{"id":"d54adb7e-6846-48a7-9b02-0d79b972cee4","arxiv_id":"2411.10216","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"A Monte Carlo study finds that a silicon direct-detection experiment needs between 0.0013 and 1.9 kg-year of exposure to statistically link a hypothetical LDMX signal to dark matter, for masses between 4 and 25 MeV.","lead":"This paper proposes a four-step analysis to test whether a future signal at the LDMX fixed-target experiment really comes from dark matter. Using simulations of both LDMX and a silicon direct-detection experiment, it estimates how much exposure the direct detector needs, from about 0.001 kg-year for 4 MeV dark matter to 1.9 kg-year for 25 MeV dark matter.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Threshold exposures rely on an uncalibrated chi-square: Eq. (4.1) is evaluated with LDMX fixed at its mean and DD predictions carrying posterior uncertainty not in the denominator, so the 95% C.L. interpretation of xi_th may be optimistic.","rationale":"The reader's weakest assumption (backgrounds) is a real limitation, but the paper presents a proof-of-principle with simulated signals, so the purely background-free treatment is an acknowledged idealisation rather than an internal inconsistency. The more load-bearing issue is statistical: the threshold exposures are derived from a test statistic whose null distribution is assumed to be chi-square with 2(nb-1) degrees of freedom, yet the way Eq. (4.1) is evaluated in Eq. (4.4) violates the conditions for that distribution. Specifically, the LDMX side is fixed at its expectation, while the DD side is a posterior prediction with its own variance that is omitted from the denominator. This makes the fitted chi^2(xi) curve artificially low at large exposure, so the intersections with chi^2_* (which define xi_th) occur at smaller exposures than a properly calibrated test would give. The central claim--that exposures of 0.0013-1.9 kg-year suffice--therefore rests on a calibration that has not been demonstrated. The concern is addressable by a Monte Carlo recalibration, and the overall four-step strategy may survive, so the verdict remains CONDITIONAL. I mark agreement as 'partial' because the background issue is related (both are about idealised uncertainties), but the calibration problem is distinct and more directly tied to the headline numbers.","tokens_in":16283,"tokens_out":8649,"duration_ms":89731,"concrete_test":"For each benchmark (m_chi, epsilon*) in Fig. 4, run a Monte Carlo calibration: generate many realizations of LDMX and DD data under the null (both Poisson-sampled from the same true histograms, with DD predictions drawn from the full posterior predictive distribution). Compute the statistic in Eq. (4.1) on each realization for exposures spanning the reported xi_th. Build the empirical null distribution and measure the false-rejection rate at the reported xi_th. If the rate differs substantially from 5%, or if the exposure at which the median chi^2 equals the empirical 95th percentile shifts by more than a factor of two, the headline thresholds require recalibration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative result (Fig. 4 and abstract: xi_th from 0.0013 to 1.9 kg-year) is obtained by solving Eq. (4.3) with the test statistic of Eq. (4.1) evaluated as in Eq. (4.4). The null hypothesis in Eq. (4.2) states that LDMX and DD data are independent Poisson samples from the same mean. In the actual evaluation, however, the 'observed' LDMX histograms are fixed at their true expectations (no Poisson fluctuations), and the DD 'predicted' histograms are posterior means from Eq. (3.14), whose variance (Eq. 3.16) is not included in the denominator of Eq. (4.1). The denominator N_LDMX,i + N_DD,i is appropriate only for two independent Poisson counts; it is not the variance of N_LDMX,i - N_DD,i under the procedure actually used. The authors acknowledge part of this in footnote 6: neglecting LDMX fluctuations makes chi^2 approach a smaller value than the true statistic. Because the fitted A + B/xi curve then crosses chi^2_* at a smaller exposure, the reported threshold values are likely underestimated. The chi-square distribution assumption in Sec. 4 is therefore not satisfied by the computed statistic, and the claimed 95% C.L. meaning of xi_th is not established. This threatens the headline exposure values directly, independent of the background modelling question.