{"id":"2d3ada63-d7c1-4ed2-b1ce-056c3d3d41d7","arxiv_id":"2603.03561","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A Frobenius-norm template-matching algorithm estimates antineutrino direction from neutron-capture patterns in segmented detectors, with an optimal segment size near the mean neutron track length (~70 mm) at low counts.","lead":"This paper tests a pattern-matching method for telling which direction an antineutrino beam came from by comparing the 2D pattern of neutron captures in a segmented detector against simulated templates. It reports that the best detector segment size in the low-count regime is about 70–75 mm, roughly the distance a neutron travels before being captured—useful for reactor monitoring and pointing to neutrino sources.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Angular uncertainty is reported as estimator scatter, never checked for 68% coverage; at n≤10 synthetic uniform angles are mixed in, so 'more realistic' is unsupported.","rationale":"The reader's weakest assumption (closed-loop simulation) is real but is a standard scope limitation of any simulation-based algorithm paper; it would be addressed by independent validation but does not by itself invalidate the algorithm's statistical behavior within the model. A more immediate and internal threat is that the paper never establishes that its reported δϑ is a calibrated confidence interval. The algorithm's uncertainty is the spread of reconstructed angles over λ iterations. For an uncertainty to be 'realistic', 68% of repeated experiments should have the true direction inside δϑ. At low n, the estimator often gives a flat FND and the authors replace those iterations with random angles; this mixture can make the circular standard deviation large without corresponding to 68% coverage. The paper's only evidence that the uncertainty is more realistic is that it is larger than Eq. (1), which conflates conservatism with correctness. A coverage study on simulated pseudo-experiments with known true direction would settle this within the paper's own framework. If coverage is close to 68%, the central claim is supported; if not, the method needs revision or the claim needs to be weakened. This is why I keep the verdict CONDITIONAL/UNCHANGED rather than moving to ACCEPT or REJECT.","tokens_in":16739,"tokens_out":7644,"duration_ms":79111,"concrete_test":"Run a coverage study on simulated pseudo-experiments: for each n in {10, 30, 100, 300} and each segment size in {5, 50, 150 mm}, generate m=1000 independent datasets from the RAT-PAC2 capture distribution at a known true angle ϑ_true. For each dataset, run the FND algorithm exactly as in the paper (including synthetic-random imputation at low n) and record the reconstructed angle and δϑ. Then compute empirical coverage: the fraction of datasets for which the circular distance between the reconstructed angle and ϑ_true is ≤ δϑ. Also compute the same coverage for Eq. (1) on the same datasets. If the FND coverage is not within statistical uncertainty of 68% (or if it undercovers while Eq. (1) overcovers), the 'more realistic' claim fails. Report coverage as a function of n and check the sensitivity to removing the synthetic imputation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline claim is that the FND pattern-matching estimator yields a 'higher (and thus more realistic) angular uncertainty' at low event counts than the conventional Chooz formula. The load-bearing premise is not just simulation fidelity; it is that the reported circular standard deviation δϑ is a calibrated 1σ uncertainty on the true source direction. The paper never tests this. δϑ is computed as the spread of point estimates across λ Monte Carlo iterations (Appendix A), which is an estimator-variance measure, not a confidence interval. For n≤10 the estimator frequently returns 'no directional information' and the authors substitute uniform random angles from [−π,π] (Appendix A, 'Use of synthetic data'). The resulting mixture distribution (spikes + uniform) is not von Mises; applying the von Mises MLE and σ = sqrt(−2 ln R) can yield a number that does not correspond to 68% coverage. A larger δϑ than Eq. (1) is not automatically 'more realistic': it is more conservative only if 68% of reconstructed angles fall within δϑ of the true direction. Without a coverage check on simulated pseudo-experiments with known true angle, the central claim that the method improves uncertainty quantification is unsupported. This concern is internal to the paper's own simulation and does not depend on external detector physics. If coverage is badly miscalibrated (e.g., 40% or 90% instead of 68%), the method's reported uncertainties are not trustworthy regardless of simulation fidelity.