{"id":"c57586b9-f2e6-4a06-ad91-906eade25343","arxiv_id":"2412.00089","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A step-reference cross-correlation seeding method improves PCGe pulse rise-time fits, cutting surface leakage by about 70% and lowering the TEXONO analysis threshold by at least 10 eV_ee.","lead":"Cross-correlation pulse-shape matching with a step-function template and a low-pass filter is used to seed rise-time fits in point-contact germanium detectors, sharpening bulk versus surface event separation near threshold. The method cuts surface leakage by roughly 70 percent in pulser tests and suggests a 10 eV_ee lowering of the TEXONO analysis threshold, which matters for dark matter and neutrino searches.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Pulser validation uses a single surface time constant, so the ~70% leakage reduction in TEXONO data has no independent ground truth; the 10 eVee threshold gain rests on one low-statistics bin.","rationale":"The pulser validation is genuine evidence: with known truth, the cross-correlation method reduces false-fit peaks at 0.3 keVee by >70% and reduces rise-time scatter without large bias (Fig. 9). My concern is not that the method is useless, but that the paper's headline outcome—lowering the analysis threshold by ≥10 eVee—depends on a real-data claim that has no independent ground truth. The pulser sample is acknowledged to be narrower than data because a single time constant is used for surface events; this is exactly the regime where the method's behavior could differ, since real surface pulses have a spread of rise times and possibly different noise correlations. The TEXONO '70% reduction' is a difference between two algorithms on the same unlabeled events, so it cannot distinguish 'removed surface leakage' from 'moved bulk events to the surface side.' The threshold gain in Table I rests on the 0.19 keVee bin, whose rate difference from smoothed fit is not statistically significant (Δ≈67 cpkkd, combined uncertainty ≈136 cpkkd). The concrete test—using the ratio method on the same data to obtain an independent bulk/surface decomposition—would settle whether the removed events are actually surface events. This does not require new data and directly probes the load-bearing assumption. I partially agree with the reader: the pulser representativeness is the source of the problem, but the sharpest issue is the absence of independent labels in the real-data validation. Verdict: conditional—the method is plausible and the pulser evidence is positive, but the headline threshold claim should not be adopted until the real-data leakage reduction is independently confirmed.","tokens_in":15737,"tokens_out":6609,"duration_ms":61529,"concrete_test":"Apply the ratio method of Yang et al. (NIM A 886, 13 (2018)) [16] to the same 49 kg-days AC−⊗CR− TEXONO dataset to obtain an independent estimate of the surface fraction in the bulk-selection region (log10τ<0) at 0.18–0.30 keVee. Then compare the events removed by the cross-correlation method against this ratio-method surface classification; if the removed events are not predominantly surface-type, or if the true bulk efficiency drops by more than ~5%, the claimed ≥10 eVee threshold gain is unsupported. This check uses existing in situ data and directly tests whether the real-data leakage reduction is physical rather than an algorithmic bias.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim—a ≥10 eVee analysis-threshold gain (Sec. VI)—rests on the assumption that the ~70% reduction in the low-energy bulk spectrum in TEXONO data (Sec. V C, Fig. 13) is removal of surface leakage, not a systematic shift of genuine bulk events. This assumption is not directly tested. The pulser validation (Sec. V B) has ground truth, but it uses a single programmed surface time constant, producing a surface spectrum the authors acknowledge is 'much narrower than in data' (Sec. IV). Real surface events have a range of rise times and pulse shapes, so the 70% false-peak suppression measured on this simplified sample does not establish that the method behaves correctly on real surface pulses. In the real-data comparison there are no event labels: the reduction is only a difference between two algorithms applied to the same events. Table I shows the 0.19 keVee bin, the basis of the threshold claim, has rate 139±53 cpkkd versus 206±125 cpkkd for the smoothed fit; the difference is ~0.5σ combined, so even under the paper's own numbers the threshold gain is not statistically supported. If the real-data reduction is not validated against an independent bulk/surface decomposition, the headline achievement could be an artifact of the new method's behavior on low-SNR pulses.