{"id":"d3dbe657-cf42-4e2e-bf9c-acb87cd3beeb","arxiv_id":"2506.16896","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"COSMIC-L is a deep three-band photometric catalog of 57,997,665 stars in the Large Magellanic Cloud, complete to magnitude 21 with detections to magnitude 22.","lead":"This paper presents COSMIC-L, a new catalog of about 58 million stars in the Large Magellanic Cloud, built from ground-based telescope images in three colors. The catalog maps the galaxy's stellar populations and is intended for studies of star formation, distances, and microlensing.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Limiting magnitudes rest on an unvalidated SNR extrapolation that is biased at the faint end; the 'deepest' claim is therefore unsecured.","rationale":"Good-faith reading: the data reduction and calibration chain (SExtractor + PSFEx + ATLAS reference) is reasonable, and the paper is honest that the depth estimates are indirect. The central claim, however, is the depth and uniqueness of the catalog, and that claim rests on the Section 3 extrapolation. The observed SNR distribution at faint magnitudes is not the true SNR distribution because of detection truncation; this is a well-known statistical selection effect, not a matter of differing conventions. The omission of any comparison with existing DECam LMC surveys also leaves the 'deepest currently available' assertion unsupported, even if the extrapolated values were correct. The reader's verdict of CONDITIONAL is exactly right, and the proposed injection-recovery test would materially change confidence. I do not see grounds to reject outright, because the catalog could well be as deep as claimed and simply needs the validation it explicitly lacks.","tokens_in":6643,"tokens_out":7936,"duration_ms":91350,"concrete_test":"Run injection–recovery simulations on representative DECam fields (e.g., central fields F9 and F14 and outer fields F1 and F23): inject roughly 10,000 synthetic stars per band at magnitudes 18–24 with realistic PSF and Poisson noise, run the same SExtractor/PSFEx pipeline, and compare the recovered 90% completeness and the magnitude at which recovered S/N = 5 with Table 3. If those recovered depths are more than about 0.2 mag brighter than the quoted values, the depth claim, and hence the 'deepest' claim, is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is the depth estimate in Section 3. The limiting magnitudes in Table 3 are obtained by fitting an exponential decay to the measured SNR–magnitude relation (Figure 3) and reading off the magnitude at which the fit crosses S/N = 5, using Equation 1. The paper itself notes that this is done 'in the absence of injection–recovery simulations.' That matters because the catalog is detection-thresholded: at a given faint magnitude, only sources whose noise fluctuations pushed them above the detection limit enter the sample, so the average measured SNR of cataloged sources is biased high relative to the true SNR of all sources at that magnitude. An exponential fit to this selected sample therefore overestimates the faint-end SNR and likely yields limiting magnitudes (g = 22.97, r = 21.64, i = 21.71) that are too faint. The completeness magnitudes (21.38, 20.82, 20.79) are likewise derived from a linear fit to the rising side of the magnitude histogram with an ad hoc 5% deviation criterion and no quoted uncertainty. Because the paper's headline claim of being the deepest ground-based LMC catalog depends on these numbers, and because no comparison with existing DECam LMC surveys such as SMASH is made, the central claim is not supported by the presented analysis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents COSMIC-L, a photometric catalog of 57,997,665 sources detected by the DECam in the g, r, and i bands over 23 fields covering the Large Magellanic Cloud. The authors reduce public DECam images with a pipeline based on SExtractor and PSFEx, calibrate instrumental magnitudes against the ATLAS reference catalog with a per-field linear fit, and release positions, magnitudes, and uncertainties. They report completeness magnitudes of 21.38 (g), 20.82 (r), and 20.79 (i) and limiting magnitudes of 22.97 (g), 21.64 (r), and 21.71 (i), derived from a linear fit to the bright side of the magnitude histograms and from an exponential fit to the signal-to-noise ratio versus magnitude relation extrapolated to S/N = 5. The paper