{"id":"244ed033-1148-4a99-a351-26a9f79e1c26","arxiv_id":"2504.19083","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"Simulation-only study of a SiPM-readout plastic scintillator claims 22.29 ps timing and 1.5 mm position resolution with CNN, but lacks experimental validation.","lead":"A GEANT4 simulation of a 200 mm by 200 mm plastic scintillator with 64 silicon photomultipliers reports millimeter-level position resolution and a 22.29 ps timing resolution. The paper also applies a convolutional neural network to timing data, claiming 1.5 mm position resolution, but no experimental measurements are presented.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Simulation-only performance claims: unvalidated GEANT4 chain is the load-bearing assumption; no benchmark or uncertainty analysis supports the 22.29 ps and 1.5 mm figures.","rationale":"This is not a claim that simulation studies are inadmissible; a well-validated GEANT4 model with a measured SPE response and checked NPE distributions could support design conclusions. The problem is that the paper's language moves from simulation to achieved performance without any anchor to experiment. The reader's rejection is therefore correct, but for a reason of evidence rather than a demonstrated numerical error. A single benchmark test would settle the question. The geometric inconsistency in Section 4.1 is real and specific, but it is narrower than the validation gap: it does not touch the CNN-based 1.5 mm result or the 22.29 ps timing result, both of which depend on the same unvalidated simulation chain. Thus the most load-bearing concern remains simulation fidelity, and the reader's weakest-assumption identification captures it accurately.","tokens_in":9405,"tokens_out":11109,"duration_ms":125417,"concrete_test":"Reproduce the exact GEANT4 + digitization chain on a benchmark with published measured data—e.g., a 3×3×50 mm³ EJ-200 bar coupled to one S13360-6025PE SiPM—and compare simulated photoelectron count and CFD timing resolution with the published experimental values. If the simulated benchmark does not match measurement within quoted uncertainties, the untested 22.29 ps and 1.5 mm numbers cannot be accepted as detector performance.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Both headline numbers—22.29 ps timing (Section 3) and 1.5 mm CNN resolution (Section 4.2)—are single-point outputs of one unbroken simulation chain: GEANT4 optical transport (§2.2), SPE waveform superposition (Eq. 1), CFD timing, and supervised CNN training. For these numbers to describe the proposed detector, that chain must be a faithful surrogate for the physical system. The manuscript supplies no evidence for that surrogacy: there is no comparison of simulated light yield, NPE spectra, or timing distributions to any measurement, no benchmark detector simulation, no uncertainty quantification on optical parameters, and no released code or data. The abstract and summary nevertheless state the results as achieved performance ('a position resolution of mm level has been achieved'; 'delivering exceptional temporal sensitivity and spatial precision'), so the central claim is readable only as a claim about the real detector. A separate internal inconsistency affects Section 4.1: Eq. 3 assumes straight-line time-of-flight immediately after the text concedes that most detected photons undergo multiple reflections; this undermines the geometric method, though not the CNN or timing results. The dominant load-bearing risk is therefore simulation fidelity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a simulation study of a position-sensitive plastic scintillator detector consisting of a 200 mm × 200 mm × 6 mm EJ-200 scintillator read out by 64 SiPMs arranged on its four sides. Using GEANT4 simulations of 3 GeV/c muons, the authors simulate optical photon transport, synthesize SiPM waveforms from individual photoelectron responses, and extract channel times with a constant-fraction discriminator. They report a weighted-average time resolution of 22.29 ps (Section 3), a geometric reconstruction based on circle intersections and a \"largest empty circle\" search with roughly 3.4 mm resolution (Section 4.1), and a CNN regressor trained on simulated timing data that achieves approximately 1.5 mm resolution in both horizontal and vertical directions (Section 4.2). The paper concludes that the design provides millimeter-level spatial resolution and sub-30 ps timing performance.","tokens_in":9558,"tokens_out":4262,"duration_ms":42395,"significance":"If the claimed performance were demonstrated on a physical detector, the design would be