{"id":"a1975244-2930-4379-bc72-7426e9b10e2f","arxiv_id":"2506.16834","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A syndrome-history algorithm, REI, detects and localizes radiation-induced correlated faults in simulated rotated surface codes, and a RadMatching pre-processing step reduces logical error rate during radiation bursts.","lead":"Quantum computers that use error correction can be knocked out when a cosmic ray hits the chip and corrupts many qubits at once. This paper introduces a software routine that reads the ongoing error-correction signals, detects such a strike, finds where it hit, and uses that information to help the decoder recover.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closed-loop validation: REI's 100% detection and localization claims are tested only on the exact radiation model the detector is built to recognize, with no sensitivity analysis and no quantitative localization error metric.","rationale":"I focused on model circularity because the paper's universal claims are about real radiation, not just about the simulated signature. The reader's weakest_assumption identifies the same load-bearing premise, and my reading agrees with that assessment. I also note two concrete gaps that make the concern testable: no quantitative localization metric appears anywhere in Section 5, and the RadMatching gain is measured on a single central-fault d=9 configuration. These are not accusations of distortion; they simply delimit what the evidence supports. The strongest independent evidence is the scale of the simulation (over 11 million shots) and the internal consistency of the figures, but scale alone does not break the closed loop when the injection model is the only source of ground truth. A sensitivity/held-out-model check would settle whether REI detects radiation per se or only the specific inverse-square-plus-exponential pattern of Eqs. 1-4. Since the paper explicitly promises open-source release [5] and the pipeline is STIM-based, this check is feasible by the authors or a third party. The conditional verdict stands unchanged.","tokens_in":18980,"tokens_out":7650,"duration_ms":85437,"concrete_test":"Rerun the full REI/RadMatching pipeline on the same simulated surface code but with held-out radiation models that break the assumed signature: (a) a spatially uniform correlated burst with S=1 inside a random radius and 0 outside, (b) a slower temporal decay without the exp(10) spike, and (c) an experimentally calibrated model based on McEwen et al. 2022 or Wilen et al. 2021. For each variant, report detection rate, centroid error in lattice units, area overlap (IoU or Dice) against the true affected set, and RadMatching logical-error improvement. If detection remains at 100% and centroid error stays within one lattice spacing across all variants, the closed-loop concern is resolved; if not, the universal claims in the abstract need to be narrowed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central abstract claim—100% fault detection, accurate centre/area identification, and up to 20-25% decoder improvement—rests on a self-consistent simulation loop. REI's temporal FIFO, spatial clustering, and weighted-centroid radius are keyed to exactly the signatures generated by the paper's own radiation model: the exp(10) coherence collapse in Eq. 1, the single-gate erasure probability in Eq. 2, the inverse-square spatial falloff in Eq. 3, and the product form in Eq. 4. If real cosmic-ray events have different temporal persistence, a different spatial profile, or affect stabiliser measurements differently, these detection and localization numbers need not transfer. The manuscript provides no robustness study over these model parameters, no validation on experimental correlated-error data (e.g., McEwen et al. 2022 or Wilen et al. 2021), and no quantitative accuracy metric for the claimed centroid and radius estimates: Sections 5.1 and 5.2 report detection rates and affected-stabiliser ratios but never compare the returned centre and radius against the injected locus. The RadMatching improvement is also evaluated only for one configuration (d=9, central fault, 384 shots). These are addressable limitations, so the appropriate status is conditional rather than accepted or rejected.