{"id":"d4a7c644-9bf3-4374-8e14-247577fc4def","arxiv_id":"2412.13131","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"JWST NIRSpec dark calibration images exclude previously allowed high-cross-section parameter space for sub-GeV dark matter coupled to an ultralight dark photon, for subcomponent fractions as low as about 0.01%.","lead":"Using dark calibration images from JWST's NIRSpec detector, this paper places new limits on dark matter particles that interact so strongly with ordinary matter that ground-based detectors cannot see them. The limits close a previously open region of parameter space for a specific dark matter model, and show that existing space telescope data can be repurposed for dark matter searches.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Dataset selection by background-only fit quality could bias the exclusion; the 16 discarded datasets may contain the DM signal the paper claims to exclude, so a signal-injection coverage test is needed.","rationale":"The reader's weakest_assumption identifies the background model and the discarding of 16 datasets; I agree this is the most load-bearing point. The paper presents a novel use of JWST dark images and makes the data public, which is a real strength, but the statistical procedure has a selection step that is not validated. A simulation-based coverage test is the natural check: if the analysis pipeline, when applied to mock data with a known injected DM signal, excludes that signal more often than the nominal 5%, then the reported exclusion cannot be taken at face value. The form-factor uncertainty is also real but is acknowledged by the authors and is conservative in the direction of weakening the limit if the true rate is lower; the dataset-selection issue is not conservative in any obvious direction. For these reasons I would keep the reader's conditional verdict unchanged, with the condition that the authors run the proposed injection test (or an equivalent closure test) and include all datasets with a more flexible background model in a robustness check.","tokens_in":20024,"tokens_out":10028,"duration_ms":103669,"concrete_test":"For each of the 30 real datasets, use the best-fit background parameters (λDC, σ1, σ2, r, h, Ntot) from the background-only fit as the true background. Generate 1000 mock realizations per dataset by drawing Poisson dark current per pixel, adding two-Gaussian readout noise, and applying the identical mask chain. Inject a DM signal with a known (mχ, σe) at the claimed 95% CL boundary (e.g., mχ = 100 MeV, σe ≈ 3×10^-23 cm², fχ = 0.4%) into each mock before the fit. Run the same χ²_fit selection (≤121) and profile-likelihood limit. If the injected σe is excluded in more than ~5% of realizations, the dataset selection or background model biases the exclusion. Repeat for several masses spanning 1 MeV-10 GeV.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central exclusion claim depends on the assumption that the 14 of 30 datasets with χ²_fit ≤ 121 are the only usable exposure and that the 16 rejected datasets fail because of unmodeled noise, not because they contain DM (Sec. IV and SM §VII). This is a selection on the same data used to set the limit. A real DM signal would skew the pixel charge distribution toward higher Ne and thereby worsen a background-only fit, so the datasets most likely to contain DM are exactly those that would be discarded. The surviving 14 datasets could then underrepresent the signal, and the profile-likelihood limit could exclude cross sections that are actually present. The paper reports no simulation or closure test showing that this selection preserves DM events. The concern is amplified because the background model is incomplete for 53% of the datasets, and because the DM rate in the open region is comparable to the fitted dark current (λDC ~ 6 e- vs ~1-4 DM e-/pixel/exposure in Fig. 1), so even small unmodeled components in the 14 'good' datasets could mimic or mask the signal.