a compact, low-cost option for beam position monitoring at facilities like XiPAF. The manuscript gives a concrete geometry, a clear simulation chain, and enough detail to reproduce the simulation setup, which is a positive feature. However, the significance is strongly limited by the fact that all headline results are single-point outputs of an unvalidated simulation; no experimental data, benchmark comparison, or uncertainty quantification is provided. In addition, the geometric algorithm in Section 4.1 contains an internal inconsistency, and the CNN evaluation in Section 4.2 measures only the network's ability to reproduce the simulation's mapping rather than the detector's physical resolution. The paper is best read as a preliminary simulation study, not as a demonstration of achieved detector performance.","major_comments":[{"comment":"The geometric reconstruction converts photon arrival times into straight-line distances using L = tc/n, assuming line-of-sight propagation. This directly contradicts the same subsection's statement that \"the vast majority undergo multiple reflections within the scintillator.\" Since the arrival time is dominated by multiply-reflected paths, Eq. (3) does not measure the distance from the emission point to the SiPM, and the circle-intersection construction in Figs. 8-11 is therefore built on an invalid premise. The 3.4 mm resolution claim and Table 1 results are not supported by a consistent physical model.","section":"Section 4.1, Eq. (3)"},{"comment":"The reported ~1.5 mm CNN position resolution is obtained by training the network on GEANT4-simulated timing data with simulated truth labels and then evaluating on the same simulation model. This is a self-consistency check of the network's ability to fit the simulation's input-output mapping; it provides no evidence about performance on a physical detector. A meaningful resolution claim would require experimental data, or at minimum a demonstration that the simulation reproduces measured quantities and an estimate of the simulation-to-reality domain gap.","section":"Section 4.2, Figs. 15-16"},{"comment":"Every headline number (22.29 ps in Section 3, ~3.4 mm in Section 4.1, ~1.5 mm in Section 4.2) depends on an unvalidated simulation chain: GEANT4 optical transport, the SPE waveform superposition of Eq. (1), CFD timing, and the CNN training. The manuscript provides no comparison of simulated NPE spectra, waveform shapes, or timing distributions to any measurement, no benchmark against a known detector, and no uncertainty analysis on critical optical parameters such as the EJ-200 refractive index, surface treatment, or SiPM PDE. Without such validation, the results are assumption-laden simulation outputs and cannot be presented as achieved detector performance.","section":"Section 2.2 and all results"},{"comment":"The 22.29 ps time resolution is presented as a single Gaussian fit to a simulated distribution, but the paper does not state the number of simulated events, the fit range, or the statistical uncertainty on the resolution. In addition, the claim that the SiPM transit-time spread (standard deviation 20 ps) \"does not affect the outcome\" is not justified, which is concerning for a result whose central quantity is a time resolution. The paper should provide the statistical and systematic uncertainty on the quoted value.","section":"Section 3, Fig. 6"}],"minor_comments":[{"comment":"The text specifies an S13360-6025PE SiPM, but the cited reference [16] links to the S13360-3025PE model. The authors should verify which model is used and correct the reference or the text accordingly.","section":"Section 2.1 and Ref. [16]"},{"comment":"The table header contains a stray \"/2\" and the column labels are poorly formatted, making the table harder to read than necessary.","section":"Table 1"},{"comment":"The largest-empty-circle algorithm depends on an unspecified square search-region side length and a 200×200 grid resolution; these parameters affect the reconstructed positions in Table 1 and Figs. 10-13 and should be stated explicitly.","section":"Section 4.1"},{"comment":"The wording \"a position resolution of mm level has been achieved\" and \"delivering exceptional temporal sensitivity\" overstates results that are purely simulated. The paper should consistently refer to simulated performance until experimental validation is available.","section":"Abstract and Section 5"},{"comment":"There are several minor grammatical and typographical errors, e.g., \"timimg\" in Section 3, \"Relation\" instead of \"Equation\" for Eq. (1), and inconsistent use of \"SiPM\" versus \"SiPMs.