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a directionality reconstruction algorithm for segmented inverse-beta-decay (IBD) antineutrino detectors. The method compares a binned 2D matrix of neutron-capture locations in an unknown 'empirical' dataset against reference matrices generated from simulated IBD events at different incident neutrino angles, using the Frobenius norm of the matrix difference (FND). The angle minimizing the FND (after fitting with |sin θ|) is taken as the reconstructed direction; the angular uncertainty δϑ is the circular standard deviation of reconstructed angles over λ Monte Carlo iterations. The authors claim this pattern-matching approach yields a 'higher (and thus more realistic)' angular uncertainty than the conventional Chooz formula at low event counts, and they identify an optimal segment size (≈73 mm for 0.1% 6Li doping) close to the simulated mean neutron track length. Validation is performed entirely within the RAT-PAC2 simulation framework, with both the 'empirical' data and the reference templates generated by the same code and, for the main results, the same underlying 10M-event dataset. The paper also reports a usable-event variant, discusses detector geometries, and outlines applications to safeguards and geo-neutrinos.","tokens_in":17087,"tokens_out":3360,"duration_ms":36333,"significance":"If the algorithm performs as claimed, it would provide a low-statistics direction estimator for segmented IBD detectors and a design heuristic linking optimal segment size to neutron mean track length. The computational pattern-matching idea is simple and potentially applicable beyond this specific problem. The paper makes a falsifiable prediction (optimal segment size ≈ mean neutron track length) and uses a publicly available simulation toolkit. However, the central validation is closed-loop: both the pseudo-data and the templates come from the same Monte Carlo, and the reported uncertainty is an estimator-scatter measure never tested for calibration. The claimed improvement in 'realistic' angular uncertainty is therefore unproven even within the paper's own framework.","major_comments":[{"comment":"The validation is closed-loop: the 'empirical' dataset and the reference template matrices are both products of RAT-PAC2, and the main results use the same 10M-event fiducial dataset (Appendix C). A self-consistency test of this kind cannot establish that the reported angular uncertainties are 'more realistic' than those from the Chooz formula. At minimum, the paper needs a cross-validation with an independent simulation (different transport code or a withheld half of the dataset not used to build templates) and a clear statement that the results currently demonstrate internal consistency only. As written, the abstract and conclusion overclaim.","section":"Abstract and Appendix C"},{"comment":"The reported δϑ is the circular standard deviation of point estimates across λ iterations; this is an estimator-variance measure, not a calibrated confidence interval. The paper never checks whether 68% of reconstructed angles actually fall within δϑ of the true direction. Moreover, for n≤10 the authors replace failed fits with synthetic uniform random angles; the resulting mixture (spikes + uniform) is not a von Mises distribution, so applying the von Mises MLE and σ = sqrt(-2 ln R) has no known coverage interpretation. A coverage plot from pseudo-experiments with known true angles is essential to support the claim that a larger δϑ is 'more realistic.' The synthetic-data substitution should also be reported separately from genuine fits.","section":"Appendix A, Eq. (A.9), 'Use of synthetic data'"},{"comment":"The central shape assumption of the method—that the FND is proportional to |sin((ϑ0−ϑ)/2)|—is cited to the authors' companion paper [42] and is not derived or even summarized here. Since this functional form is load-bearing for both the angle extraction and the uncertainty interpretation, the manuscript should include a self-contained derivation or at least a clear statement of the conditions under which it holds, with a check that those conditions are satisfied in the simulation settings used.","section":"Section 'DEVELOPING AN ALGORITHM', Fig. 12"},{"comment":"The optimal segment size of 73 mm is found from a polynomial fit to FND-based angular uncertainties, all obtained with the same RAT-PAC2 simulation and with the stated approximations (single-segment prompt, neglected 1/r2, negligible core size, no escaping neutrons). The proximity of the optimum to the simulated mean track length (70 mm) is presented as 'interesting,' but both quantities are outputs of the same transport model, so the near-equality is partly a test of model self-consistency. A sensitivity study varying these assumptions, or at least a discussion of how they could shift the optimum, is needed before this can be a design guideline.","section":"Fig. 14"}],"minor_comments":[{"comment":"Equation numbering begins at (A.3); A.1 and A.2 are missing. Also, 'CFND V-plots' is used in the text but the abbreviation CFND is not defined.","section":"Appendix A"},{"comment":"The number of iterations λ used for each point in Fig. 13 is not stated (only Fig. 14 mentions λ=300). The correction in Eq. (A.11) depends on λ and λ_synthetic, so the reader cannot judge the error bars. Please state λ for every plotted point and indicate which points used the synthetic-data correction.","section":"Fig. 13 and Table III"},{"comment":"The fit parameters in Table III have very large uncertainties (e.g., b = 98.66 ± 49.79 for 5 mm). This suggests the fit is poorly constrained in the low-count region that is the paper's main focus; a statement about the fit range and stability is needed.","section":"Eq. (2), Table III"},{"comment":"The text refers to 'the inflection point (n = 30 in Fig. 17)', but Fig. 17 does not show an obvious inflection; please clarify the criterion or point to the specific feature.","section":"Conclusion"},{"comment":"The code name is written inconsistently: 'RAT-PAC2' in the abstract, 'RAT-PAC 2' elsewhere, and 'RATPAC2' in captions and Appendix C. Also, some figure captions (e.g., Fig. 16) are incomplete.