. The reader's CONDITIONAL verdict is appropriate: the core idea is plausible and numerically supported, but the theorem as stated is not yet a rigorous guarantee for the practical coding scheme. The concern is addressable by adding a pilot-based p_err estimator and either proving or empirically validating approximate independence, or replacing (19) with a correlation-aware correction. Therefore I recommend no change to the reader's verdict.","tokens_in":7473,"tokens_out":11001,"duration_ms":98426,"concrete_test":"Transmit random (e.g., all-zero) codewords over the same AWGN channel and LDPC code used in Fig. 4 at a fixed rate and SNR. After decoding, group the output bits into blocks of size w (matching the observable weight) and compute the empirical average of (−1)^{Σ_{j=1}^w e_j} over many such groups. Compare this empirically measured value with (1−2 p_err)^w, where p_err is the measured BER. If the two differ by more than the Monte-Carlo standard error, Eq. (23) fails and the debiasing factor (19) is miscalibrated for this code, invalidating the unbiasedness claim in Eq. (20).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result Proposition 1 rests on the debiasing factor a_m = 3^{|S_m|}/(1−2 p_err)^{|S_m|} in Eq. (19), derived in Eqs. (22)–(24) under the assumption that each decoded outcome bit is flipped independently with known probability p_err = BER(Nn, R_b). This assumption appears twice in a load-bearing way. First, unbiasedness (Eq. 20) requires E[∏_{j∈S_m} \\hat b_{i,j} | compatible] = (1−2p)^{|S_m|}⟨O_m⟩, which is only true when the residual bit flips e_j are independent. LDPC decoding does not produce independent residual errors, and interleaving before transmission does not decorrelate errors after decoding. Second, the Hoeffding bound in Eq. (25) requires the N summands to be independent; the paper asserts this 'holds approximately' but provides no bound on the approximation error. Furthermore, the decoder is not told how to obtain p_err; a mismatch between the true and assumed p_err biases every estimate, and the bias is amplified by 3^w/(1−2p)^w. Without a specified estimator for p_err and a correlation-aware correction, the guarantee (21) is not a theorem about the implemented system.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers an encoder with access to N copies of an n-qubit state. The encoder measures each copy in a random Pauli basis from {I,H,HS^dagger}^n, then transmits the basis choices and measurement outcome bits over a classical noisy channel using unequal error protection: basis bits are encoded with a stronger code (rate R_u) than outcome bits (rate R_b > R_u). The decoder selects measurements compatible with each of M Pauli observables of weight at most w, applies a debiasing factor that depends on the bit-error probability p_err, and forms estimates of the observable expectation values. Proposition 1 gives a lower bound on N that is logarithmic in M and exponential in w, and numerical simulations compare the scheme with shadow tomography with equal error protection and with quantized full state transmission.","tokens_in":7862,"tokens_out":9650,"duration_ms":99106,"significance":"The underlying idea is attractive and potentially useful: it connects classical shadow tomography with classical error-control coding, and it isolates the observation that basis errors are catastrophic while outcome errors can be partially corrected by debiasing. The derivation of the debiasing factor is clean and parameter-free, and the bound in Proposition 1 is a genuine sample-complexity statement for the idealized Bernoulli-error model assumed in the proof. The numerical comparison with relevant baselines is useful and shows a consistent advantage for the UEP strategy in the tested regime. However, the paper overstates the communication-complexity result and does not fully bridge the gap between the theorem's assumptions and the LDPC-coded channel used in simulations.","major_comments":[{"comment":"The claimed communication complexity is not independent of the number of qubits n. Since S = 3^n, log_2(S) = n log_2(3), so Eq. (11) gives B = (n/R_b + n log_2(3)/R_u) N. With N from Eq. (21), B is linear in n: B = Theta(n * 9^w * log(M/delta) / ((1-2p_err)^{2w} epsilon^2)). The result is a sample-complexity bound independent of n, not a bit-complexity bound independent of n. The Abstract and the 'Main Contributions' section should be revised to avoid this conflation, since the stated contribution is communication complexity in bits.","section":"Sec. III-A, Eq. (11); Abstract"},{"comment":"Proposition 1 rests on two assumptions that are not satisfied by the simulated system: (i) the post-decoding bit flips are independent, and (ii) the decoder knows p_err exactly. The debiasing factor in Eq. (19) and the Hoeffding bound in Eq. (25) both require independent flips. Interleaving before transmission does not make LDPC residual errors independent after decoding, and the paper does not explain how the decoder obtains p_err. If the assumed and true p_err differ, every estimate is biased, with the bias amplified by 3^w/(1-2p)^{2w}. Thus Eq. (21) is a theorem for an idealized Bernoulli-error model, not for the LDPC-coded AWGN channel used in Section IV. The authors should either state this model explicitly as an assumption and align the simulations with it, or provide a correlation- and mismatch-robust version of the guarantee.","section":"Sec. III-B, Eqs. (13)-(19); Appendix (22)-(25)"}],"minor_comments":[{"comment":"The formal requirement should read Pr(max_m |o_m_hat - <O_m>| > epsilon) < delta, i.e. the union bound over m is needed. As written, 'for all m' inside the probability states a different, weaker event that does not match the union-bound proof in the Appendix.","section":"Sec. II, Eq. (4)"},{"comment":"The text refers to 'Proposition 2', but the paper contains only Proposition 1. The reference should be corrected.","section":"Sec. IV, Fig. 4"},{"comment":"The caption uses 'STT-UER' in one place; this should be 'STT-UEP'.","section":"Sec. IV, Fig. 