{"id":"87213241-731e-449f-b70e-d32ada9ef22b","arxiv_id":"2502.00110","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Combined MIGHTEE and CHILES stacking yields log MHI = (0.32 +/- 0.04) log M* + (6.65 +/- 0.36) at z ~ 0.36, placing gas content between the z=0 and z=1 relations.","lead":"By stacking faint 21 cm radio signals from thousands of distant galaxies, this paper measures the atomic hydrogen content of typical star-forming galaxies at redshift 0.36. Combining two independent radio surveys, it finds that galaxies hold more atomic gas than local galaxies with the same stellar mass, a benchmark for gas evolution ahead of the SKA era.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table 1 and Eq. 4 are internally inconsistent: the 10% confusion correction is not applied to the highest-mass bin, and the printed corrected masses do not reproduce the quoted fit.","rationale":"The reader's weakest assumption concerns the sensitivity of the result to the updated spec-z catalog. That is a valid external robustness concern, but the most load-bearing issue I find is internal and more easily settled: the paper's own Table 1 is inconsistent with its headline Eq. 4. The first three rows of MHI,corr show exactly a 10% reduction from MHI, while the fourth row does not; a fit to the printed column does not return the quoted parameters. This directly affects the central measurement, not just its future stability. If the table is corrected to 11.84, Eq. 4 is reproducible, so the concern may be a simple typo; but as published, a reader cannot verify the main result from the data table. Related numerical inconsistencies support the need for a check: the text's '~60% higher than z=0' conflicts with Table 2 (9.83 vs 9.55 is 0.28 dex, about 90%), and the quoted evolution index 1.99±0.13 is not what a three-point fit to the Table 2 values gives (about 1.77). None of these necessarily overturn the measurement, but they lower confidence in the headline numbers until resolved. The reader's CONDITIONAL verdict remains appropriate; my concern adds a specific, testable condition rather than changing the verdict.","tokens_in":27106,"tokens_out":18287,"duration_ms":170545,"concrete_test":"Recompute the parametric bootstrap least-squares fit using Table 1's MHI,corr column exactly as printed. If the recovered slope and intercept do not match Eq. 4, replace the last-row corrected mass with 13.15×0.9=11.84 and refit. If only the second fit matches Eq. 4, the table contains a typo and the fit used corrected masses; if neither matches, the published relation is not supported by the table and the paper must clarify which masses were fit.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Table 1 reports MHI,corr = 3.26, 4.75, 8.64, and 12.84 for the four bins. The first three are the uncorrected MHI values reduced by 10% (3.62→3.26, 5.28→4.75, 9.61→8.64), but 13.15→12.84 is only a 2.4% reduction; the stated 10% confusion correction (Section 3.2) would give 11.84. A weighted least-squares fit to the printed MHI,corr column yields a slope of about 0.35 and intercept about 6.34, not Eq. 4 (0.32±0.04, 6.65±0.36). If the last value is corrected to 11.84, the fit reproduces Eq. 4, suggesting a typo. As published, the central relation cannot be reproduced from the paper's own final masses, and it is unclear whether Eq. 4 was fit to corrected or uncorrected masses. This is a concrete, checkable internal inconsistency in the core measurement.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper combines MIGHTEE-HI Early Science and CHILES 21-cm data in the COSMOS field to perform spectral-line stacking of star-forming galaxies at mean redshift <z> = 0.36. The authors split the sample into four stellar mass bins, apply RFI masking, a confusion correction, and a spectroscopic redshift quality cut, and derive a best-fit scaling relation log10(M_HI/Msun) = (0.32 +/- 0.04) log10(M*/Msun) + (6.65 +/- 0.36). They compare this relation with z ~ 0 and z ~ 1 results, infer an evolutionary index M_HI proportional to (1+z)^1.99 at fixed stellar mass, and discuss implications for the baryon cycle and H I-to-H2 conversion. The paper includes extensive validation: noise scaling with N, Gaussianity tests, cross-survey consistency checks, and an appendix isolating the origin of the difference with the earlier MIGHTEE-only result of S22.","tokens_in":27235,"tokens_out":6934,"duration_ms":71890,"significance":"If the central measurement is correct, this is the most statistically robust M_HI-M* relation at z ~ 0.36 to date, based on four stacks with S/N > 5 and on two independent surveys that agree within 1.5 sigma in all bins. The technical validation is a genuine strength: Figure 5 verifies the expected 1/sqrt(N) noise scaling, Appendix A.1 tests Gaussianity and applies a conservative outlier cut, and Appendix A.2 demonstrates consistency between MIGHTEE and CHILES. The main caveats are that the printed Table 1 does not reproduce the quoted Eq. (4), and that the evolutionary-index claim in Section 4.3 and Figure 11 is partly circular because it uses the paper's own z = 0.36 point and then rescales a local relation with that same