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a pulse-shape analysis method for p-type point-contact germanium detectors in which a cross-correlation shape-matching step and a low-pass filter are used to seed the four-parameter hyperbolic tangent rise-time fit. The method is developed on programmable pulser-generated bulk-like and surface-like pulses with known true time offsets and rise times, where it is reported to reduce the size of false-fit peaks by more than 70% at 0.3 keV_ee, improve the ROC curve for bulk-surface discrimination, and roughly halve the computation time. The method is then applied to TEXONO reactor data, where the authors report a nearly 70% reduction in the low-energy bulk spectrum and claim that the analysis threshold can be lowered by at least 10 eV_ee, from 200 to 190 eV_ee.","tokens_in":16047,"tokens_out":3831,"duration_ms":37176,"significance":"If the reported improvements are robust, the method would be a simple and portable enhancement to low-energy pulse-shape discrimination in pPCGe detectors, with direct relevance to dark-matter and coherent-elastic-neutrino-nucleus-scattering searches. The use of pulser pulses with known truth information is a genuine strength, as is the explicit ROC-curve comparison and the demonstration of consistency across two detector systems. The main limitation is that the headline quantitative claims—the 70% leakage reduction in real data and the 10 eV_ee threshold gain—rest on comparisons without independent ground truth and on a single low-statistics energy bin, respectively. The core algorithmic idea appears sound and testable, but the current manuscript overstates the strength of the evidence for those specific performance gains.","major_comments":[{"comment":"The claimed lowering of the analysis threshold by at least 10 eV_ee is not supported by the quoted numbers. The 0.19 keV_ee bin, which is the only new bin rendered 'analysable', has rates of 206 ± 125 cpkkd (smoothed fit) and 139 ± 53 cpkkd (cross-correlation). The difference of 67 cpkkd has a combined uncertainty of approximately 136 cpkkd, i.e., a significance of roughly 0.5σ. The paper should either present a quantitative threshold-setting procedure that accounts for this uncertainty or explicitly retract the 'at least 10 eV_ee' claim.","section":"Section V C, Table I"},{"comment":"The pulser surface sample is generated with a single programmed time constant, and the authors acknowledge in Section IV that the resulting surface spectrum 'is much narrower than in data'. The quantitative 70% false-peak suppression measured on this simplified sample (Figure 10) therefore does not by itself establish that the method will suppress surface leakage by the same amount for real surface events, which have a range of rise times and pulse shapes. The manuscript should either obtain pulser surface samples spanning the observed rise-time distribution, or present the real-data result as a qualitative trend rather than a quantitative transfer of the 70% figure.","section":"Section IV and Section V B"},{"comment":"In the TEXONO data comparison there is no independent ground truth: the 'reduction' is the difference between two algorithms applied to the same unlabeled events. The near-70% decrease in the low-energy bulk spectrum could in principle be caused by a systematic shift of genuine bulk events into the surface region or by a bias in the new t0 estimate at low signal-to-noise ratios, rather than by true removal of surface leakage. The authors should cross-check the result against an independent bulk-surface decomposition, for example the ratio method of reference [16], or demonstrate with injected pulser bulk pulses that genuine bulk events are not lost by the cross-correlation method.","section":"Section V C, Figure 13"},{"comment":"The choice of the time segments 3 to 5.5 microseconds after t0 and the 6 microsecond rise-time upper bound is justified only by the statement that this covers 99% of surface events, but no quantitative support or sensitivity study is provided. Because the same pulser sample is used both to motivate this choice and to validate the method, there is a mild circularity. The paper should discuss how the results depend on these ad hoc parameters, or at least provide a reference measurement for the claimed 99% coverage.","section":"Section III B"}],"minor_comments":[{"comment":"The discrete cross-correlation sum in Eq. (3) uses the notation S_{j+\\Delta t} without specifying how indices are handled at the array edges; please clarify the boundary treatment.","section":"Section III A, Eq. (3)"},{"comment":"The pulser samples were taken with an nPCGe detector at Academia Sinica while the TEXONO data are from a pPCGe detector at KSNL. The paper normalizes energies by pedestal RMS, but it should also discuss whether differences in detector geometry and electronics could affect the transferability