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript uses 30 JWST NIRSpec dark calibration images, applies custom masks to reject high-energy background events, and fits the pixel charge distribution with a background model of Poisson dark current smeared by two Gaussians. Fourteen datasets pass a background-only fit-quality cut, and a profile-likelihood analysis is used to set 95% CL upper limits on the DM-electron cross section for a fermion coupled to an ultralight dark photon. The authors report that for fχ=0.4% the previously unconstrained high-cross-section region between about 1 MeV and 10 GeV is disfavored, with reach to fχ~0.01% for some masses.","tokens_in":20314,"tokens_out":8616,"duration_ms":84924,"significance":"If the result is correct, it is a novel and important constraint on strongly interacting sub-GeV dark matter, using archival JWST calibration data and closing a window that ground-based detectors cannot reach because of atmospheric and crustal stopping. The paper is commendably transparent in several respects: the datasets are provided via a DOI, all 14 accepted background fits are shown, the statistical procedure is standard, and several choices are deliberately conservative (lower of the HgTe/CdTe rates, a factor-two reduction for an assumed impenetrable Mo shield). The central concern is that the exclusion is conditioned on a data-dependent selection step whose effect on the DM signal has not been tested; this must be addressed before the constraint can be taken at face value.","major_comments":[{"comment":"The selection of the 14 'good' datasets is made by fitting the background-only model to the same data that are then used to set the limit. A real DM signal would shift the pixel charge distribution toward larger Ne and, for fixed background, would tend to increase the background-only χ²_fit; the 16 rejected datasets are therefore not statistically independent of the DM hypothesis. The paper states in Section IV that the poor fits are 'likely due to an inadequate modeling of all noise sources,' but this is an assumption: no signal-injection or closure test is reported that shows that datasets containing a DM signal at the claimed 95% CL cross section would still pass the χ²_fit ≤ 121 cut and that the profile-likelihood procedure recovers the injected rate. Without such a test, the reported exclusion could be biased, and the 16 rejected datasets could contain the events the paper claims to exclude. In addition, the sentence 'We show the charge distributions and the background fit for the other 13 datasets in the SM' is numerically wrong (30 - 14 = 16), and the SM does not show the rejected fits, so the failure mode of the discarded datasets is not documented.","section":"Section IV, SM §VII"},{"comment":"The DM signal model uses QEDark form factors for HgTe and CdTe, with the lower rate chosen per charge yield, while the exact HgCdTe form factor is deferred to a future publication [84]. Because the profile likelihood compares the shape of the predicted charge distribution with the data, not only the total rate, the Q-dependence of the form factor matters. The claim that taking the lower of the HgTe and CdTe rates is conservative needs justification for the alloy Hg0.7Cd0.3Te, whose rate need not be bounded by the endpoint-crystal rates. Please either provide the exact calculation or show that plausible form-factor variations move the 95% CL contour by less than the uncertainties in the shielding and background selection.","section":"Section II, Eq. (1)"},{"comment":"The shielding model is essential for the high-cross-section reach, but it is based on a simplified geometry (20 mm SiC on most sides, 12 mm Mo treated as opaque, and an ad hoc factor-of-two flux reduction) and on a private communication [89]. No uncertainty on the shield thickness or composition is propagated into the limits. Since a thicker or denser shield would reduce the DM flux at the detector and weaken the exclusion, the paper should include a sensitivity study (e.g., varying the SiC thickness and the treatment of the Mo shield) and show the resulting change in the σe upper limit. This is particularly important because the cross-section boundary of the excluded region is set by the shield opacity.","section":"Section II, SM §I"}],"minor_comments":[{"comment":"The phrase 'between HgCd and CdTe' appears to be a typo; it should be 'between HgTe and CdTe' (or 'HgCdTe') to match the preceding sentence. Please make the target-crystal nomenclature consistent throughout.","section":"Section III"},{"comment":"Please report the number of bins and degrees of freedom for the quoted χ²_fit and p-values, so that the reader can interpret the fit-quality cut χ²_fit ≤ 121 and the claimed p-values.","section":"Figure 2 and Fig. 8"},{"comment":"Equation (9) treats each Ne bin as an independent Poisson variable; please state the bin width and clarify how the asymptotic chi-square approximation for qμ was validated over the full scan range, not just for the background-only case.","section":"SM §III"}],"recommendation":"major_revision","confidential_remarks":"This is a promising paper from an experienced group, and the archival-data approach is timely. The main blocking issue is the data-dependent dataset selection; it should be addressable with signal-injection tests and by publishing the rejected fits. The form-factor and shielding caveats are secondary but should be accompanied by sensitivity checks. The paper is within scope for a particle-astrophysics journal and I would not reject it on novelty grounds."