index. After correcting the table/equation inconsistency, the paper would be a valuable reference measurement for the redshift evolution of the atomic gas content of star-forming galaxies.","major_comments":[{"comment":"The confusion correction is applied inconsistently to the highest-mass bin. Section 3.2 states that a 10% confusion correction is applied, and the first three entries in Table 1 are indeed the uncorrected M_HI values reduced by 10% (3.62 -> 3.26, 5.28 -> 4.75, 9.61 -> 8.64). However, the fourth entry is 13.15 -> 12.84, which is only a 2.4% reduction; a 10% reduction would give 11.84. A least-squares fit to the printed corrected masses gives a slope of about 0.35 and an intercept of about 6.34, not the quoted Eq. (4) values (0.32 +/- 0.04, 6.65 +/- 0.36). If the fourth bin is corrected to 11.84, the fit reproduces Eq. (4). As published, the central relation cannot be reproduced from the paper's own final masses, and it is not stated whether Eq. (4) was fit to corrected or uncorrected masses. This must be fixed and clarified.","section":"Table 1, Eq. (4), Section 3.2"},{"comment":"The evolutionary index 1.99 +/- 0.13 is derived by fitting a power law to three points: z = 0 from G21, z = 0.36 from this work, and z = 1 from C22. The z = 0.36 point is the paper's own measurement. The left panel of Figure 11 then rescales the G21 local relation using this same index and shows agreement with the same three datasets. This agreement is therefore partly built into the fit and does not constitute an independent validation of the evolutionary index. The index itself is a legitimate fit, but the text and Figure 11 should be reframed to make clear that this is a consistency check of the adopted power-law form, not an independent confirmation.","section":"Section 4.3, Figure 10, Figure 11"},{"comment":"The confusion correction is a load-bearing assumption for the normalization of the final relation. Section 3.2 adopts the ~10% contamination level derived from MeerKAT-like simulations in S22 and applies it unchanged to CHILES and to the combined stack, despite the different synthesized beam sizes (VLA ~7 arcsec versus MeerKAT ~17 arcsec in this configuration) and different spatial resolutions. The correction is global, so it does not affect the slope, but it directly sets the zero-point of Eq. (4). A sensitivity test varying the correction within a plausible range (e.g., 0-20%) should be reported so that the quoted normalization uncertainty reflects this assumption rather than only the spectral noise.","section":"Section 3.2, Appendix B"}],"minor_comments":[{"comment":"The caption says the local relation is rescaled with an evolutionary power-law index of 1.8, while the text and Figure 10 report 1.99 +/- 0.13. Please harmonize the numbers.","section":"Figure 11 caption"},{"comment":"The text in Section 2 says the MIGHTEE Early Science data cover a total area of ~5 deg^2, while Section 3.1 says MIGHTEE covers the full COSMOS field of ~2 deg^2. Please clarify which area applies to the Early Science data cubes used here.","section":"Section 2 and Section 3.1"},{"comment":"The definition of the integrated S/N in Eq. (3) is not fully transparent: it is written as a ratio between an integrated flux and a quantity involving N_ch and sigma, but the symbols are not defined precisely in the text. Please specify what N_ch is and how sigma is computed so that the formula can be evaluated directly.","section":"Eq. (3)"},{"comment":"The statement that, for a galaxy in the overlapping region, 'we will have two spectra, which we will treat as two separate, independent instances' overstates statistical independence: the galaxy is the same, so the source properties are correlated even if the instrumental noise is independent. 'Independent noise realizations' would be more accurate.","section":"Section 3.1"},{"comment":"The description of the KS test result as 'turned positive' is ambiguous: a p-value of 0.05 rejects Gaussianity at the usual 5% level, and after the 3-sigma cut the test does not reject. Reporting the actual p-values before and after the cut would be clearer.","section":"Appendix A.1"}],"recommendation":"major_revision","confidential_remarks":"The internal inconsistency between Table 1 and Eq. (4) is concrete and checkable; it appears to be a fixable typo rather than a deep methodological flaw, but as printed the paper's headline relation cannot be reproduced from its own table. I would ask the authors to correct the table/equation and to clarify whether Eq. (4) is based on corrected or uncorrected masses, and to reframe the Figure 11 comparison as a consistency check rather than an independent validation. The measurement itself is potentially important and the validation appendices are a real strength."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look: this is the first stacking result combining MIGHTEE and CHILES for the MHI-M* relation, with 6,598 spectra in four mass bins and S/N above 5 in every stack. That makes it the most statistically robust measurement at z~0.36 so far. The paper also does good housekeeping: it verifies the noise scales as N^{-1/2}, tests Gaussianity, applies RFI masking, and shows the two surveys agree within 1.5 sigma in every bin. Appendix B, where the authors dissect why their MIGHTEE-only result differs from their own S22 relation, is honest and informative. The updated spectroscopic catalog and a proper redshift quality cut are plausibly the driver, and the authors say so rather than hiding the discrepancy.