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The central difficulty is the closed-loop simulation validation combined with the use of estimator scatter as a calibrated uncertainty. The authors may reasonably argue that all Monte Carlo studies are self-consistent to some degree, but the specific claim of 'more realistic' uncertainty requires a calibration test. A well-executed coverage study, plus separation of the synthetic-angle cases, would substantially strengthen the paper. I also note that the key mathematical claim is delegated to the authors' own companion paper [42]; for a standalone publication this needs to be addressed. The paper has merit as a new analysis technique description, but in its current form the quantitative results (including the 73-mm optimum) are not fully supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this is a coherent, honest simulation study, but the headline claim goes beyond the evidence. The stress-test note is right—there is no coverage test anywhere in the paper. δϑ is reported as the spread of point estimates across λ iterations, which is an estimator-variance measure, not a calibrated 1σ interval. For n≤10, uniform random angles are mixed in when the fit fails, so the reported circular standard deviation from a von Mises fit on a spiked-plus-uniform distribution is not a meaningful 68% number. 'Higher' does not automatically mean 'more realistic' unless you show that 68% of reconstructed angles fall within δϑ of the truth.\n\nWhat is actually new: the optimal segment-size scan (73 mm vs the ~70 mm mean neutron track length) and the repeated-iteration uncertainty scheme. The paper also does well at stating its assumptions—1/r², core angular size, escaping neutrons—and the simulation validation appendix is substantive. It is upfront that both the 'empirical' pseudo-data and the templates come from RAT-PAC2, so this is a self-consistency test, not a test against nature. That is a limitation, but not a fatal one for a method paper.\n\nSoft spots, in order of importance. The coverage issue is the big one. Without a check on simulated pseudo-experiments with known true angles, the entire 'more realistic uncertainty' claim is unsupported. Second, the comparison to the conventional Chooz method is not head-to-head: Fig. 3 is from a previous paper, not computed on the same simulated data. That should be fixed. Third, the synthetic-data imputation is a patch; it should be justified or the low-n points removed. Fourth, the 73 mm optimum is a fitted minimum interpreted post hoc; the 1.04 ratio is suggestive, not a result, and the error bars on that minimum are not shown.\n\nNone of these are fatal to the method itself. The algorithm may well be useful for low-statistics directionality in segmented detectors. But the paper as written needs revision: matched pseudo-experiments with coverage checks, a head-to-head against Eq. (1), and a decision on how to handle failed fits.\n\nWho is this for? People designing segmented IBD detectors for safeguards, small-reactor monitoring, or geo-neutrino work. It deserves a serious referee—yes—but the referee should push for the coverage analysis and the comparison. I would send it to review rather than desk-reject, with the expectation of major revision.","headline":"A clean, readable simulation study of a template-matching directionality estimator, but the central claim that its uncertainties are 'more realistic' at low counts is not actually demonstrated because the paper never checks coverage.","tokens_in":17602,"tokens_out":2199,"would_cite":false,"duration_ms":26306,"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":"A pattern-matching algorithm that compares binned neutron-capture maps estimates antineutrino direction more realistically than the standard resolution formula, and finds an optimal segment size near the neutron's mean travel distance.","keywords":["antineutrino directionality","inverse beta decay","segmented scintillator detector","pattern matching","Frobenius norm","angular resolution","neutron capture","reactor monitoring"],"falsifier":"Compare the simulated neutron-capture displacement distribution (mean and spread of the Euclidean distance from IBD vertex to capture) with measurements from a running segmented 6Li detector at a known reactor; if the pattern-matching angular uncertainty at a given event count, or the segment-size optimum, differs from the simulation by more than the quoted error bars, the central claim is contradicted.","tokens_in":16575,"feed_emoji":"🧭","tokens_out":6867,"duration_ms":58360,"temperature":0.7,"pith_summary":"The paper proposes that antineutrino direction in segmented inverse-beta-decay (IBD) detectors should be reconstructed by pattern matching: bin the positions of neutron captures relative to the prompt vertex into a 2D histogram, rotate a simulated template through all angles, and take the angle that minimizes the Frobenius norm of the difference between template and data. This replaces the conventional resolution formula, which the authors argue can return misleadingly small angular uncertainties at low event counts because it treats the limitation as instrumental resolution rather than the intrinsic kinematic spread of the neutron. Using Monte Carlo 'empirical' datasets, the algorithm converges to a higher, more realistic angular uncertainty at low counts and identifies an optimal segment size of roughly 73 mm for 0.1% lithium-6 loading, close to the 70 mm mean neutron track length. A sympathetic reader would care because the method provides a data-driven way to estimate direction when events are scarce and gives a design