of the pulse-shape conclusions.","section":"Section IV"},{"comment":"The ROC curve in Figure 11 is presented without confidence bands or a statement of the sample sizes used to construct it; adding these would make the improvement more quantitatively assessable.","section":"Section V B, Figure 11"},{"comment":"The uncertainties quoted in Table I appear to be much larger than Poisson counting statistics alone (e.g., 206 ± 125 cpkkd at 0.19 keV_ee). The text should specify the statistical and systematic components of these uncertainties and how they were propagated.","section":"Table I"},{"comment":"The phrase 'at least 10 eV_ee' should be softened to reflect the large uncertainty in the 0.19 keV_ee bin, e.g., 'a possible reduction of the analysis threshold by about 10 eV_ee', until a more precise analysis is available.","section":"Abstract and Section VI"},{"comment":"There are several typographical and wording issues: 'starts to convolve with each other' in the abstract, 'receiver-operation-characteristic' for 'receiver operating characteristic', 'hallow' for 'hollow' in the Figure 15 caption, and 'T est' in the Section V C heading.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The core method—cross-correlation seeding with a low-pass filter—seems sensible and the pulser-based validation with true parameters is a positive feature. However, the headline threshold gain is not statistically supported by Table I, and the real-data leakage reduction is not validated against any independent ground-truth decomposition. I would encourage the authors to reframe the paper as a technical demonstration of improved seeding and rise-time resolution, supported by the pulser studies, and to either strengthen or de-emphasize the threshold-lowering claim. The manuscript may be better suited to a specialized instrumentation journal than a general physics journal, but the scope decision is the editor's."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThe core methodological point is credible: cross-correlation seeding with a step-function reference, plus low-pass filtering for amplitude and pedestal estimates, improves hyperbolic tangent fits to PCGe pulses. The pulser validation is the strongest part. With programmed true time offsets and rise times, the method narrows the t0 distribution and suppresses false-fit peaks; the ROC curve at 0.3 keV_ee shows a real gain. Cutting the number of fits from four to two is a practical bonus.\n\nThe soft spots are in the real-data claims. The pulser surface sample uses a single time constant, which the authors admit is 'much narrower than in data.' That makes the quantitative 70% false-peak reduction a weak predictor for real surface events with a spread of rise times. On TEXONO data, the ~70% bulk-spectrum reduction has no independent ground truth; it is just two algorithms applied to the same events. The threshold-lowering claim (at least 10 eV_ee) rests on the 0.19 keV_ee bin, where the smoothed and cross-correlation rates agree within about half a sigma. The paper also leaves the low-pass filter cutoff and correlation window unspecified, which hurts reproducibility.\n\nNone of this kills the idea. Better t0 seeding is real and likely useful. But the headline real-data results need either independent validation—calibration data with known bulk/surface composition, or a pulser model that draws surface rise times from a realistic distribution—or they need to be cut back to what the statistics actually support.\n\nWho should read this: anyone working on pPCGe pulse-shape analysis, especially in TEXONO, CDEX, or CONUS circles. It deserves a serious referee, but with major revisions. I would ask the authors to present the pulser results as the primary quantitative evidence, treat the TEXONO comparison as qualitative, and release code and parameters so the ROC curve can be reproduced.\n\nRecommendation: engage with it, but push for the real-data claims to be either strengthened or substantially softened.","headline":"Solid pulser-validated methods paper whose real-data threshold claim outruns its statistics—worth reviewing, needs major revisions.","tokens_in":16546,"tokens_out":4501,"would_cite":false,"duration_ms":40874,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["29.40.Wk"],"model":"deepseek-v4-flash","headline":"A cross-correlation and low-pass filter seeding step for pulse rise-time fits reduces surface-event leakage by roughly 70% at near-threshold energies and lowers the TEXONO analysis threshold by at least 10 eV_ee.","keywords":["point-contact germanium detectors","pulse-shape discrimination","rise-time analysis","cross-correlation","low-pass filter","surface event leakage","TEXONO","analysis threshold"],"falsifier":"Compare the cross-correlation method and the smoothed-fit method on