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Radiation Event Identification (REI), a runtime algorithm that analyzes syndrome measurements of a Rotated Surface code to detect cosmic-ray-induced correlated errors, estimate the impact centre, and identify the affected area. The authors also introduce RadMatching, a radiation-aware pre-processing step that flips stabiliser measurements in the estimated affected region before feeding the syndrome to a MWPM decoder. The claims—100% fault detection, accurate centre/area identification, sub-0.3% decoding overhead, and up to 20–25% improvement in decoder output correctness—are supported by a large STIM-based simulation campaign (over 11 million shots) using a physics-derived radiation model specified by Equations 1–4.","tokens_in":19229,"tokens_out":4213,"duration_ms":40798,"significance":"The topic is timely and practically relevant: radiation-induced correlated errors are a recognized obstacle to fault-tolerant superconducting quantum computing, and algorithmic mitigation at the decoder level would complement hardware shielding approaches. The paper's strengths include the scale of the simulation campaign, the explicit goal of runtime integration, the low overhead of REI, and the concrete proposal of a radiation-aware decoding step. If the claims hold, this would be a useful step toward radiation-aware QEC. However, the significance is contingent on the transferability of the results from the assumed physics model to real device behaviour; the current evidence is entirely simulation-based and the detector is matched to the model that generates its test data.","major_comments":[{"comment":"The REI detection logic (FIFO temporal correlation, spatial clustering test, weighted centroid, radius formula) is constructed to recognize exactly the correlated signatures generated by the paper's own radiation model: exp(10) coherence collapse, single-gate erasure probability, inverse-square spatial falloff, and product form. The manuscript tests the detector on this same model, so the reported 100% detection and localization accuracy may be an artifact of this closed loop. No sensitivity analysis over model parameters (e.g., the temporal decay exponent, the spatial falloff exponent, or event duration) is provided, and there is no validation on experimental correlated-error data such as those in McEwen et al. [36] or Wilen et al. [45]. This is load-bearing for the central claim; please add a systematic robustness study or an experimental benchmark.","section":"Section 3.3 (Eqs. 1–4) and Algorithm 1"},{"comment":"The abstract claims that REI \"accurately identify[ies] both the radiation impact centre and the area affected,\" but the evaluation never compares the returned (x, y) and radius against the injected locus. The reported metrics are detection rate and afflicted-stabiliser ratio over time; there is no quantitative localization error metric. Without e.g. the mean/median absolute error in centre coordinates and radius, as a function of time, code distance, and injection site, the localization claim is unsupported.","section":"Sections 5.1 and 5.2"},{"comment":"RadMatching is evaluated only for one configuration: d=9, central fault, 384 shots. The abstract's \"up to 20%\" improvement conflicts with the text's \"upwards of 25%\" in Section 5.5 and \"about 25%\" in the Conclusion. The improvement should be reported as an average over a meaningful ensemble with error bars, and should be tested across multiple fault positions, code distances, and radiation intensities to support the general claim. It should also be clarified whether the improvement is measured relative to MWPM alone or to all the compared decoders, since the comparison set matters for the abstract's wording.","section":"Section 5.5 and Figure 7"},{"comment":"The claim that the technique \"detects 100% of injected faults\" is ambiguous. The detection-rate plots in Figures 3 and 4 show 100% only near event onset, with the rate falling before the end of the 1 ms event (earlier for peripheral faults and higher-distance codes). If the intended claim is that every injected event is detected at least once during its lifetime, this should be stated explicitly and supported with per-event detection counts. If the claim is a per-syndrome detection rate, it is contradicted by the presented data.","section":"Abstract, Figures 3 and 4"}],"minor_comments":[{"comment":"The abstract states overhead \"lower than 0.3% the decoding time,\" but Section 5.3 reports measured overheads between 0.1% and 0.2% for distances 3–19. These numbers should be reconciled.","section":"Abstract and Section 5.3"},{"comment":"The improvement figure is reported inconsistently: \"up to 20%\" in the abstract, \"upwards of 25%\" in Section 5.5, and \"about 25%\" in the Conclusion. Use a single, precisely defined quantity.","section":"Section 5.5 and Conclusion"},{"comment":"Reference [5] is an anonymous placeholder URL with \"to be disclosed after the review process.