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper has a promising core idea and the Appendix derivation is sound under the stated Bernoulli-noise model. The main obstacles are the overstated bit-complexity claim and the mismatch between the theorem's independence/known-p_err assumptions and the LDPC-based simulations. Both are fixable in a revision, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read it. The core idea is worth taking seriously: when you send classical shadow data over a noisy bit channel, protect the basis bits more than the outcome bits. The paper makes that case cleanly, and the numerical work backs it up.\n\nWhat is new: the observation that basis-selection errors are catastrophic while outcome bit flips are statistically debiasable, and the UEP protocol that codes bases at a lower rate. The appendix proof is correct under its stated assumptions: the debiasing factor 3^w/(1-2p)^w is the right fix if residual bit flips are independent and the decoder knows p_err. The Hoeffding step is standard, and (21) follows honestly from those assumptions.\n\nTwo soft spots matter.\n\nFirst, the \"independent of n\" claim is false for transmitted bits. There are 3^n possible random bases, so specifying the bases costs n log2(3) bits per copy. Total bits are B = O(n N) = O(n 9^w log M). Sample complexity is n-independent; bit complexity is not. The authors conflate the two in the abstract and Section I-C. This should be corrected.\n\nSecond, the guarantee (21) relies on independent, known p_err after LDPC decoding. LDPC residual errors are neither independent nor known to the decoder, and interleaving before transmission does not decorrelate errors after decoding. The paper only says \"approximately\" and moves on. If the decoder's assumed p_err is off, the debiasing factor amplifies bias by 3^w/(1-2p)^w. This is a genuine gap between the theorem and the system simulated. The authors need either to specify how p_err is estimated and show robustness to mismatch, or to prove a bound that holds under correlated residual errors.\n\nMinor things: a few typos (ST-UEP, STT-UER) and the abstract oversells \"arbitrary observables\" when the results are for local Pauli observables.\n\nProportionate verdict: the protocol is plausible, the math is correct under stated assumptions, and the experiments are honest. These are fixable issues, not a dead end. The paper deserves a serious referee, but the referee should demand a corrected n-scaling claim and an honest treatment of the residual-error model before acceptance.","headline":"Useful UEP-for-classical-shadows idea, but the n-independence claim is wrong and the main guarantee assumes independent known residual errors the implemented system doesn't deliver.","tokens_in":8293,"tokens_out":5028,"would_cite":false,"duration_ms":50920,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P47"],"pacs":["03.67.-a","03.67.Hk"],"model":"deepseek-v4-flash","headline":"This paper claims that a receiver can estimate the expected values of arbitrary local Pauli observables of an unknown quantum state sent over a noisy classical channel using a number of transmitted bits that grows logarithmically with the n","keywords":["shadow tomography","classical shadows","unequal error protection","quantum communication","Pauli observables","sample complexity","debiasing","noisy classical channel"],"falsifier":"Transmit through a channel with correlated (bursty) errors without interleaving, use the claimed debiasing factor with a fixed $p_{err}$, and measure the empirical bias of the estimator over many trials. If the bias exceeds $\\epsilon$, or if the required $N$ from (21) fails to satisfy the accuracy requirement (4), the central claim is refuted.","tokens_in":7428,"feed_emoji":"⚛️","tokens_out":6708,"duration_ms":59637,"temperature":0.7,"texified_at":"2026-08-05T20:41:52.209941+00:00","pith_summary":"The paper introduces STT-UEP, a protocol for sending properties of an unknown quantum state over a classical noisy channel without requiring the receiver to reconstruct the full state. It measures copies of the state in random Pauli bases, then transmits the basis descriptions and the measurement outcomes separately, giving the basis bits a stronger error-correcting code. A debiasing factor derived from the bit-error probability exactly removes the statistical bias that outcome bit flips introduce, so the number of state copies needed grows only logarithmically with the number of requested observables and exponentially only with the largest observable weight, independent of system size. If correct, this removes the exponential-in-qubits communication bottleneck for distributed quantum computing and sensing scenarios where the receiver only needs specific properties.","texify_model":"deepseek-v4-flash","texify_usage":{"total_tokens":8940,"prompt_tokens":735,"completion_tokens":8205,"prompt_tokens_details":{"cached_tokens":0},"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":735,"completion_tokens_details":{"reasoning_tokens":7529}},"feed_headline":"Bases get stronger codes, so quantum properties send in 9^w samples","feed_subtitle":"Protocol STT-UEP estimates many Pauli observables over noisy channels, sample count scaling with observable weight alone.","key_machinery":"The central mechanism is the pairing of basis-compatibility