calibration-source data (for example 241Am or 137Cs) in the 0.18–0.4 keV_ee range, where the true bulk and surface spectral shapes are known independently; if the measured leakage reduction is substantially smaller than 70%, the pulser's single-time-constant surface sample overstates the real-world gain.","tokens_in":15536,"feed_emoji":"🔬","tokens_out":7918,"duration_ms":60729,"temperature":0.7,"pith_summary":"Point-contact germanium detectors are used in searches for dark matter and neutrinos because they can see sub-keV energy deposits, but pulses originating near the detector's passive surface are slower and are treated as background. At energies near the analysis threshold the rise-time spectra of bulk and surface events overlap, so misclassified surface events leak into the signal spectrum. This paper proposes seeding the pulse-rise-time fit with parameters obtained by cross-correlating the pulse against a step-function reference and low-pass filtering the pulse to estimate its amplitude and baseline. On pulser-generated pulses the new seeding reduces the false-fit peaks in both bulk and surface rise-time spectra by more than 70%, and on TEXONO reactor data it reduces the near-threshold surface leakage into the bulk spectrum by nearly 70%, allowing the analysis threshold to be lowered by at least 10 eV_ee, from 200 to 190 eV_ee, while cutting computation time roughly in half.","feed_headline":"Cross-correlation fit lowers germanium detector threshold by 10 eV","feed_subtitle":"Surface-event leakage drops about 70% and analysis runs nearly twice as fast.","key_machinery":"The key machinery is a two-step seeding procedure for the four-parameter hyperbolic-tangent fit $A/2\\tanh(s(t-t_0))+C$. First, the pulse is cross-correlated with a step-function reference that rises from $-n$ to $n$ (a hyperbolic tangent of infinite time constant), whose peak position yields the time offset $t_0$; the step is chosen because its correlation peak is not widened by the reference's own rise time. Second, a low-pass filter smooths the pulse so that amplitude $A$ and pedestal $C$ can be estimated from the asymptotic end segments. The resulting seeds let the nonlinear fit converge faster and more accurately, reducing rise-time fitting errors at low signal-to-noise ratios, where the smoothed-fit method frequently lands on false secondary peaks.","core_discovery":"The central claim is that the quality of the hyperbolic-tangent rise-time fit in pPCGe detectors is controlled by the initial parameter estimates ('seeding'), and that replacing the Savitzky–Golay smoothed-fit seeding with cross-correlation shape-matching plus low-pass filtering sharpens rise-time resolution at near-threshold energies. The paper argues that a step function is the optimal reference shape for cross-correlation because it has no finite rise time of its own, so the correlation peak width reflects only the signal; the correlation maximum gives the pulse time offset $t_0$, and a low-pass filter reduces noise when estimating amplitude $A$ and pedestal $C$. In pulser-generated samples at 0.3 keV_ee, the false-fit peaks in both bulk and surface rise-time spectra shrink by more than 70%, and the primary peaks become more concentrated. In 49 kg-days of TEXONO reactor data, the near-threshold (0.18–0.3 keV_ee) bulk spectrum is reduced by nearly 70% relative to the smoothed-fit method, consistent with suppression of surface leakage, and the full-spectrum analysis shows the lowest analysable bin shifting from 0.20 to 0.19 keV_ee, i.e., a threshold gain of at least 10 eV_ee. The improved seeding also lets the fit be run twice instead of four times, a ~50% computation-time saving.","pith_inferences":["Because the improvement comes purely from better initial conditions for an existing nonlinear fit, the same cross-correlation seeding idea could benefit other experiments that fit sigmoidal or template pulse shapes at low signal-to-noise ratio.","The 10 eV_ee threshold gain is a software-only improvement, so it should compound with hardware noise reduction rather than substitute for it; the paper itself notes hardware advances are still needed.","The quantitative leakage-reduction claim rests on pulser-generated surface pulses that use a single time constant, while real surface events span a wider rise-time range; a direct test with calibration sources would show whether the 70% figure survives in situ.","Since the bulk-selection curve in the full analysis is optimised on pulser-generated bulk pulses, applying the method to other detectors will require re-optimisation with their own pulser settings."],"forward_implications":["The analysis threshold for TEXONO pPCGe data can be lowered from 200 eV_ee to at least 190 eV_ee, extending the reach of low-energy rare-event searches.","Surface-event leakage into