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is the application, not the framework. This is the first use of JWST NIRSpec dark calibration images to constrain strongly-interacting sub-GeV dark matter, and it closes a genuinely open region for the ultralight dark-photon mediator at fχ = 0.4%. The signal calculation uses existing tools (QEDark, DaMaSCUS-CRUST), and the statistics are a standard profile likelihood over the pixel charge distribution. That's fine. The authors are careful with the JWST pipeline, describe the custom masks in enough detail to reproduce, show all fourteen fitted charge distributions, and make the data public with a DOI. Credit where due: this is a solid, reproducible new constraint.\n\nThe soft spots are real and roughly in proportion. The biggest is the selection of 14 of 30 datasets by background-only fit quality (χ²_fit ≤ 121). The stress-test concern is not manufactured: if a DM signal were present, it would skew the charge distribution and could make a background-only fit fail. The sixteen discarded datasets are exactly the ones where the signal might be hiding. The paper offers no signal-injection or coverage test showing the selection preserves DM events. That's a load-bearing gap. A referee should demand a simulation-based test: inject the expected DM signal into all 30 datasets at the claimed exclusion boundary, rerun the selection and the limit, and show the coverage is still ~95%. Without that, the exclusion boundary could be over-optimistic.\n\nTwo smaller issues. The HgCdTe crystal form factor is approximated; the paper says the exact calculation is 'to appear.' That's an acknowledged systematic that could shift the rate by an O(1) factor. The detector shielding model leans on a private communication and a simplified geometry; again acknowledged, but the limit in the high-cross-section turnover region will depend on it. These are minor-to-moderate, not fatal.\n\nOverall: the paper is honest, the analysis is mostly transparent, and the central idea is sound. It should go to peer review, with the selection-bias closure test as the main condition. I'd bring it to reading group; there's a good discussion here about fitting versus scanning and about when it is legitimate to discard bad datasets.","headline":"First real use of JWST NIRSpec dark frames to constrain strongly-interacting sub-GeV dark matter, with a novel exclusion for the ultralight dark-photon model — but the limit rests on a dataset-selection step that needs a signal-injection closure test before I'd trust the boundary.","tokens_in":20805,"tokens_out":1961,"would_cite":true,"duration_ms":20311,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"JWST's dark calibration frames can be repurposed as a detector for strongly interacting sub-GeV dark matter, already disfavoring the open high-cross-section window for a 0.4% subcomponent.","keywords":["dark matter","strongly interacting dark matter","sub-GeV dark matter","JWST NIRSpec","dark photon","direct detection","pixel charge distribution","dark current"],"falsifier":"Inject synthetic dark-matter signals into the raw dark frames at the excluded cross sections, process them through the identical masks and dataset selection, and check whether the 95% upper limits contain the injected cross section in at least 95% of trials; if the coverage is short, the reported exclusion is biased.","tokens_in":19822,"feed_emoji":"🔭","tokens_out":20068,"duration_ms":167750,"temperature":0.7,"pith_summary":"The paper's central claim is that JWST's NIRSpec \"dark\" calibration frames—images taken with an opaque filter to characterize instrument noise—can be used as a dark-matter detector, and that they already place new upper limits on strongly interacting sub-GeV dark matter. The target is a fermion that makes up a small fraction $f_\\chi$ of the dark matter and interacts with ordinary electrons through an ultralight dark-photon mediator; such particles would be stopped in the atmosphere or crust before reaching underground detectors, which is why the high-cross-section window has remained open. The authors add five custom masks to the standard JWST pipeline, model the