filtering with a multiplicative debiasing factor. The decoder selects only the approximately $3^{-|S_m|}$ fraction of measurements whose random Pauli basis matches the target observable on its support, and then scales the averaged product outcome by $a_m = \\frac{3^{|S_m|}}{(1 - 2 p_{err})^{|S_m|}}$. This factor simultaneously corrects for the chance that a basis is compatible and for the expected attenuation of the product observable under independent bit flips. Unequal error protection with code rates $R_u < R_b$ ensures the basis-identifying bits survive the channel, since an error there is catastrophic, while outcome bit flips enter only as a kn","core_discovery":"STT-UEP combines classical shadow tomography with unequal error protection: the encoder measures $N$ copies of the state in randomly chosen single-qubit Clifford bases (Z, X, or Y), and transmits the basis descriptions under a low-rate code and the outcome bits under a higher-rate code. The decoder keeps only measurements whose basis is compatible with each target observable, forms a product estimate on the support, and multiplies by the debiasing factor $a_m = \\frac{3^{|S_m|}}{(1 - 2 p_{err})^{|S_m|}}$, which exactly cancels the selection bias (probability $3^{-|S_m|}$) and the bit-flip bias ($(1 - 2 p_{err})^{|S_m|}$). Proposition 1 shows that with probability $1 - P_{\\text{outage}}$, the accuracy requirement $\\Pr(|\\hat{o}_m -$","pith_inferences":["The UEP principle likely extends beyond Pauli observables: in any shadow-tomography pipeline where the transmitted classical description of the random basis is more safety-critical than the outcome bits, protecting the basis-identifying bits with a stronger code should yield the same kind of logarithmic communication advantage.","If the decoder does not know p_err exactly, the debiasing factor becomes mismatched and the estimator develops a bias proportional to the error in p_err; the bound suggests this could be handled by estimating p_err from pilot bits and paying a multiplicative penalty of (1 - 2 p_err)^{-w}.","A testable extension: on a channel with bursty errors and no interleaving, the independence assumption for bit flips fails, so the debiasing factor should no longer cancel the bias; one would expect the outage-free accuracy guarantee to degrade exactly in proportion to the residual error correlation.","The rate-splitting idea generalizes to other semantic communication settings: bits that determine 'which question is being asked' (the basis) are worth more than bits that answer the question, so giving them unequal protection should be near-optimal whenever the question identity is discrete and the answer is a real-valued statistic."],"forward_implications":["The total number of transmitted bits is B = (n/R_b + ceil(log2 S)/R_u) N, with N given by (21), so communication cost scales logarithmically with M and exponentially only in the maximum observable weight w, not the system size n.","Outcome bit errors can be both corrected and debiased without retransmission: as long as the bit-flip probability p_err is below 1/2 and known, the estimator \\(\\hat o_m = a_m \\hat o^{biased}_m\\) is unbiased and satisfies the Hoeffding-based guarantee.","In the numerical experiments, giving the basis bits a stronger code than the outcomes (R_u < R_b, including uncoded outcomes with R_b = 1) achieves the highest success probability for a fixed total bit budget, outperforming equal error protection and conventional state-vector quantization.","The protocol is agnostic to the receiver's chosen observables: the encoder need not know which M Pauli observables the decoder will request, only the maximum weight w.","A channel outage, caused by failure to decode the basis bits despite the CRC check, sets the overall failure probability to P_outage = BLER(ceil(log2 S) N, R_u)."],"fun_headline_variants":["STT-UEP: send quantum property info with logarithmic bit cost","Shadow tomography plus UEP: quantum observables sent efficiently","Unequal error protection shrinks quantum property transmission cost","STT-UEP: logarithmic bits for quantum observable estimation","Quantum property send: bases coded stronger than outcomes"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof treats bit errors on the transmitted measurement outcomes as independent across all $N n$ bits with a known, fixed probability $p_{err}$, and the debiasing factor collapses if residual errors are correlated after decoding or if $p_{err}$ is unknown.","fun_headline_variants_meta":{"raw":{"variants":["STT-UEP: send quantum property info with logarithmic bit cost","Shadow tomography plus UEP: quantum observables sent efficiently","Unequal error protection shrinks quantum property transmission cost","STT-UEP: logarithmic bits for quantum observable estimation","Quantum property send: bases coded stronger than outcomes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00036,"raw_usage":{"total_tokens":1771,"prompt_tokens":720,"completion_tokens":1051,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":464,"completion_tokens_details":{"reasoning_tokens":971}},"tokens_in":464,"tokens_out":1051,"duration_ms":7014,"temperature":1.0,"reasoning_tokens":971,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:29:01.946246+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Transmit through a channel with correlated (bursty) errors without interleaving, use the claimed debiasing factor with a fixed $p_{err}$, and measure the empirical bias of the estimator over many trials. If the bias exceeds $\\epsilon$, or if the required $N$ from (21) fails to satisfy the accuracy requirement (4), the central claim is refuted.","supporting_citations":[],"review_version":1}