the bulk rise-time spectrum is suppressed by roughly 70% at near-threshold energies (0.3 keV_ee in pulser samples; 0.18–0.3 keV_ee in TEXONO data).","Rise-time analysis computation time drops by about 50% because the number of fits per pulse is reduced from four to two.","The method is robust across detectors: improvements were seen on both the TEXONO pPCGe detector and a local nPCGe test detector."],"supporting_citations":[{"why":"Establishes the smoothed-fit seeding method and the spectral-shape approach to bulk-surface discrimination that this paper improves on.","marker":"[11]"},{"why":"Introduces the ratio method for in-situ calibration of bulk vs. surface spectra, which the new rise-time resolution aims to enhance.","marker":"[16]"},{"why":"Motivates the use of programmable pulser-generated pulse samples to evaluate pulse-shape methods with known truth values.","marker":"[29]"},{"why":"Describes the TEXONO experiment and detectors whose reactor data are used for the in-situ validation.","marker":"[30]"},{"why":"Supplies the noise-edge cleaning cuts and data-selection methods used in the full energy-spectrum analysis.","marker":"[3]"},{"why":"Provides the noise-edge cleaning and threshold-setting methodology applied to the final spectrum.","marker":"[31]"},{"why":"Defines the smoothing filter used in the baseline smoothed-fit method against which the cross-correlation method is compared.","marker":"[20]"}],"fun_headline_variants":["Cross-correlation cuts germanium surface noise by ~70%","Faster fit sharpens germanium detector threshold by 10 eV","Shape-matching pulse fit quickens analysis, cuts leakage","Rise-time fit upgrade halves compute time, lowers threshold","Germanium detector gains 10 eV reach with smarter fitting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the pulser-generated surface events, which use one programmed time constant and so form a much narrower rise-time distribution than real surface events, faithfully represent the behaviour of true near-threshold surface events well enough that the roughly 70% leakage suppression measured on these simplified samples transfers to TEXONO reactor data.","fun_headline_variants_meta":{"raw":{"variants":["Cross-correlation cuts germanium surface noise by ~70%","Faster fit sharpens germanium detector threshold by 10 eV","Shape-matching pulse fit quickens analysis, cuts leakage","Rise-time fit upgrade halves compute time, lowers threshold","Germanium detector gains 10 eV reach with smarter fitting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000678,"raw_usage":{"total_tokens":3174,"prompt_tokens":1132,"completion_tokens":2042,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":748,"completion_tokens_details":{"reasoning_tokens":1958}},"tokens_in":748,"tokens_out":2042,"duration_ms":14884,"temperature":1.0,"reasoning_tokens":1958,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:26:17.023809+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the cross-correlation method and the smoothed-fit method on calibration-source data (for example 241Am or 137Cs) in the 0.18–0.4 keV_ee range, where the true bulk and surface spectral shapes are known independently; if the measured leakage reduction is substantially smaller than 70%, the pulser's single-time-constant surface sample overstates the real-world gain.","supporting_citations":[{"cited_title":"Soma et al., Characterization and performance of ger- manium detectors with sub-kev sensitivities for neutrino and dark matter experiments, Nucl","cited_arxiv_id":null,"evidence_quote":"Establishes the smoothed-fit seeding method and the spectral-shape approach to bulk-surface discrimination that this paper improves on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the ratio method for in-situ calibration of bulk vs. surface spectra, which the new rise-time resolution aims to enhance."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates the use of programmable pulser-generated pulse samples to evaluate pulse-shape methods with known truth values."},{"cited_title":"Malavolta et al., Atmospheric stellar parameters from cross-correlation functions, Mon","cited_arxiv_id":null,"evidence_quote":"Describes the TEXONO experiment and detectors whose reactor data are used for the in-situ validation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the noise-edge cleaning cuts and data-selection methods used in the full energy-spectrum analysis."},{"cited_title":"Blake et al","cited_arxiv_id":null,"evidence_quote":"Provides the noise-edge cleaning and threshold-setting methodology applied to the final spectrum."},{"cited_title":"Yue et al","cited_arxiv_id":null,"evidence_quote":"Defines the smoothing filter used in the baseline smoothed-fit method against which the cross-correlation method is compared."}],"review_version":1}