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a four-step strategy to validate the dark matter interpretation of a future LDMX excess by combining LDMX data with next-generation silicon direct-detection (DD) data. The authors simulate an LDMX signal with MadGraph, simulate DD data with DarkELF for increasing exposures, perform Bayesian inference with MultiNest, use the resulting posterior to predict the LDMX electron recoil energy and transverse momentum distributions, and compare predictions with the simulated LDMX signal via a chi-square test. The central quantitative result is a threshold DD exposure, ξth, above which the test would not reject a common DM origin; the authors report ξth growing from 0.0013 kg-year at mχ = 4 MeV to 1.9 kg-year at mχ = 25 MeV. The paper also studies a cross-check in which the LDMX and DD data are generated from different DM masses, and shows that the test can often reject the wrong mass.","tokens_in":16705,"tokens_out":10203,"duration_ms":99482,"significance":"The proposed four-step strategy is a sensible and valuable proof-of-concept for cross-experiment validation of a sub-GeV DM signal, and the manuscript is clearly written. It uses standard public tools (MadGraph, DarkELF, MultiNest) and gives explicit equations for the likelihood, posterior, and test statistic, which aids reproducibility. The mass-mismatch study in Sec. 5 (Figs. 5-7) is a useful sanity check. However, the headline threshold exposures are not yet firmly established: the test statistic is not evaluated under the null distribution that is assumed, and the idealized no-background setup is not reflected in the unqualified abstract numbers. The paper is therefore best viewed as a methodology proposal whose quantitative outcome requires further calibration.","major_comments":[{"comment":"The test statistic as evaluated does not have the claimed chi-square distribution under the null. Under the null in Eq. (4.2), if both datasets were independent Poisson counts with equal means, the denominator in Eq. (4.1) would be the variance of the difference; but in the actual evaluation (Eq. (4.4)), N_LDMX,i is fixed at the true mean and N_DD,i is a posterior mean, whose variance is not included in the denominator. Footnote 6 acknowledges that fixing the LDMX fluctuations makes chi^2 smaller than the true statistic, but the same issue applies to the DD posterior uncertainty, and no Monte Carlo calibration of chi^2_* under the null is provided. Since ξth is obtained by solving Eq. (4.3) with this uncalibrated chi^2_*, the quoted threshold exposures are likely underestimated, and their stated 95% C.L. meaning is not established. I recommend calibrating the threshold by simulating the full pipeline under the null, or using a statistic whose null distribution is known (for example, including Poisson fluctuations in both datasets and using the variance of N_LDMX,i - N_DD,i in the denominator).","section":"Section 4, Eqs. (4.1)-(4.5)"},{"comment":"The simulated LDMX 'signal' and the DD data are generated as pure DM signals with no backgrounds, even though the LDMX signal is defined in Sec. 3.1 as an excess over inclusive single-electron backgrounds and the DD analysis uses only DM-induced electron-hole pairs. The abstract and conclusion report threshold exposures without this caveat. Real backgrounds at LDMX or in a silicon detector would change the effective counts entering Eq. (4.1) and could materially shift ξth. The authors should either add a background-injection study for at least one benchmark, or clearly state in the abstract and conclusions that the quoted thresholds apply to a background-free, proof-of-principle scenario.","section":"Secs. 3.1 and 3.2"},{"comment":"The normalization of the LDMX signal is ambiguous. Eq. (3.4) defines NMG(mχ,ε) = ηLσee(mχ,ε), which for fixed luminosity and efficiency is the expected number of events; however, the caption of Fig. 1 states 'Simulations are performed with MadGraph assuming NMG = 10^4 events per simulation.' If NMG is fixed to 10^4 rather than computed from the luminosity and cross section, the LDMX histograms are not the expected event counts for the stated beam conditions, and the comparison between N_LDMX,i and N_DD,i in Eq. (4.1) is not on an absolute scale. The authors should clarify whether 10^4 is a Monte Carlo generation size used only for shape, and if so, explain how the rescaling in Eq. (3.6) restores the physical normalization. This is load-bearing for the absolute values of ξth.","section":"Eq. (3.4) and Fig. 1 caption"},{"comment":"The predicted LDMX histograms are approximated by evaluating N_i,MG at the posterior means mχ and ε, rather than by computing the actual posterior expectation ∫ dmχ dε f_DD N_i,MG. This 'plug-in' approximation is not generally valid for a nonlinear function, and it could bias the predicted histograms and hence χ2. The error estimate in Eq. (3.16) accounts for