{"id":"acbf0fd4-7974-4eec-a11e-0a575fe98efd","arxiv_id":"2506.19068","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Using lognormal simulations, the authors find that BINGO 21 cm intensity mapping cross-correlated with LSST photometric galaxies can detect the HI signal even with photo-z errors, though with significance comparable to autocorrelation.","lead":"This paper simulates BINGO's 21 cm hydrogen signal and the LSST galaxy survey to test whether cross-correlating them can detect cosmic structure despite photometric redshift errors. It finds the signal stays detectable, with significance similar to BINGO's own autocorrelation, but only in simulations whose signal comes from the same model used for detection.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"GNILC-deb detectability numbers rely on a covariance from 'fast' simulations that Sec. 4.2.2 admits are not representative of cleaned maps; until this mismatch is quantified, the central detectability claim is not fully supported.","rationale":"The reader's conditional verdict identifies the same weakest link: covariance mismatch. I agree this is the load-bearing issue. The paper's own Sec. 4.2.2 admits the fast simulations are not representative; because the sqrt(delta chi^2) values are the quantitative basis for 'detectable,' this cannot be dismissed. Other concerns—closed-loop use of UCLC_l for both signal injection and detection model, post hoc selection of 99<ell<309, and the absence of 1/f noise and polarization leakage—are real but secondary: they affect the robustness of the forecast as a statement about the real sky, not the internal consistency of the pipeline comparison. The relative finding (cross-correlation significance comparable to autocorrelation) is supported by the simulations and would survive a covariance recalibration in either direction; hence I would not move the verdict. The paper should either recalibrate the covariance or temper the abstract's 'realistic conditions' phrasing, which is exactly what a conditional recommendation captures.","tokens_in":22965,"tokens_out":7128,"duration_ms":75598,"concrete_test":"Generate a new ensemble of at least 200 GNILC-cleaned BINGO-like maps (same pipeline, independent noise and foreground realizations) and compute the cross-APS covariance directly from these maps over the 99<ell<309 band. Recompute sqrt(delta chi^2) for the central HI-LSST pairs (HI14xG1, HI23xG2, HI30xG3) and for the 30-bin auto/cross panels of Fig. 8, applying the Hartlap factor. If the significance values stay within about 1 sigma of the published values, the fast-simulation covariance is adequate; if they drop below the autocorrelation values or below the detection thresholds, the detectability claim needs revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive numbers in Sec. 4.2.2—the sqrt(delta chi^2) values for the GNILC-deb scenario—are computed with a covariance estimated from 1500 'fast' simulations (HI+WN+Fg), not from actual GNILC-cleaned maps. The fast simulations are constructed by adding white noise and a residual foreground template to the input HI maps; the foreground residual is repeated every 50 realizations, and the fast maps do not include the cleaning-induced signal loss or the non-white noise structure that GNILC actually produces. The authors explicitly state in Sec. 4.2.2 that the fast simulations are not representative of the cleaned maps (green vs purple points diverge in Fig. 8). If the fast-simulation covariance is not the covariance of the cleaned maps, then every significance value quoted for the GNILC-deb scenario is miscalibrated; the central claim that the HI signal 'remains detectable' under LSST-like photo-z is an absolute statement built on those values. The relative comparison to the autocorrelation is more robust, but the abstract's feasibility claim is not fully supported until the covariance mismatch is quantified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript uses lognormal FLASK simulations to assess whether the BINGO 21 cm intensity mapping survey can detect the HI signal by cross-correlating with LSST photometric galaxy catalogs. The authors generate 30 BINGO frequency bins with thermal noise and foregrounds, apply GNILC foreground cleaning with debiasing to 50 realizations, and build LSST-like galaxy maps in three photo-z bins with realistic number densities and photo-z errors. They measure auto- and cross-angular power spectra, compute detection significance via sqrt(Delta chi^2), and fit the degenerate amplitude b_HI Omega_HI r. The main claim is that photo-z errors add noise that reduces the cross-correlation significance to levels comparable to the autocorrelation, but the signal remains detectable, and the degenerate astrophysical parameters can be constrained.","tokens_in":23066,"tokens_out":4547,"duration_ms":51540,"significance":"If