\" This is not acceptable for reproducibility; please provide a working repository link or a data-availability statement.","section":"References"},{"comment":"The table caption contains a typo: \"Instrinsic noise model\" should be \"Intrinsic noise model.\"","section":"Table 1"},{"comment":"The detection-rate plots claim \"no false positives\" without reporting a quantitative false-positive rate. Please state the number of false-positive detections or the false-positive rate outside the event window.","section":"Figures 3 and 4"},{"comment":"The 11 million shot figure is not reconciled with the reported sample sizes (128–512 per configuration). Please provide a breakdown of how the total number of shots is accumulated.","section":"Section 1 and Section 5"}],"recommendation":"major_revision","confidential_remarks":"The paper's core ideas are timely and the framework is promising, but the closed-loop validation and the absence of quantitative localization metrics make the headline claims premature. The \"first algorithm\" novelty claim is also strong; the authors should either substantiate it with a more thorough literature search or soften it. I would encourage the editor to send the revision back to the same referees to verify that the added robustness study and localization metrics are adequate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this paper has a real new idea—REI, a syndrome-history-based runtime detector for radiation events, plus RadMatching, a pre-decoding correction that flips suspect stabilisers before MWPM. I don't see that in the prior literature, and it deserves attention. The simulation campaign is large (11M shots) and the overhead measurements are honest. The multi-code logical error analysis in Section 5.4 is also a usable result about correlated logical faults.\n\nWhat is soft: the validation is closed-loop. REI's FIFO backlog, spatial clustering, weighted centroid, and radius formula are keyed to the exact temporal decay and inverse-square spatial falloff the authors generate in Equations 1–4. So the 100% detection and \"accurate\" localization are statements about the model, not about real cosmic-ray events. There is no sensitivity analysis over the model parameters, and no comparison with the published experimental correlated-error data (McEwen, Wilen). That is the load-bearing gap. Also, the abstract says \"accurately identify both the centre and the area,\" but Sections 5.1 and 5.2 never compare the returned radius and centroid against the injected locus—there is no quantitative localization accuracy metric. The RadMatching gain is shown for a single configuration (d=9, central fault, 384 shots), and the abstract's \"up to 20%\" doesn't match the text's \"upwards of 25%,\" though that is minor. Key thresholds (K_max, alpha, radius multiplier, correlation threshold) are unstated or unjustified, and the code is promised but not shipped. These are all addressable.\n\nOn the other hand, the core observation—that radiation creates persistent spatially clustered defect patterns distinguishable from intrinsic noise—is plausible and internally consistent. The authors do not oversell into a hardware solution; they present a simulation result with clear limits. The citation pattern to the radiation and QEC literature is fine.\n\nBottom line: send it out for peer review. But the referee should ask for (1) robustness tests across radiation-model parameters, (2) a quantitative localization error metric, and (3) ideally a test on recorded syndrome data from the experimental papers. Without those, the headline claims only hold inside the simulation.","headline":"A genuinely novel runtime radiation detector for surface codes, but the 100% numbers are only demonstrated inside the simulation loop built around the same model.","tokens_in":19758,"tokens_out":2299,"would_cite":false,"duration_ms":25047,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.Pp","85.25.