also presents color-magnitude and color-color diagrams and identifies a foreground population of K- and M-type stars using Gaia EDR3 distances. The central claim is that COSMIC-L is the deepest currently available ground-based photometric catalog of LMC stars.","tokens_in":7022,"tokens_out":2369,"duration_ms":28146,"significance":"If the depth claims are reliable, COSMIC-L would be a valuable community resource: it is a large, homogeneous, publicly derived catalog with per-field photometric calibration, and the color-magnitude products demonstrate immediate scientific utility for studying LMC stellar populations and foreground contamination. The authors correctly follow a published pipeline from Franco et al. (2025) and make the data products publicly available. However, the headline depth claims rest on two indirect estimation procedures that are not validated against injection-recovery simulations or external catalogs, and no comparison is made with existing DECam-based LMC surveys such as SMASH. The significance of the catalog is therefore conditional on strengthening the depth analysis.","major_comments":[{"comment":"The limiting magnitudes in Table 3 are obtained by fitting an exponential decay to the measured SNR-magnitude relation and extrapolating to S/N = 5. This is problematic because the catalog is detection-thresholded: at a given faint magnitude, only sources whose noise fluctuations pushed them above the detection limit enter the sample, so the mean measured SNR of cataloged sources is biased high relative to the true SNR of all sources at that magnitude. The exponential fit to this selected sample therefore likely overestimates the faint-end SNR and yields limiting magnitudes that are too faint. The authors explicitly note that this is done 'in the absence of injection-recovery simulations,' but they do not quantify the resulting bias or provide any external validation. I request an injection-recovery analysis, or a cross-check against a deeper external catalog, and a statement of the propagated uncertainty on the quoted limiting magnitudes.","section":"Section 3, Eq. (1), Figure 3, Table 3"},{"comment":"The completeness magnitudes are derived from a linear fit to the rising side of the magnitude histogram with an ad hoc 5% deviation criterion. No uncertainties are quoted for the fitted slopes, intercepts, or the resulting 5% crossing magnitudes, and the choice of the fitting range is not justified. Because the completeness values are a central component of the 'complete to about 21st magnitude' claim, the authors should quote uncertainties, justify the fitting range and criterion, and ideally validate the completeness estimate with injected sources or a comparison to a deeper catalog.","section":"Section 3, Figure 5, Table 3"},{"comment":"The statement that COSMIC-L is 'the deepest ground-based photometric catalog of the LMC stars currently available' is not supported by the analysis as presented. No comparison is made with existing LMC surveys such as SMASH, which also uses DECam, nor with other deep ground-based catalogs. The depth comparison should be made with a consistent limiting-magnitude definition, and the claim should either be substantiated with such a comparison or softened to a description of the present catalog's measured depth.","section":"Section 4 (Conclusions)"},{"comment":"The paper states that 'numerous sources are detected down to magnitudes as faint as m 25,' yet the fitted limiting magnitudes are about 22 in all bands. This discrepancy is not discussed. If sources at m 25 are real, the exponential SNR fit is clearly not describing the faint end of the catalog, which further undermines the extrapolated limiting magnitudes; if they are spurious, the catalog's faint-end contamination should be quantified. Either way, this point needs to be addressed.","section":"Section 3, Figure 3 and Figure 4"}],"minor_comments":[{"comment":"The abstract describes the LMC as containing 'approximately 30 billion stars,' while the catalog contains about 58 million sources; the wording should be clarified to distinguish the total stellar content of the galaxy from the number of cataloged sources.","section":"Abstract"},{"comment":"Figure 1 applies vertical shifts of 0.2 and 1.0 to the r- and i-band distributions for readability, but the shifts are only described in the caption; the same information should be indicated in the figure