the spread of the posterior, but it does not correct a biased mean. The authors should justify this approximation for the ranges of mχ and ε considered, or test it by comparing the plug-in prediction with the full integral for at least one benchmark.","section":"Eq. (3.14)"}],"minor_comments":[{"comment":"The header abstract and the abstract in the full text give different numbers for the threshold exposure: the former quotes 0.012 kg-year at 4 MeV and 1 kg-year at 25 MeV, while the latter quotes 0.0013 kg-year and 1.9 kg-year. Please check and harmonize the numbers.","section":"Header abstract vs. full-text abstract"},{"comment":"The threshold exposure ξth is extracted from a fit A + B/ξ to the mean χ2(ξ) over 10 realizations, but no uncertainty on ξth is provided. Given the large scatter shown in Fig. 4, the statistical error on ξth could be substantial; please report an uncertainty or a range.","section":"Fig. 4 and Sec. 5"},{"comment":"The test statistic adds contributions from the recoil energy and transverse momentum distributions even though these are expected to be correlated with each other; the text acknowledges this but does not discuss the impact on the null distribution. A sentence explaining the expected effect, or a calibration check, would be helpful.","section":"Sec. 4, Eq. (4.1)"},{"comment":"The text says χ2 is calculated for 10 different realisations of the DD data, but Eq. (4.4) uses posterior means; please clarify exactly which quantities vary between realisations and how the 10 values are combined.","section":"Sec. 5, Fig. 3"},{"comment":"The phrase 'statistically dependent' is used to mean 'compatible with a common DM origin,' but a chi-square test can only reject or fail to reject the null. Consider using more standard terminology such as 'statistically compatible' (which the paper also uses) for clarity.","section":"Throughout"},{"comment":"In Eq. (3.11), the symbol ω is used both for the deposited energy and for the average energy per electron-hole pair (3.6 eV); consider using a distinct symbol such as ω_pair to avoid confusion.","section":"Eq. (3.11)"}],"recommendation":"major_revision","confidential_remarks":"The main unresolved issue is the LDMX normalization ambiguity. If the authors confirm that NMG is fixed to 10^4 rather than the luminosity-expected event count, the absolute threshold exposures would need to be recomputed, which could change the paper's conclusion. Please ask the authors to clarify this prominently during revision. The paper is otherwise within JCAP scope, and I do not see a circularity problem: the comparison is a genuine cross-check between independently simulated datasets."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper proposes a four-step strategy to validate the DM origin of a future LDMX signal using direct detection data, and quantifies the required exposure. The strategy itself is the main contribution: earlier WIMP complementarity work is extended to sub-GeV DM with electron scattering, and the exposure-threshold curve as a function of mass is new. The numerical work is transparent: MadGraph for LDMX, DarkELF for silicon DD, MultiNest for the posterior, and a chi-square comparison. They also test a negative case (different masses at the two experiments) to show the test can reject the wrong model. That is good practice.\n\nThe main soft spot is the statistical calibration of the chi-square test. The test statistic in Eq. (4.1) is treated as chi-square distributed, but the way it is evaluated in Eq. (4.4) does not satisfy the assumptions. The LDMX histograms are fixed at their true expectations (no Poisson fluctuations), and the DD predicted histograms are posterior means whose variance is not included in the denominator. The denominator N_LDMX + N_DD is appropriate only for two independent Poisson counts. As a result, the computed chi-square is smaller than the true statistic, so the threshold exposures are likely underestimated. The authors acknowledge the LDMX side in footnote 6 but still use the chi-square critical value. This is not fatal to the strategy, but it means the 95% C.L. label is not established. The fix is straightforward: calibrate the test statistic under the null via Monte Carlo, or use a proper likelihood ratio.\n\nTwo other issues: no backgrounds are simulated for either experiment, so the threshold exposures are for a signal-only idealization. And the abstract in the submitted version gives different numbers (0.012 and 1 kg-year) than the body and Fig. 4 (0.0013 and 1.9 kg-year). That needs fixing. No code or data are released, which makes reproduction harder.