guideline for choosing segment geometry in future detectors.","feed_headline":"Capture-map matching beats standard formula at low counts","feed_subtitle":"A new estimator yields realistic angular uncertainty and sets optimal segment size near the neutron's travel distance.","key_machinery":"The central object is the Frobenius norm of the difference between two binned matrices: one holds the neutron-capture counts per segment (relative to the prompt segment) for the measured 'empirical' dataset, the other holds the same for a simulated template rotated through angle θ. The norm is the square root of the sum of squared element-wise differences. The algorithm computes this norm at many rotations, fits the resulting curve with an absolute-value sine function, and reads the minimum as the reconstructed neutrino direction. The same norm functions as a general 2D pattern-matching distance and can be applied beyond neutrino physics. Uncertainty is quantified by repeating the reconstruc","core_discovery":"The central claim is that directional information in a segmented IBD detector resides in the full pattern of neutron-capture positions relative to the prompt vertex, not just in the mean prompt-delayed displacement. The authors show that by binning capture positions into matrices and comparing them via the Frobenius norm of the difference, the reconstructed direction is the angle that minimizes this norm, with the angular dependence well described by an absolute-value sine function. On Monte Carlo data, the algorithm yields angular uncertainties that do not collapse to zero at low event counts, in contrast to the conventional formula, and it places the optimal segment size at 73 mm for 0.1%","pith_inferences":["Editorial inference: If the optimal segment size is set by the mean neutron track length, then changing the 6Li loading (which shortens the diffusion path) or switching target materials should shift the optimum; a simulation scan at 0.5% loading would test this prediction directly.","Editorial inference: Because the algorithm is validated only against the same simulation code that generates the templates, the reported angular uncertainties and the 73 mm optimum should be treated as simulation-level estimates; real detector data could move them.","Editorial inference: The Frobenius-norm template matching is a generic technique; it could be applied to other direction-sensing problems such as gamma or fast-neutron imaging with coded apertures, wherever a simulated template can be rotated against binned measurements."],"forward_implications":["At low event counts (tens to hundreds of IBDs), the pattern-matching algorithm gives a larger, more realistic angular uncertainty than the conventional formula, which is relevant for geo-neutrino studies and supernova pointing where events are scarce.","The optimal segment size for 0.1% 6Li-loaded scintillator is about 73 mm, nearly equal to the 70 mm mean neutron travel distance; this gives a concrete design rule: match segment size to the neutron moderation length to balance sparse matrices against central-bin dilution.","Events whose prompt and delayed signals occur in the same segment, which the conventional method discards as carrying zero information, can be included in the pattern-matching framework and weighted as needed.","The algorithm is not limited to 2D segmentation: the authors state it can be extended to 3D segmented geometries and, more broadly, to any computationally efficient 2D pattern-matching problem."],"fun_headline_variants":["Matrix-distance algorithm sharpens neutrino direction at low counts","Better angular uncertainty for antineutrino detectors with new estimator","Optimal segment size found for neutrino directionality in detectors","Pattern-matching beats standard method for neutrino direction sensing","Frobenius norm tuning improves reactor antineutrino detection"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the Monte Carlo simulation used to generate the 'empirical' data faithfully represents a real segmented detector; if the simulated neutron transport, 6Li capture distribution, or segment response differs from nature, every reported angular uncertainty and the 73 mm optimum would change.","fun_headline_variants_meta":{"raw":{"variants":["Matrix-distance algorithm sharpens neutrino direction at low counts","Better angular uncertainty for antineutrino detectors with new estimator","Optimal segment size found for neutrino directionality in detectors","Pattern-matching beats standard method for neutrino direction sensing","Frobenius norm tuning improves reactor antineutrino detection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000191,"raw_usage":{"total_tokens":1159,"prompt_tokens":703,"completion_tokens":456,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":447,"completion_tokens_details":{"reasoning_tokens":374}},"tokens_in":447,"tokens_out":456,"duration_ms":4683,"temperature":1.0,"reasoning_tokens":374,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T19:03:41.897883+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the simulated neutron-capture displacement distribution (mean and spread of the Euclidean distance from IBD vertex to capture) with measurements from a running segmented 6Li detector at a known reactor; if the pattern-matching angular uncertainty at a given event count, or the segment-size optimum, differs from the simulation by more than the quoted error bars, the central claim is contradicted.","supporting_citations":[],"review_version":1}