itself or in a clearer legend.","section":"Section 2, Figure 1"},{"comment":"The table gives completeness and limiting magnitudes without uncertainties; since both quantities come from fits, the fit uncertainties should be reported in the table or in the text.","section":"Section 3, Table 3"},{"comment":"The catalog access information is incomplete: the paper states that the data are publicly available from NOIRLab and describes the catalog columns, but it does not state where the derived COSMIC-L catalog itself can be downloaded, nor does it provide a DOI or a persistent archive link.","section":"Section 3"},{"comment":"The paper cites OGLE and other surveys but does not cite SMASH, the most relevant DECam-based LMC survey for a depth comparison. Adding SMASH and related LMC photometric catalogs to the discussion would strengthen the context for the 'deepest' claim.","section":"References"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"COSMIC-L is a large, homogeneous gri photometric catalog of the LMC built from 23 DECam fields with PSF photometry calibrated to ATLAS. That is the real content: 58M sources, ~18.7M in all three bands, with CMD and color-color diagrams showing the expected LMC populations plus a clean foreground K/M dwarf sequence. For stellar population work and transient follow-up this is a plausible resource, assuming the depth claims hold.\n\nWhat the paper does well: the reduction follows the previously described COSMIC-S pipeline, the calibration slopes are per-field and per-band with sensible values, and the paper is honest that depth estimates are made 'in the absence of injection-recovery simulations.'\n\nThe soft spot is exactly the depth estimate, and it is load-bearing. Section 3 derives completeness from a linear fit to the histogram just before the peak with a 5% deviation criterion, and limiting magnitudes by fitting an exponential to the measured SNR versus magnitude and reading off S/N = 5. The measured SNR of cataloged sources is biased high at faint magnitudes because the catalog is detection-thresholded: only sources that scattered above the limit are included, so the faint-end exponential overestimates the true SNR. That likely makes 22.97/21.64/21.71 too faint, and no uncertainty is quoted anywhere. The 'deepest ground-based LMC catalog' claim in Section 4 is unsupported without comparing to existing DECam surveys like SMASH, which reach fainter in r/i. The paper does not mention SMASH or MCPS at all.\n\nAlso, the central product is not made available. A catalog paper with no catalog link and no data availability section is hard to evaluate; the reader cannot check the photometry or use the entries.\n\nNone of this kills the underlying value. If the authors release the catalog, do an injection-recovery test, and compare depths against SMASH/MCPS, the paper would be a solid resource. As it stands, it is a useful dataset in need of validation. I would send it to peer review, but require those revisions.","headline":"A potentially useful 58M-source LMC catalog, but the headline 'deepest' claim rests on an SNR extrapolation that is biased at the faint end and there is no comparison with existing surveys.","tokens_in":7445,"tokens_out":1732,"would_cite":false,"duration_ms":17259,"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":"COSMIC-L presents a photometric catalog of 57,997,665 stars in the Large Magellanic Cloud, complete to about 21st magnitude with detections to about 22nd magnitude in the g, r, and i bands.","keywords":["catalogs","photometry","Large Magellanic Cloud","DECam","PSF photometry","stellar populations","Magellanic Clouds","signal-to-noise depth"],"falsifier":"Run injection-recovery simulations on the same DECam images: plant artificial stars of known brightness in the g, r, and i frames, re-run the detection pipeline, and measure the magnitude at which the recovery fraction drops to 50%. If that 50% recovery magnitude is significantly brighter than the claimed limiting magnitudes (22.97, 21.64, 21.71 in g, r, i), the extrapolated signal-to-noise depth is not real and the catalog is not as deep as claimed.","tokens_in":6455,"feed_emoji":"🌌","tokens_out":5487,"duration_ms":48268,"temperature":0.7,"pith_summary":"This paper presents COSMIC-L, a photometric catalog of 57,997,665 stars in the direction of the Large Magellanic Cloud, built from 23 DECam pointings in the SDSS g, r, and i bands. The