-j"],"model":"deepseek-v4-flash","headline":"Radiation-induced faults in superconducting quantum chips can be detected and localised at runtime from syndrome measurements alone.","keywords":["quantum error correction","syndrome decoding","radiation faults","cosmic rays","superconducting qubits","surface code","fault detection","fault injection"],"falsifier":"Take a superconducting chip instrumented with a co-located cosmic-ray detector or a known radiation source, record the real syndrome stream during confirmed strikes, and run REI to compare its detection rate, impact centre, and radius against the known events; a detection rate below 100%, false alarms outside the strike window, or a systematic bias in the estimated centre would falsify the claim that syndrome correlations alone identify radiation events.","tokens_in":18781,"feed_emoji":"⚛️","tokens_out":6010,"duration_ms":63479,"temperature":0.7,"pith_summary":"This paper argues that the syndrome stream of a quantum error correction code carries enough of a cosmic-ray strike's signature to detect it at runtime, locate its impact point, and estimate its affected area with negligible overhead. The authors introduce REI, a QEC-agnostic routine that accumulates recent syndrome measurements, filters for spatially clustered per-qubit defect rates, and returns a weighted centroid and radius. Across more than 11 million simulated Rotated Surface code shots using a physics-derived radiation model, REI detects 100% of injected faults regardless of impact position, with overhead below 0.3% of decoding time. The paper then uses that information to flip stabiliser measurements inside the affected area before minimum-weight perfect matching decoding, a step they call RadMatching, which lowers logical error by roughly a quarter compared with radiation-unaware decoders. If the results transfer to real devices, radiation-aware decoding becomes a practical software complement to shielding and other hardware mitigations.","feed_headline":"Syndrome data catch 100% of radiation hits on quantum chips","feed_subtitle":"A runtime routine also finds the impact point and cuts logical error by about 25%.","key_machinery":"The carrying mechanism is the Radiation Event Identification (REI) subroutine, built around the physics-derived erasure model P(Δs, Δt_g, t) = S(Δs) · T(Δt_g, t), where the spatial falloff is S(Δs) = 1/(Δs+1)^2, the temporal erasure probability is T(Δt_g, t) = 1 − exp(−Δt_g/τ_rad(t)), and the radiation-shortened coherence time is τ_rad(t) = τ_1 exp(10((t−t_rad)/Δ_t_rad − 1)). REI turns this expected signature into a detection rule: it accumulates the last K_max syndrome rounds, converts them into per-qubit defect incidence rates, prunes values below $\\alpha$ = 1/((rounds+1)·S), verifies spatial correlation through the average nearest-neighbour distance, and estimates the impact centre as the defect-squared-weighted centroid with radius equal to twice the weighted average distance. RadMatching then applies the identified circular area to invert stabiliser measurements before MWPM decoding, giving the decoder a pre-cleaned syndrome.","core_discovery":"The central claim is that radiation-induced correlated errors can be detected and localised solely from syndrome information before decoding, by exploiting the spatial and temporal correlation of the fault: per-qubit defect rates near the impact point rise above the intrinsic-noise level and stay elevated across consecutive syndrome rounds. REI keeps a FIFO backlog of the last K_max syndrome vectors, normalises per-qubit defect incidence over time, prunes qubits below a threshold, checks that the surviving qubits are spatially clustered, and returns a weighted centroid and a radius estimate for the event. On simulated Rotated Surface codes of distance 5 to 19, with faults injected at central and peripheral positions, REI detects 100% of injected faults with no false positives outside the fault window, regardless of code basis or impact position. Feeding the estimated affected area into a minimum-weight perfect matching decoder as bit flips on affected stabiliser measurements yields RadMatching, which reduces logical error rate by upwards of 25% relative to standard MWPM and other graph-based decoders at the fault's most intense phase.","pith_inferences":["If real cosmic-ray events match the model's temporal and spatial profile, REI's output could trigger hardware-level responses such as pausing correction, flagging affected QEC rounds as unreliable, or dynamically remapping logical qubits away from the damaged region.","The correlated multi-code error spikes imply that future large chips should schedule or place logical qubits with radiation geometry in mind, for example by duplicating logical qubits in well-separated regions rather than clustering them around one chip area.","REI's radius and centroid could be used to adjust decoder edge weights continuously during a radiation transient, not just to flip binary stabiliser measurements, which might capture partial erasures more accurately than RadMatching's thresholded flip.","The 100% detection number is a simulation result tied to the injected model; a natural next test is to run REI on real syndrome streams from a chip with a co-located cosmic-ray detector