\" A careful proofread is needed.","section":"Throughout"}],"recommendation":"reject","confidential_remarks":"The central claims are simulation-only and are stated as achieved performance in the abstract and summary. The lack of any experimental validation or comparison to measured data, together with the internal inconsistency in the geometric algorithm and the self-referential CNN evaluation, make the headline numbers unsupported. This is a substantive issue that cannot be fixed by local revision; the work would need either experimental data or a fundamental reframing as a design study with explicit caveats and uncertainty quantification, which likely falls outside a standard revision cycle."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Briefly, this is a GEANT4-only design study, and the abstract sells simulation outputs as achieved detector performance. That framing is the main problem. But there is real content here: the specific geometry, a bare 200x200x6 mm EJ-200 plate with 64 SiPMs coupled around the edges, and the maximal void circle reconstruction idea, are new to me. The simulation chain is standard but decently described: optical transport, SPE waveform superposition from a measured SPE, CFD timing, and a photoelectron-weighted average. The CNN part is a standard regression, though the 2x16x2 encoding of the 64 channels is a thoughtful way to exploit the detector symmetry.\n\nWhere the paper gets soft is in what the numbers mean. The 22.29 ps time resolution and the 1.5 mm CNN position resolution are single-point outputs of one unbroken simulation chain. No measurement, no benchmark comparison, no uncertainty on optical parameters, no code or data. The abstract says mm level has been achieved, which a normal reader understands as a claim about the physical detector. The authors should be much more careful.\n\nThe geometric algorithm has a genuine internal inconsistency. Eq. 3 converts arrival time to distance using the bulk speed of light, immediately after the text tells us that most photons undergo multiple reflections before reaching the SiPMs. Multiply-scattered photons do not travel along straight paths, so that step is not justified. This undermines the geometric method's claim, though it does not affect the CNN or timing results.\n\nOn the CNN: I would not call it circular. Training on GEANT4 truth and evaluating on GEANT4 test events is a valid test of whether the network can invert the simulation's mapping. It is not, however, a detector resolution. If the simulation is wrong, the 1.5 mm is meaningless. And the void-circle search still has unspecified hand-tuned parameters, which makes the result hard to reproduce.\n\nWho should read this? People working on compact scintillator beam monitors, especially at XiPAF-like facilities, and anyone interested in ML-based position reconstruction. As a design exploration it is useful; as a detector characterization it is not. The paper deserves peer review rather than desk rejection, because the geometry and the void-circle idea are testable and the simulation work is coherent. The revision burden is a reframing, a fix or honest discussion of the geometric assumption, some uncertainty quantification, and ideally a bench measurement of a small prototype.\n\nMy own take: with the current framing I would not cite the numbers, but I would follow the work.","headline":"Simulation-only design study with a new geometry and a flawed geometric reconstruction; the headline numbers are simulation outputs, not measured performance, but the work is coherent and deserves a referee.","tokens_in":10174,"tokens_out":2875,"would_cite":false,"duration_ms":29535,"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":"The paper claims a flat 200×200×6 mm plastic scintillator with 64 SiPMs can reconstruct simulated muon hits to about 1.5 mm using a CNN trained only on timing data, with 22.29 ps timing resolution.","keywords":["plastic scintillator","silicon photomultiplier","position-sensitive detector","time resolution","spatial resolution","convolutional neural network","GEANT4 simulation","beam position monitoring"],"falsifier":"Build the described 200×200×6 mm scintillator with 64 edge-coupled SiPMs, send a collimated source or scanned beam to known positions, and apply the same weighted-time and CNN algorithms: measured residuals far from 1.5 mm, or a time spread far from 22.29 ps, would falsify the central claim. A faster check is to measure the actual single-photoelectron waveform and transit-time jitter and rerun the simulation with those measured inputs; if the predicted resolution degrades to match the measurement, the