\n\nOverall, the idea is useful and the paper is worth a serious referee. The strategy is novel in this mass range and the quantitative result, though not yet calibrated, is a reasonable first estimate. I would engage with it in review and ask for a Monte Carlo calibration of the test statistic before accepting.\n\nBest","headline":"A sensible strategy for validating the DM origin of an LDMX signal, but the headline threshold exposures rest on a chi-square test that is not calibrated as claimed.","tokens_in":17203,"tokens_out":4551,"would_cite":true,"duration_ms":40372,"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 silicon-based direct detection experiment with 0.0013–1.9 kg-year exposure can statistically validate the dark matter origin of a future LDMX signal.","keywords":["dark matter","sub-GeV dark matter","LDMX","fixed-target experiment","direct detection","dark photon","silicon detector","Bayesian inference"],"falsifier":"Repeat the same four-step pipeline with simulated background events included at realistic levels, such as the inclusive single-electron background at LDMX and environmental or radioactive counts in a silicon detector. If the threshold exposure for the $m_\\chi=10\\,\\mathrm{MeV}$ benchmark moves substantially from the quoted $\\sim0.03\\,\\mathrm{kg\\,year}$, or if $\\chi^2(\\xi)$ no longer crosses the 95% C.L. line within reachable exposures, the background-free estimate is falsified.","tokens_in":16086,"feed_emoji":"⚛️","tokens_out":8826,"duration_ms":75855,"temperature":0.7,"pith_summary":"This paper argues that a future positive signal at a fixed-target experiment such as LDMX cannot by itself establish that the new particle is dark matter; an invisible particle carrying beam energy could have a non-cosmological origin. To close that gap, the authors propose a four-step strategy: record the LDMX electron recoil-energy and transverse-momentum excess, operate a silicon direct-detection experiment with increasing exposure, extract a Bayesian posterior for the dark matter mass and coupling from the direct-detection data, and compare the resulting predicted LDMX distributions with the recorded ones in a chi-square test. Simulating complex scalar dark matter with a dark photon mediator, they find that the exposure needed to avoid wrongly rejecting a common dark matter origin grows with mass, from $0.0013\\,\\mathrm{kg\\,year}$ at $m_\\chi=4\\,\\mathrm{MeV}$ to $1.9\\,\\mathrm{kg\\,year}$ at $m_\\chi=25\\,\\mathrm{MeV}$. These exposures are within reach of current or near-future silicon detectors, so the strategy offers a concrete cross-check between fixed-target and direct searches.","feed_headline":"0.0013–1.9 kg-years of exposure can pin an LDMX excess as dark matter","feed_subtitle":"A silicon detector needs only that much exposure to cross-check a fixed-target dark matter hint.","key_machinery":"The machinery is a four-step analysis chain built on three components. First, Monte Carlo simulation of dark photon production in electron-tungsten scattering, $e^-W\\to e^-W A'\\to e^-W\\chi\\chi$, with $m_{A'}=3m_\\chi$ and couplings on the relic-density line, produces the recorded LDMX signal in 30 bins of electron recoil energy and transverse momentum. Second, direct-detection data are simulated from the DM-electron scattering rate in silicon, expressed through the dielectric function and converted into electron-hole pair counts, and a Bayesian posterior for $(m_\\chi,\\sigma_e)$ is extracted from those counts. Third, the posterior is projected into predictions for the LDMX histograms, and the test statistic $\\chi^2_{\\rm TS}$ of Eq.~(4.1) compares predicted and recorded histograms; the threshold exposure $\\xi_{\\rm th}$ solves $\\chi^2(\\xi)=\\chi^2_*$ at 95% C.L. The statistical engine is the chi-square test, while the physical engine is the complementary information carried by the $E_e$ and $P_T$ distributions.","core_discovery":"The central discovery is a quantitative threshold for validation: a silicon-based direct-detection experiment operating with exposure $\\xi$ can assert or reject the dark matter origin of a hypothetical LDMX signal. The authors compute $\\chi^2(\\xi)$ from a test statistic comparing recorded and predicted $E_e$ and $P_T$ histograms, and define $\\xi_{\\rm th}$ as the exposure at which $\\chi^2(\\xi)$ crosses the 95% C.L. threshold $\\chi^2_*$ under the null hypothesis of a common dark matter origin. They find that $\\xi_{\\rm th}$ increases with the dark matter mass, from $0.0013\\,\\mathrm{kg\\,year}$ at $4\\,\\mathrm{MeV}$ to $1.9\\,\\mathrm{kg\\,year}$ at $25\\,\\mathrm{MeV}$; at $10\\,\\mathrm{MeV}$ it is about $0.03\\,\\mathrm{kg\\,year}$. They also show that when the LDMX and direct-detection