and measure the true detection rate and localisation bias.",""],"forward_implications":["REI can be inserted before any graph-based decoder with under 0.3% overhead, so radiation awareness is compatible with real-time decoding time constraints.","Because REI is designed to be agnostic to the QEC code and the error basis, similar syndrome-correlation pre-processing should transfer to stabiliser codes beyond the Rotated Surface code.","Larger code distances dilute a radiation event across more stabilisers, so higher-distance codes preserve a lower fraction of corrupted stabilisers and stay detectable for a shorter tail of the event.","RadMatching lowers logical error by upwards of 25% at peak intensity relative to MWPM, while all four radiation-unaware graph decoders studied exceed roughly 50% logical error during the event.","On a chip hosting several distance-15 surface codes, correlated logical-error spikes track proximity to the impact point, with equidistant codes spiking together.",""],"supporting_citations":[{"why":"provides the stabiliser-circuit simulator used to generate the more than 11 million syndrome-measurement shots.","marker":"[24]"},{"why":"supplies the minimum-weight perfect matching decoder that serves as the baseline and as the core of the proposed RadMatching pipeline.","marker":"[28]"},{"why":"documents the temporal persistence and chip-wide extent of cosmic-ray error bursts, motivating the correlated temporal-spatial signature REI detects.","marker":"[36]"},{"why":"reports synchronous detection of cosmic rays and correlated qubit errors, grounding the assumption that such events imprint on syndrome measurements.","marker":"[27]"},{"why":"establishes that ionizing radiation reduces superconducting qubit coherence, the physical effect the model expresses through a reduced tau_rad.","marker":"[44]"},{"why":"shows correlated charge noise and relaxation errors in superconducting qubits, supporting the chip-wide correlated corruption assumption.","marker":"[45]"},{"why":"provides the SI1000 intrinsic-noise model that sets the baseline noise rate p = 10^-5 against which radiation events are distinguished.","marker":"[25]"},{"why":"previously argued that surface codes alone cannot compensate for radiation events, motivating the need for radiation-aware decoding.","marker":"[42]"}],"fun_headline_variants":["Syndrome data reveal every radiation hit on quantum chips","100% of radiation hits on qubits caught via syndrome","Radiation fault locator for quantum chips: 100% hit rate","Qubit faults from radiation: 100% caught, fixed in real time","Syndrome-only method catches 100% of quantum chip radiation hits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim rests on the radiation model in Equations 1 to 4 faithfully describing real cosmic-ray strikes: if actual events decay with a different time profile, spread across the chip differently, or corrupt syndrome measurements in a way not proportional to S times T, then the 100% detection rate and the localization accuracy are not guaranteed to transfer.","fun_headline_variants_meta":{"raw":{"variants":["Syndrome data reveal every radiation hit on quantum chips","100% of radiation hits on qubits caught via syndrome","Radiation fault locator for quantum chips: 100% hit rate","Qubit faults from radiation: 100% caught, fixed in real time","Syndrome-only method catches 100% of quantum chip radiation hits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001381,"raw_usage":{"total_tokens":5604,"prompt_tokens":967,"completion_tokens":4637,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":583,"completion_tokens_details":{"reasoning_tokens":4547}},"tokens_in":583,"tokens_out":4637,"duration_ms":32849,"temperature":1.0,"reasoning_tokens":4547,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:18:18.593227+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a superconducting chip instrumented with a co-located cosmic-ray detector or a known radiation source, record the real syndrome stream during confirmed strikes, and run REI to compare its detection rate, impact centre, and radius against the known events; a detection rate below 100%, false alarms outside the strike window, or a systematic bias in the estimated centre would falsify the claim that syndrome correlations alone identify radiation events.","supporting_citations":[{"cited_title":"Bench- marking the Planar Honeycomb Code","cited_arxiv_id":null,"evidence_quote":"reports synchronous detection of cosmic rays and correlated qubit errors, grounding the assumption that such events imprint on syndrome measurements."}],"review_version":2}