simulation assumptions are the identified cause.","tokens_in":9145,"feed_emoji":"🎯","tokens_out":11254,"duration_ms":99483,"temperature":0.7,"pith_summary":"The paper tries to show that a flat, unwrapped plastic scintillator panel surrounded by 64 silicon photomultipliers can serve as a cheap, compact position-sensitive detector, replacing expensive photomultiplier tubes and complex fiber couplings used in earlier designs. All reported numbers come from a GEANT4 simulation of 3 GeV/c muons: a timing resolution of 22.29 ps, a geometric position reconstruction at about 3.4 mm, and a convolutional-neural-network position reconstruction at about 1.5 mm in both X and Y. A sympathetic reader would care because, if the simulation is faithful, this geometry gives millimeter-level beam or particle positioning with straightforward mechanics and low-cost readout.","feed_headline":"Simulated plastic scintillator plus neural net hits 1.5 mm positions","feed_subtitle":"A 64-channel SiPM panel localizes hits from timing data alone, with 22 ps timing resolution in simulation.","key_machinery":"The mechanism that carries the argument is the spatial encoding of photon arrival times. Photons created at the muon track reach nearby SiPMs sooner, so the 64 readout times form a pattern that shifts with hit position, and the detector's fourfold symmetry makes this pattern translationally invariant. The geometric branch uses $L = tc/n$ to turn each arrival time into a circle, pairs the 64 SiPMs to draw 2016 intersecting circles, and locates the hit at the center of the largest circle empty of intersection points. The learning branch reshapes the 64 times into a $(2,16,2)$ tensor that respects the left/right and up/down symmetry and feeds it through two convolutional layers with ReLU, batch normalization, average pooling, and dropout, followed by fully connected regression trained with MSE loss and the Adam optimizer. The timing branch uses a photoelectron-weighted average, Eq. (2), to suppress the strong dependence of single-channel timing on hit position.","core_discovery":"The central claim is that the timing pattern across the 64 edge-mounted SiPMs contains enough spatial information to localize a hit well below the detector's 200 mm scale. The paper shows that a CNN trained to regress hit coordinates from the per-channel times reconstructs both coordinates with about 1.5 mm resolution, roughly twice as good as a geometric algorithm that converts arrival-time differences into intersecting circles and finds the largest empty circle. The paper also claims that a photoelectron-weighted average of the 64 channel times removes position-dependent timing jitter and yields a single-peaked arrival-time distribution with 22.29 ps resolution. These results are presented as simulation outcomes, demonstrating that a bare scintillator with symmetric SiPM readout can combine fast timing with precise position in one simple structure.","pith_inferences":["The 1.5 mm figure is a simulation ceiling; dark counts, crosstalk, imperfect optical coupling, and electronic noise will add real-world position jitter, so the measured resolution should be expected to be somewhat worse until the simulation is tuned to bench data.","The CNN may be fitting simulation-specific artifacts such as the assumed 20 ps transit-time spread and idealized optical surfaces, so transfer to a real detector may require retraining on measured waveforms or domain adaptation.","Following the paper's own suggestions, adding photoelectron count or pulse amplitude as extra input channels, or replacing the largest-empty-circle fit with Hough circle detection, are natural next steps that should improve both methods.","The paper's motivating application demands better than 1 mm, so the simulated 1.5 mm result does not yet meet that target; the contribution is evidence that the compact geometry is in the right regime, with the final push left to further algorithms or hardware refinements."],"forward_implications":["A detector of this geometry could act as a beam-position monitor with millimeter-level accuracy while avoiding the mechanical complexity of slotted scintillators and fiber readout.","Because the CNN uses only timing information, position reconstruction could in principle proceed without precise per-channel gain or charge calibration.","The 22.29 ps weighted-average timing resolution would allow the same panel to serve as a time-of-flight counter while simultaneously reporting hit position.","The two reconstruction methods cross-check each other: the geometric result anchors the interpretation, and the CNN improvement quantifies the