the central claim holds, this is a useful feasibility study for a relatively unexplored observational route: using photometric galaxy surveys, rather than spectroscopic ones, for HI intensity mapping cross-correlations. The paper has clear strengths: it builds a reasonably detailed end-to-end simulation pipeline including realistic foregrounds, GNILC cleaning, debiasing, a null test, and parameter estimation; it uses public tools (FLASK, UCLC_l, NaMaster); and it makes an explicit comparison with a previous spectroscopic-based study. The novelty of using the full photo-z bin, rather than slicing to match the HI bin width, is a genuine contribution. However, the quantitative detectability claim rests on a significance calculation whose covariance is taken from 'fast' simulations that the authors themselves state are not representative of the cleaned maps, and the injection and detection template share the same theoretical power spectrum. Both issues need to be addressed before the abstract's 'remains detectable' statement is fully supported.","major_comments":[{"comment":"The sqrt(Delta chi^2) values for the GNILC-deb scenario are computed using the covariance matrix from the 1500 fast simulations (HI+WN+Fg), not from the GNILC-cleaned maps. The fast simulations omit the cleaning-induced signal loss and the non-white noise structure produced by GNILC, and the authors themselves note that the green and purple points diverge in Fig. 8, concluding that the fast simulations 'are not representative enough of the foreground cleaned maps.' Since the abstract's detectability claim is an absolute statement built on these significance values, the covariance mismatch must be quantified. I recommend comparing the fast-simulation covariance with one estimated from cleaned maps (or from an analytic model of the cleaning-induced noise), and showing how the reported significance values and their error bars change. Without this, the central claim is not fully supported.","section":"Sec. 4.2.2, Fig. 8"},{"comment":"The simulated HI and galaxy maps are generated with FLASK using the UCLC_l angular power spectra, and the detection template used in the chi^2 statistic of Sec. 4.2.2 is the same UCLC_l model. The analysis is therefore a closed-loop recovery of an injected signal from a known model: any error or bias in the theoretical power spectrum is common to both simulation and template and cannot be detected. This inflates the significance relative to what would be obtained if the true sky differed from the model. I suggest testing robustness by injecting the signal with an alternate power spectrum (e.g., a different set of astrophysical parameters or a different nonlinear prescription) while keeping the detection template fixed, and reporting the resulting significance. This would turn the circularity concern into a quantitative statement.","section":"Secs. 2 and 3; Eq. (2.7)"},{"comment":"The multipole range 99 < ell < 309 used for the significance calculation is described as chosen to 'avoid multipole with poor SNR and to maximize the statistical significance.' Selecting the analysis range after inspecting the results, without a penalty or a pre-specified criterion, introduces an a posteriori selection that can bias the reported significance upward. The authors should either fix the range before the analysis, show the significance for a set of pre-defined ranges, or correct for the number of trial ranges considered. This is especially important because the same range is used for the parameter constraints in Sec. 5.","section":"Sec. 4.2.2"},{"comment":"The null test uses Gaussian random maps with standard deviation sigma to mimic non-correlated galaxies, but such maps have white-noise statistics rather than a realistic galaxy field with clustering and shot noise. The test therefore demonstrates only that adding an uncorrelated random field increases the variance of the cross-spectrum, not specifically that photometric redshift uncertainty has that effect. The interpretation that photo-z errors are the cause of the enlarged 68% regions should be supported by comparing against simulations where the photo-z scatter is realized in the galaxy positions, or by using a galaxy mock with an explicit photo-z error model. Additionally, the parameter constraints in Sec. 5 inherit the covariance-mismatch issue noted above, so the reported 1-sigma dispersions on b_HI Omega_HI and b_HI Omega_HI r may be underestimated until the covariance is validated.","section":"Secs. 4.2.1 and 5; Figs. 5 and 9"}],"minor_comments":[{"comment":"The text reads 'the three three scenarios'; this should be 'the three scenarios.'","section":"Sec. 4.2.2"},{"comment":"The caption refers to 'scenarios (i) to (vi)' but the text in Sec. 4.2.1 defines only scenarios (i) through (iv). Please make the numbering consistent.","section":"Sec. 5, Fig. 9 caption"},{"comment":"There are several typographical issues, including missing spaces ('Hisignal'), 'thje' instead of 'the', and 'scenarions' instead of 'scenarios.' A careful proofreading pass is recommended.","section":"Throughout"},{"comment":"The notation phi'(z) is used for both the galaxy selection function n(z) and the HI projection kernel, but the prime is not defined explicitly; please define it at first use to avoid confusion.","section":"Sec. 2.1, Eq. (2.6)"},{"comment":"The description of the fast simulations states that the foreground residual is repeated every 50 realizations. This introduces correlations among realizations that should be acknowledged when using these simulations for the covariance matrix, since the effective number of independent realizations is smaller than 1500.","section":"Sec. 4.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid simulation-based feasibility study, but the central detectability claim depends on a significance calculation with an acknowledged covariance mismatch and on a closed-loop injection/template setup. These are fixable within the scope of the manuscript: quantify the covariance mismatch, add a model-robustness test, and pre-specify or penalize the multipole-range choice. I do not see a need to reject, but the current version overstates the degree to which the results support the abstract's feasibility claim. The paper also fits the journal's scope and the authors appear to have used appropriate public tools and cited relevant prior work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's core comparison—that photo-z errors push cross-correlation significance down to autocorrelation levels—is well supported and worth taking seriously. The absolute \"remains detectable\" numbers are less solid, because the significance for the cleaned maps is computed with a covariance from fast simulations that the authors themselves say don't match the cleaned maps.\n\nWhat's new: this is the first harmonic-space forecast of BINGO x LSST cross-correlation using the full photo-z bins, rather than matching bin widths as Cunnington et al. did. The null test with Gaussian random maps is a clean demonstration that photo-z scatter acts like noise. The pipeline is internally consistent, uses standard tools (FLASK, GNILC, NaMaster), and the relative comparisons across scenarios are believable.\n\nThe soft spots are real but not fatal. In Sec. 4.2.2, the GNILC-deb significance uses the covariance from 1500 HI+WN+Fg fast sims; the paper then notes the green and purple points diverge in Fig. 8, meaning those sims don't capture cleaning-induced signal loss and noise structure. So the quoted significance values are miscalibrated, and we don't know if the true detection significance is higher or lower. This weakens the abstract's \"detectable under realistic conditions\" claim. Also, the injected signal and the detection template both come from UCLC_l, so this is a closed-loop forecast—standard practice, but it makes the absolute numbers optimistic. The multipole range 99<ell<309 was selected after looking at the SNR, so there's a mild post-hoc selection effect. The missing systematics (1/f noise, polarization leakage) are explicitly named in the text, which is commendable, but it means \"realistic\" is doing a lot of work.\n\nThis paper is for people planning BINGO/MeerKAT/SKAO cross-correlations with LSST-like photometric surveys. They'll take the relative photo-z result and the null-test methodology as a useful starting point. The parameter-constraint section is secondary.\n\nWorth a serious referee. A revision that quantifies the covariance mismatch—say, by estimating the covariance from cleaned maps or perturbing it to show how significance shifts—and that fixes the multipole range a priori would make the central claim solid. As is, conditional accept.","headline":"A solid, clearly-written simulation forecast of BINGO x LSST cross-correlation; the relative photo-z effect is convincing, but the absolute detectability numbers rest on a covariance the authors admit is not representative.","tokens_in":23807,"tokens_out":2317,"would_cite":true,"duration_ms":23227,"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":"The 21 cm signal from BINGO remains detectable when cross-correlated with LSST photometric galaxies, despite photo-z uncertainties that reduce its significance to the level of the autocorrelation.","keywords":["21 cm intensity mapping","BINGO","LSST","cross-correlation","angular power spectrum","photometric redshift","GNILC foreground cleaning","neutral hydrogen"],"falsifier":"Take the 50 cleaned realizations, estimate the covariance directly from their measured power spectra (or generate enough cleaned maps to do so), and recompute $\\sqrt{\\Delta\\chi^2}$ in the same multipole range; if the significance drops below the detection threshold, the detectability claim fails. Alternatively, apply the same pipeline to real BINGO and LSST data and check whether the measured cross-power spectrum amplitude is consistent with the model at the claimed significance.","tokens_in":22663,"feed_emoji":"📡","tokens_out":5355,"duration_ms":50533,"temperature":0.7,"pith_summary":"This paper tries to establish that the 21 cm neutral-hydrogen signal observed by the BINGO radio telescope can be detected through cross-correlation with photometric galaxies from the LSST survey, despite the blurring introduced by photometric redshift errors. Using lognormal sky simulations over BINGO's frequency range, with thermal noise, foregrounds, and a needlet-based cleaning step, the authors show that photo-z uncertainty degrades the cross-correlation significance down to about the level of the autocorrelation detection, but does not erase it. The detection survives across the full redshift range where the surveys overlap, centered near mean redshifts 0.25, 0.35, and 0.45. A sympathetic reader would care because most previous 21 cm cross-correlations have relied on spectroscopic galaxy surveys; showing that a photometric survey works would widen the pool of usable data for intensity mapping.","feed_headline":"BINGO's 21 cm signal survives LSST photo-z blur","feed_subtitle":"Simulations show the HI signal stays detectable in cross-correlation, at the same significance as BINGO's autocorrelation.","key_machinery":"The argument rests on three pieces: lognormal realizations of the cosmological HI and galaxy fields generated from theoretical angular power spectra; a foreground-cleaning pipeline, Generalized Needlet Internal Linear Combination, a component-separation method that exploits the smooth frequency structure of foregrounds, followed by debiasing of the power spectra; and the measured angular power spectrum, estimated through pseudo-$C_\\ell$ mode coupling in the multipole range $99<\\ell<309$. The decisive choice is to cross-correlate each narrow HI bin with the entire photometric bin rather than selecting galaxies that match the HI bin width. That choice preserves the cross-correlation amplitude while producing a wider scatter, and the null test with Gaussian random maps attributes the scatter to photo-z scatter.","core_discovery":"The central claim is that the HI signal remains detectable through the angular power spectrum cross-correlation between foreground-cleaned BINGO-like maps and LSST-like photometric galaxy bins, even when photometric redshift errors are as large as LSST's. The photo-z errors do not bias the average cross-correlation amplitude; they add a noise-like contribution from galaxies that do not physically overlap the narrow HI redshift bin, enlarging the error bars until the statistical significance is comparable to the autocorrelation. For the HI bins closest to the center of each photometric bin, the significance is sufficient to claim detection, and the cross-correlation amplitude constrains the degenerate product $b_{\\rm HI}\\Omega_{\\rm HI} r$ without the bias that appears in the autocorrelation at higher redshifts.","pith_inferences":["A testable extension would be to recompute the detection significance using a covariance estimated directly from the cleaned maps rather than from the fast simulations; if the fast-simulation covariance is optimistic, the reported significance could shrink.","The same pipeline should transfer to other photometric surveys with narrower photo-z errors, and if photo-z scatter is the dominant noise term, narrower photo-z should directly raise the significance.","Co-adding the 30 BINGO bins within each photo-z bin might increase the signal-to-noise ratio beyond the per-bin analysis, an avenue the paper mentions as future work.","The null test suggests a practical diagnostic for real data: cross-correlating a cleaned HI map with a shuffled galaxy catalog should reproduce the noise-like error inflation seen in the simulations."],"forward_implications":["If the paper is right, BINGO and LSST data can jointly detect the 21 cm signal in cross-correlation with significance comparable to BINGO's autocorrelation for the three central redshift bins.","The full photometric bin preserves the cross-correlation amplitude, so photometric surveys need not be sliced into narrow redshift bins to be useful for intensity mapping.","The cross-correlation provides an unbiased constraint on the product $b_{\\rm HI}\\Omega_{\\rm HI} r$, with a 1$\\sigma$ uncertainty around 20% of the theoretical value at the central bins, whereas the autocorrelation estimate drifts at higher redshift.","Galaxies in the photometric bin that do not overlap the HI bin act like extra noise, so their contribution explains