signals are generated by different dark matter masses, the test rejects a common origin at 95% C.L. for essentially all exposures, provided the mass difference is large.","pith_inferences":["Editorial inference: if real backgrounds are added, the quoted threshold exposures are optimistic floors; the same pipeline would likely yield larger $\\xi_{\\rm th}$, particularly at low mass where the LDMX signal is small.","Editorial inference: the four-step posterior-predictive comparison is model-agnostic in structure, so it could be repurposed for other sub-GeV benchmark models or for comparing other fixed-target and direct-detection pairs.","Editorial inference: because the paper uses posterior means to build predictions, its stated insensitivity to astrophysical halo uncertainties may be fragile; a fully marginalized check over halo parameters would be a direct stress test of that claim."],"forward_implications":["A future LDMX excess can be checked with exposures already in reach: $0.0013\\,\\mathrm{kg\\,year}$ at $4\\,\\mathrm{MeV}$, roughly $0.03\\,\\mathrm{kg\\,year}$ at $10\\,\\mathrm{MeV}$, and $1.9\\,\\mathrm{kg\\,year}$ at $25\\,\\mathrm{MeV}$.","If the two signals share a dark matter origin, exposures below the threshold would lead one to wrongly reject that origin; the paper therefore defines a practical minimum exposure for a validation campaign.","The electron recoil energy and transverse momentum carry complementary information: $E_e$ contributes more when the dark matter masses are equal, while $P_T$ dominates when the underlying masses differ.","When the LDMX signal is produced by a different dark matter mass than the direct-detection signal, the chi-square test rejects a common origin at 95% C.L. for essentially all exposures if the mass difference is large, and for finite exposures if the masses are close."],"supporting_citations":[{"why":"Defines the LDMX setup (beam energy, tungsten target, luminosity) and supplies the benchmark relation $m_{A'}=3m_\\chi$ and relic-density couplings used to simulate the signal.","marker":"[8]"},{"why":"Supplies the Monte Carlo event generation used to simulate $e^-W\\to e^-W A'\\to e^-W\\chi\\chi$ at LDMX.","marker":"[18]"},{"why":"Provides the tungsten nuclear form factor and fixed-target dark gauge force framework used in the event generation.","marker":"[20]"},{"why":"Provides the dielectric function and DM-electron scattering rates in silicon used to simulate the direct-detection data.","marker":"[21]"},{"why":"Gives the electron-hole pair production relation that converts deposited energy into the observable event counts in silicon.","marker":"[24]"},{"why":"Supplies the nested sampling algorithm used to extract posterior pdfs for $(m_\\chi,\\sigma_e)$ from the direct-detection data.","marker":"[25]"}],"fun_headline_variants":["0.0013–1.9 kg-years pin LDMX excess as dark matter","Direct detection confirms LDMX signal as dark matter","Sub-kg exposures validate LDMX dark matter origin","Few kg-years expose LDMX signal's dark matter truth","LDMX dark matter hint testable with small exposures"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quantitative threshold exposures assume that both the LDMX excess and the direct-detection data consist purely of dark-matter events, with no ordinary background counts; any real background will change the required exposure.","fun_headline_variants_meta":{"raw":{"variants":["0.0013–1.9 kg-years pin LDMX excess as dark matter","Direct detection confirms LDMX signal as dark matter","Sub-kg exposures validate LDMX dark matter origin","Few kg-years expose LDMX signal's dark matter truth","LDMX dark matter hint testable with small exposures"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000697,"raw_usage":{"total_tokens":3240,"prompt_tokens":1124,"completion_tokens":2116,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":740,"completion_tokens_details":{"reasoning_tokens":2028}},"tokens_in":740,"tokens_out":2116,"duration_ms":14868,"temperature":1.0,"reasoning_tokens":2028,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:49:58.369118+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the same four-step pipeline with simulated background events included at realistic levels, such as the inclusive single-electron background at LDMX and environmental or radioactive counts in a silicon detector. If the threshold exposure for the $m_\\chi=10\\,\\mathrm{MeV}$ benchmark moves substantially from the quoted $\\sim0.03\\,\\mathrm{kg\\,year}$, or if $\\chi^2(\\xi)$ no longer crosses the 95% C.L. line within reachable exposures, the background-free estimate is falsified.","supporting_citations":[],"review_version":1}