nonlinear information available in the timing pattern."],"supporting_citations":[{"why":"Supplies the scintillator's 0.9 ns rise time and 2.1 ns decay time, which set the photon emission timing in the simulation.","marker":"[14]"},{"why":"Justifies the bare, unwrapped scintillator design by showing that fewer reflections improve timing performance.","marker":"[15]"},{"why":"Supplies the SiPM parameters used in digitization: 6×6 mm² area, 25 µm microcell, gain 10^6, and 40% quantum efficiency at 420 nm.","marker":"[16]"},{"why":"Provides the GEANT4 toolkit used to simulate muon transport, energy deposition, and photon propagation.","marker":"[17]"},{"why":"Gives the single-photoelectron waveform relation, Eq. (1), used to build each SiPM's output signal.","marker":"[19]"},{"why":"Establishes the per-channel timing behavior of this detector structure and motivates the photoelectron-weighted average time.","marker":"[20]"},{"why":"Supplies the convolution definition and tensor formulation used to build the CNN's position regression.","marker":"[28]"},{"why":"Motivates the batch normalization layers that speed convergence and improve generalization in the CNN.","marker":"[29]"},{"why":"Motivates the dropout layers used in the CNN to reduce overfitting.","marker":"[30]"},{"why":"Defines the Adam optimizer update rule used to train the position reconstruction network.","marker":"[32]"}],"fun_headline_variants":["Neural net boosts scintillator position resolution to 1.5 mm","SiPM scintillator hits 1.5 mm position via CNN","Scintillator+SiPM: 1.5 mm position, 22 ps timing","CNN sharpens plastic scintillator to 1.5 mm"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole case rests on the GEANT4 simulation faithfully representing the real detector; if the simulated optical surfaces, SiPM photon detection efficiency, transit-time jitter, or electronic noise are optimistic, the measured position and timing resolutions will be worse than the reported 1.5 mm and 22.29 ps.","fun_headline_variants_meta":{"raw":{"variants":["Neural net boosts scintillator position resolution to 1.5 mm","SiPM scintillator hits 1.5 mm position via CNN","Scintillator+SiPM: 1.5 mm position, 22 ps timing","CNN sharpens plastic scintillator to 1.5 mm"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1387,"prompt_tokens":843,"completion_tokens":544,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":459,"completion_tokens_details":{"reasoning_tokens":464}},"tokens_in":459,"tokens_out":544,"duration_ms":5169,"temperature":1.0,"reasoning_tokens":464,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T06:01:32.873071+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build the described 200×200×6 mm scintillator with 64 edge-coupled SiPMs, send a collimated source or scanned beam to known positions, and apply the same weighted-time and CNN algorithms: measured residuals far from 1.5 mm, or a time spread far from 22.29 ps, would falsify the central claim. A faster check is to measure the actual single-photoelectron waveform and transit-time jitter and rerun the simulation with those measured inputs; if the predicted resolution degrades to match the measurement, the simulation assumptions are the identified cause.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the scintillator's 0.9 ns rise time and 2.1 ns decay time, which set the photon emission timing in the simulation."},{"cited_title":"IEEE Transactions on Nuclear Science, 2015, 62(5): 1972-","cited_arxiv_id":null,"evidence_quote":"Justifies the bare, unwrapped scintillator design by showing that fewer reflections improve timing performance."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the SiPM parameters used in digitization: 6×6 mm² area, 25 µm microcell, gain 10^6, and 40% quantum efficiency at 420 nm."},{"cited_title":"GEANT4—a 36 simulation toolkit[J]","cited_arxiv_id":null,"evidence_quote":"Provides the GEANT4 toolkit used to simulate muon transport, energy deposition, and photon propagation."},{"cited_title":"CDF Note 5358, 2000","cited_arxiv_id":null,"evidence_quote":"Gives the single-photoelectron waveform relation, Eq. (1), used to build each SiPM's output signal."},{"cited_title":"Deep learn- ing[M]","cited_arxiv_id":null,"evidence_quote":"Supplies the convolution definition and tensor formulation used to build the CNN's position regression."},{"cited_title":"Using deep neural networks to improve the precision of fast-sampled particle timing detectors","cited_arxiv_id":"2312.05883","evidence_quote":"Motivates the batch normalization layers that speed convergence and improve generalization in the CNN."}],"review_version":1}