why cross- and auto-correlation end up with similar detection significance.","Because contaminating signals in the two datasets are uncorrelated, the cross-correlation keeps a nearly scale-independent contamination factor even where the autocorrelation is badly affected at $\\ell\\gtrsim300$."],"supporting_citations":[{"why":"establishes the GNILC foreground-cleaning procedure for BINGO-like simulated maps that the analysis inherits.","marker":"[52]"},{"why":"sets the BINGO simulation methodology and fiducial cosmological model used for the HI realizations.","marker":"[53]"},{"why":"supplies the LSST photometric redshift selection function, photo-z errors, and galaxy bias for the ten-year survey.","marker":"[55]"},{"why":"provides the previous cross-correlation analysis with photometric galaxies whose redshift-matched selection this paper contrasts with the full-bin approach.","marker":"[39]"},{"why":"demonstrates the cross-correlation power-spectrum detection and the amplitude-fitting method for $b_{\\rm HI}\\Omega_{\\rm HI} r$ that the parameter estimation follows.","marker":"[26]"},{"why":"shows that GNILC extracts the HI cosmological signal from 21 cm intensity maps, supporting the cleaning step.","marker":"[15]"},{"why":"provides the lognormal simulation method used to generate correlated tomographic realizations of the HI and galaxy fields.","marker":"[59]"},{"why":"supplies the angular power spectrum modeling code and projection kernels used to build the theoretical input spectra.","marker":"[44]"}],"fun_headline_variants":["Photo-z blur can't hide BINGO 21 cm","BINGO 21 cm detectable despite LSST photo-z scatter","Cross-correlation rescues BINGO 21 cm from photo-z noise","BINGO's 21 cm signal withstands photo-z errors","Photo-z errors don't erase BINGO 21 cm detection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The detection significance is computed with a covariance matrix built from fast simulations that add noise and foreground residuals to the HI signal, not from the actual cleaned maps; the paper itself notes these fast simulations are not representative of the cleaned maps, so the reported significances may be biased.","fun_headline_variants_meta":{"raw":{"variants":["Photo-z blur can't hide BINGO 21 cm","BINGO 21 cm detectable despite LSST photo-z scatter","Cross-correlation rescues BINGO 21 cm from photo-z noise","BINGO's 21 cm signal withstands photo-z errors","Photo-z errors don't erase BINGO 21 cm detection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.002015,"raw_usage":{"total_tokens":7858,"prompt_tokens":950,"completion_tokens":6908,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":566,"completion_tokens_details":{"reasoning_tokens":6816}},"tokens_in":566,"tokens_out":6908,"duration_ms":46830,"temperature":1.0,"reasoning_tokens":6816,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:37:44.295221+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the 50 cleaned realizations, estimate the covariance directly from their measured power spectra (or generate enough cleaned maps to do so), and recompute $\\sqrt{\\Delta\\chi^2}$ in the same multipole range; if the significance drops below the detection threshold, the detectability claim fails. Alternatively, apply the same pipeline to real BINGO and LSST data and check whether the measured cross-power spectrum amplitude is consistent with the model at the claimed significance.","supporting_citations":[{"cited_title":"Mericia, L.C","cited_arxiv_id":null,"evidence_quote":"establishes the GNILC foreground-cleaning procedure for BINGO-like simulated maps that the analysis inherits."},{"cited_title":"Novaes, J","cited_arxiv_id":null,"evidence_quote":"sets the BINGO simulation methodology and fiducial cosmological model used for the HI realizations."},{"cited_title":"Zhang, C","cited_arxiv_id":null,"evidence_quote":"supplies the LSST photometric redshift selection function, photo-z errors, and galaxy bias for the ten-year survey."},{"cited_title":"Cunnington, L","cited_arxiv_id":null,"evidence_quote":"provides the previous cross-correlation analysis with photometric galaxies whose redshift-matched selection this paper contrasts with the full-bin approach."},{"cited_title":"Cunnington, Y","cited_arxiv_id":null,"evidence_quote":"demonstrates the cross-correlation power-spectrum detection and the amplitude-fitting method for $b_{\\rm HI}\\Omega_{\\rm HI} r$ that the parameter estimation follows."},{"cited_title":"Olivari, M","cited_arxiv_id":null,"evidence_quote":"shows that GNILC extracts the HI cosmological signal from 21 cm intensity maps, supporting the cleaning step."},{"cited_title":"Loureiro, B","cited_arxiv_id":null,"evidence_quote":"supplies the angular power spectrum modeling code and projection kernels used to build the theoretical input spectra."}],"review_version":2}