expected few-electron signal in the HgCdTe detector, and compare the shape of the masked pixel-charge distribution with a background of dark current plus readout noise. For a subcomponent fraction of 0.4%, they disfavor all previously allowed high-cross-section parameter space between roughly 1 MeV and 10 GeV at 95% confidence, with some sensitivity persisting down to about 0.01%. If correct, this means existing space-based calibration data can probe a regime that terrestrial direct-detection experiments cannot reach.","feed_headline":"JWST dark images exclude a slice of strongly interacting dark matter","feed_subtitle":"A 0.4% dark-matter subcomponent with an ultralight dark photon is disfavored for masses from 1 MeV to 10 GeV.","key_machinery":"The load-bearing object is the per-pixel charge distribution $F_{\\rm obs}(N_e)$ of the NRS2 dark frames after masking, compared with the model $F(N_e) = N_{\\rm tot} h \\sum_n f(n; \\sigma_e, \\lambda_{\\rm DC}) [r\\,\\mathcal{N}(N_e; n, \\sigma_1) + (1-r)\\,\\mathcal{N}(N_e; n, \\sigma_2)]$, where $f(n; \\sigma_e, \\lambda_{\\rm DC})$ is the dark-current distribution convolved with the DM signal and the two normal components describe readout noise with standard deviations $\\sigma_1$ and $\\sigma_2$. The DM signal comes from the scattering rate integrated over the shield-attenuated DM speed distribution, using the HgTe and CdTe crystal form factors and a dielectric screening function, with the deposited energy converted to electron-hole pairs through $Q(E_e) = 1 + \\lfloor (E_e - E_{\\rm gap})/\\epsilon_{eh}\\rfloor$. What makes the background model credible is the masking chain: a jump mask removes pixels with an inter-frame step above 60 DN, a cluster-and-halo mask removes mosaics near high-energy clusters and their surrounding halos, an edge mask trims the image border, a hot-column mask drops the 60 noisiest mosaic columns, and a brightness mask removes pixels with $|y_{\\rm int}| > 150$ DN plus a two-pixel radius, leaving about 5% of pixels. The final comparison is made within $|N_e - N_e^{\\rm peak}| \\leq 55$, and the 95% confidence limit is set by the one-sided profile-likelihood test statistic $q_\\mu$. This machinery converts otherwise-discarded calibration exposures into a shape-based direct-detection limit.","core_discovery":"The central discovery, stated on the paper's own terms, is that the JWST NIRSpec dark images provide novel 95% confidence limits on the DM-electron scattering cross section for sub-GeV dark matter coupled to an ultralight dark-photon mediator. After applying a subset of the JWST calibration pipeline (superbias subtraction, reference-pixel correction, linearity correction, and saturation flagging) plus five custom masks (jump, cluster-and-halo, edge, hot-column, and brightness), the surviving pixel charge distribution is fit by a background model of a single random dark-current component smeared by two readout-noise components. A dark-matter signal is added as the convolution of a DM charge spectrum, computed from the crystal form factor, dielectric screening, and a Monte-Carlo-simulated, shield-attenuated DM velocity distribution, with the same dark current and readout kernel, and a one-sided profile likelihood ratio is used to set the 95% upper limit. The result is that the previously allowed high-cross-section region for a 0.4% subcomponent is disfavored for DM masses from about 1 MeV to 10 GeV, with weaker but nonzero constraints for subcomponent fractions as low as about 0.01%.","pith_inferences":["A natural stress test would be to apply the identical masks and likelihood to the rejected datasets alone; if they show a positive best-fit dark-matter cross section or an excess that follows the predicted signal shape, the reported exclusion could be biased by the selection of the well-fit images.","Because the boundary of the excluded region sits where the roughly 20 mm silicon-carbide shield becomes opaque, the constraint is sensitive to the assumed shield geometry; cross-checking against the NRS1 detector, which may have slightly different surroundings, could quantify that systematic.","The same shape-based logic could be extended to other strongly interacting dark-matter models, such as those with different form factors or mediators, simply by replacing the predicted electron-recoil spectrum inside the likelihood.","A future low-noise, minimally shielded space detector with single-electron counting should push this technique to much smaller subcomponent fractions and to even higher cross sections, because the limiting factor here