authors' central claim is that the catalog is complete to about 21st magnitude and reaches detections near 22nd magnitude in all three bands, making it the deepest ground-based photometric catalog of LMC stars currently available. Roughly 18.7 million sources have reliable photometry in all three bands, enough to build color-magnitude and color-color diagrams that resolve the LMC main sequence, giant branch, red clump, and a foreground population of Milky Way K- and M-type stars. A catalog this deep matters because the LMC hosts standard candles such as RR Lyrae stars and Cepheids, so precise multiband photometry feeds the extragalactic distance scale and searches for microlensing transients.","feed_headline":"57,997,665 stars mapped in the Large Magellanic Cloud","feed_subtitle":"COSMIC-L reaches about 22nd magnitude in gri, the deepest ground-based LMC catalog to date.","key_machinery":"The analysis is carried by a PSF-photometry pipeline: SExtractor detects sources and PSFEx builds point-spread-function models, which are then used to measure calibrated magnitudes for every source, with the ATLAS All-Sky Stellar Reference Catalog supplying the photometric zero points through a linear fit per field and band. Depth estimates rest on two auxiliary fits: a linear fit to the rising side of the magnitude histogram, used to define the completeness magnitude where counts fall 5% below the line, and an exponential decay fit to signal-to-noise versus magnitude, extrapolated to the signal-to-noise threshold of 5 to define the limiting magnitude. The color-magnitude and color-color diagrams, built from the 18.7 million three-band sources, are what allow the authors to separate LMC populations from foreground Galactic stars.","core_discovery":"On its own terms, the paper establishes that a uniformly processed, PSF-based photometric reduction of 23 public DECam fields yields a catalog of 57,997,665 sources in the LMC region, with 18,676,294 measured in all three gri bands. Photometric calibration against the ATLAS reference catalog gives mean uncertainties of 0.03 mag in g, 0.033 in r, and 0.035 in i. The authors estimate completeness magnitudes of 21.38 (g), 20.82 (r), and 20.79 (i) from the point where the magnitude histogram deviates 5% from a linear fit, and limiting magnitudes of 22.97, 21.64, and 21.71 by extrapolating an exponential fit of signal-to-noise versus magnitude to a signal-to-noise ratio of 5. The catalog is therefore presented as the deepest current ground-based photometric catalog of the Large Magellanic Cloud.","pith_inferences":["A natural extension the paper does not pursue is a time-series analysis: the 23 fields were observed between February 2018 and January 2020, so the multi-epoch exposures could be stacked or differenced to yield variability and proper-motion information rather than a single-epoch catalog.","The completeness and limiting magnitudes could be checked directly with injection-recovery simulations on the same images; the paper explicitly notes it uses indirect methods in the absence of such simulations.","Cross-matching COSMIC-L with space-based photometry in the same region would test whether the exponential signal-to-noise extrapolation holds at the faint end, where crowding and noise may depart from the fit.","If the faint-end depths hold, the catalog becomes a resource for identifying low-luminosity stellar populations and rare evolved stars that shallower ground-based surveys have missed."],"forward_implications":["COSMIC-L provides a deep, homogeneous gri photometric base for studying the LMC's stellar populations, star formation history, and red clump structure.","The catalog's standard-candle stars (RR Lyrae stars and Cepheids) can be used to refine extragalactic distance estimates and, in turn, the local calibration of the Hubble constant.","The detection of nearly 57 million sources, including sources to about 25th magnitude, offers a dense target list for transient searches such as microlensing events.","The identification of 76,267 foreground K- and M-type stars with distances mostly within 5 kpc provides a sample for mapping the Milky Way's thin and thick disks.","Because the catalog is built from public DECam data using a reproducible pipeline, the same strategy can be applied to other Magellanic Cloud fields or neighboring dwarf galaxies."],"supporting_citations":[{"why":"Supplies the ATLAS All-Sky Stellar