is dark-current stability and readout noise rather than the number of dark-matter events."],"forward_implications":["For a 0.4% subcomponent with an ultralight dark-photon mediator, the previously open high-cross-section region between about 1 MeV and 10 GeV is excluded at 95% confidence.","Some parameter space is constrained for subcomponent fractions as low as about 0.01%, so even a very small dark-matter subcomponent can perturb the pixel charge distribution of a space-based detector.","For subcomponent fractions above about 0.4%, cosmic-microwave-background bounds already cover much of this parameter space, making the JWST constraint most useful in the small-subcomponent gap where earlier limits vanish.","Applying the same analysis to the NRS1 detector or to additional dark-calibration frames would extend or sharpen the excluded region, since the method is limited by the number of well-fit exposures."],"supporting_citations":[{"why":"It supplies the DM-electron scattering rate formula used to predict the JWST signal.","marker":"[1]"},{"why":"It establishes that strongly interacting sub-GeV dark matter is stopped by Earth's atmosphere and crust, defining the open parameter region the JWST analysis targets.","marker":"[19]"},{"why":"It provides the cosmic-microwave-background bound that vanishes for subcomponent fractions of about 0.4% or less, which is the gap the JWST constraint fills.","marker":"[29]"},{"why":"It provides the crystal form factor calculation for HgTe and CdTe targets used in the electron-recoil spectra.","marker":"[75]"},{"why":"It supplies the dielectric screening function used to model low-energy DM-electron scatterings in the HgCdTe detector.","marker":"[85]"},{"why":"It documents the NIRSpec instrument structure used to estimate the detector shielding composition and thickness.","marker":"[87]"},{"why":"It provides additional NIRSpec detector and shielding details used in the conservative flux-reduction estimate.","marker":"[88]"},{"why":"It provides the Monte-Carlo simulation code used to compute the dark-matter velocity distribution after passage through the detector shield.","marker":"[90]"},{"why":"It describes the underlying Monte-Carlo method for dark-matter propagation through shielding that the flux calculation relies on.","marker":"[91]"},{"why":"It defines the profile-likelihood ratio and one-sided test statistic used to set the 95% confidence upper limits.","marker":"[101]"}],"fun_headline_variants":["JWST dark images put new limits on strongly-interacting DM","Sub-GeV dark matter constrained by JWST's dark frames","JWST's dark images shrink the room for strongly-interacting DM","JWST calibration data curb strongly-interacting dark matter","JWST dark frames disfavor strongly-interacting dark matter"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole exclusion rests on assuming that, after masking, the surviving pixel charges are one clean, well-understood background plus detector readout noise, and that images failing this background description can be thrown away as noise rather than counted as possible dark-matter signals.","fun_headline_variants_meta":{"raw":{"variants":["JWST dark images put new limits on strongly-interacting DM","Sub-GeV dark matter constrained by JWST's dark frames","JWST's dark images shrink the room for strongly-interacting DM","JWST calibration data curb strongly-interacting dark matter","JWST dark frames disfavor strongly-interacting dark matter"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001054,"raw_usage":{"total_tokens":4426,"prompt_tokens":947,"completion_tokens":3479,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":3394}},"tokens_in":563,"tokens_out":3479,"duration_ms":27027,"temperature":1.0,"reasoning_tokens":3394,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T13:23:04.681842+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inject synthetic dark-matter signals into the raw dark frames at the excluded cross sections, process them through the identical masks and dataset selection, and check whether the 95% upper limits contain the injected cross section in at least 95% of trials; if the coverage is short, the reported exclusion is biased.","supporting_citations":[{"cited_title":"DaMaSCUS- CRUST v1.1, [ascl:1803.001] Available at https://github.com/temken/ and archived under [DOI: 10.5281/zenodo.2846401],","cited_arxiv_id":null,"evidence_quote":"It provides the Monte-Carlo simulation code used to compute the dark-matter velocity distribution after passage through the detector shield."}],"review_version":1}