Reference Catalog used as the photometric calibration reference for all three bands.","marker":"Tonry et al. 2018"},{"why":"Provides SExtractor, the source detection and measurement tool in the pipeline.","marker":"Bertin & Arnouts 1996"},{"why":"Provides PSFEx, used to extract the PSF models essential for crowded-field photometry.","marker":"Bertin 2011"},{"why":"Describes DECam, the instrument whose 23 gri fields make up the dataset.","marker":"Flaugher et al. 2015"},{"why":"Establishes the analysis strategy, error propagation, and catalog format that COSMIC-L adopts from the COSMIC-S SMC catalog.","marker":"Franco et al. 2025"},{"why":"Supplies distance corrections applied to the foreground K/M-star sample identified in the color-magnitude diagram.","marker":"Bailer-Jones et al. 2021"},{"why":"Provides the Gaia EDR3 parallaxes from which foreground star distances are derived.","marker":"Gaia Collaboration: Vallenari et al. 2023"},{"why":"Provides the spectral-type and luminosity-class sequences overlaid on the color-color diagram to classify stellar populations.","marker":"Covey et al. 2007"}],"fun_headline_variants":["COSMIC-L: 58M stars in the Large Magellanic Cloud","Deepest ground-based LMC catalog: 58M stars","58 million stars mapped in the LMC with DECam","LMC catalogs 18.7M full-gri stars out of 58M","New LMC catalog reaches magnitude 22 in gri"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quoted limiting magnitudes assume that the exponential fit to signal-to-noise versus magnitude remains valid when extrapolated to the signal-to-noise threshold of 5, and that the noise does not flatten or turn over at fainter magnitudes; if the fit is wrong, the claimed depth and the 'deepest ground-based catalog' claim would be unsupported.","fun_headline_variants_meta":{"raw":{"variants":["COSMIC-L: 58M stars in the Large Magellanic Cloud","Deepest ground-based LMC catalog: 58M stars","58 million stars mapped in the LMC with DECam","LMC catalogs 18.7M full-gri stars out of 58M","New LMC catalog reaches magnitude 22 in gri"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000353,"raw_usage":{"total_tokens":1896,"prompt_tokens":894,"completion_tokens":1002,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":910}},"tokens_in":510,"tokens_out":1002,"duration_ms":10039,"temperature":1.0,"reasoning_tokens":910,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:15:59.747645+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run injection-recovery simulations on the same DECam images: plant artificial stars of known brightness in the g, r, and i frames, re-run the detection pipeline, and measure the magnitude at which the recovery fraction drops to 50%. If that 50% recovery magnitude is significantly brighter than the claimed limiting magnitudes (22.97, 21.64, 21.71 in g, r, i), the extrapolated signal-to-noise depth is not real and the catalog is not as deep as claimed.","supporting_citations":[{"cited_title":"L., Denneau, L., Flewelling, H., Heinze, A","cited_arxiv_id":null,"evidence_quote":"Supplies the ATLAS All-Sky Stellar Reference Catalog used as the photometric calibration reference for all three bands."},{"cited_title":"and Arnouts, S","cited_arxiv_id":null,"evidence_quote":"Provides SExtractor, the source detection and measurement tool in the pipeline."},{"cited_title":"2011, Astronomical Data Analysis Software and Systems XX","cited_arxiv_id":null,"evidence_quote":"Provides PSFEx, used to extract the PSF models essential for crowded-field photometry."},{"cited_title":"T., Honscheid, K., Abbott, T","cited_arxiv_id":null,"evidence_quote":"Describes DECam, the instrument whose 23 gri fields make up the dataset."},{"cited_title":"A., De Paolis, F","cited_arxiv_id":null,"evidence_quote":"Establishes the analysis strategy, error propagation, and catalog format that COSMIC-L adopts from the COSMIC-S SMC catalog."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies distance corrections applied to the foreground K/M-star sample identified in the color-magnitude diagram."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Gaia EDR3 parallaxes from which foreground star distances are derived."},{"cited_title":"R., Ivezic, Z., Schlegel, D., Finkbeiner, D","cited_arxiv_id":null,"evidence_quote":"Provides the spectral-type and luminosity-class sequences overlaid on the color-color diagram to classify stellar populations."}],"review_version":2}