\n\nThe stress-test note is right, and it matters. Table 1 lists corrected HI masses 3.26, 4.75, 8.64, and 12.84. The first three are consistent with a 10% confusion correction, but 13.15 -> 12.84 is only a 2.4% reduction; a 10% correction gives 11.84. Fitting the printed corrected column gives a slope around 0.35 and intercept around 6.34, not Eq. 4's 0.32 and 6.65. If you replace 12.84 with 11.84, the fit reproduces Eq. 4. So this is very likely a typo, but as published the central relation cannot be reproduced from the paper's own final masses. That needs a correction and a note.\n\nOther soft spots are minor but worth naming. The errors on the corrected masses do not include the 10% confusion systematic or the choice of the +/-350 km/s window; those enter only as a global correction. The evolutionary index 1.99 comes from a three-point power-law fit that includes this paper's own z=0.36 point, and Figure 11 rescales the local relation with that same index before comparing, so it is not an independent validation. The main data products are still in preparation, so independent replication is not possible yet.\n\nNone of this breaks the core measurement. The central stacked relation is carefully built and internally consistent once the typo is fixed. The paper deserves a serious referee, and if it were newly submitted I would send it out with a request to fix Table 1, propagate the confusion correction into the error budget, and make the three-point nature of the evolutionary fit explicit. For anyone working on HI scaling relations or SKA precursors, this is now the reference to cite at z~0.4.","headline":"The combined MIGHTEE+CHILES stacked relation at z~0.36 is a solid new reference, but Table 1 has an internal inconsistency that makes Eq. 4 unreproducible as printed.","tokens_in":28043,"tokens_out":2267,"would_cite":true,"duration_ms":22714,"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":"Stacking 6,598 spectra from two radio surveys yields a precise MHI–M* relation at z≈0.36, showing HI content grows roughly as (1+z)^2 while the slope stays constant.","keywords":["HI galaxies","21 cm line","spectral stacking","scaling relations","galaxy evolution","MIGHTEE survey","CHILES survey","COSMOS field"],"falsifier":"Run the identical stacking pipeline on the same data cubes but with spectroscopic redshifts from an independent, complete survey in COSMOS, and compare the highest-stellar-mass bin: if the recovered HI mass moves by more than ~0.2 dex, the claimed slope of 0.32 and the constant-slope conclusion would not hold.","tokens_in":26769,"feed_emoji":"📡","tokens_out":6126,"duration_ms":52494,"temperature":0.7,"pith_summary":"This paper tries to establish the most statistically powerful measurement to date of how atomic hydrogen mass relates to stellar mass in star-forming galaxies at redshift ~0.36. By stacking 21 cm spectra from two independent radio surveys, MIGHTEE and CHILES, it obtains (in log space) a slope of 0.32 ± 0.04 and a normalization of 6.65 ± 0.36. If correct, the relation shows that at fixed stellar mass galaxies at z ≈ 0.36 are roughly 60% richer in atomic gas than today's galaxies, and about 50% poorer than galaxies at z ≈ 1. The slope staying unchanged across redshift would imply that the processes governing atomic gas gain and loss do not depend strongly on stellar mass.","feed_headline":"Stacking 6,598 galaxies sharpens the HI–mass relation at z≈0.36","feed_subtitle":"Combining MIGHTEE and CHILES shows HI grows as (1+z)^2 while the slope stays flat from z=1 to today.","key_machinery":"The machinery is spectral-line stacking: extract a small 3D cubelet around each galaxy using its spectroscopic redshift, collapse to a spectrum, resample to a common 100 km/s velocity grid, weight each spectrum by the inverse noise, co-add, and integrate over ±350 km/s to get the mean HI mass in a stellar mass bin. Its two survey inputs are complementary: MIGHTEE is wide and shallower, CHILES is narrower and deeper, and their combination yields four mass bins with high signal-to-noise. A ~10% source-confusion correction is applied based on earlier simulations.","core_discovery":"The central claim is that the combined MIGHTEE+CHILES stacking yields log10(MHI/Msun) = (0.32 ± 0.04) log10(M*/Msun) + (6.65 ± 0.36) at mean redshift 0.36, from 6,598 coadded spectra in four stellar mass bins (S/N > 5 in each). The paper further claims that this relation has a slope statistically indistinguishable from the z ≈ 0 and z ≈ 1 relations, and that its normalization evolves as MHI ∝ (1+z)^(1.99 ± 0.13) at fixed stellar mass. The author would state this as the best-constrained HI–stellar mass relation at this redshift to date, superseding the earlier MIGHTEE-only result, with the improvement coming from a larger merged spectroscopic catalog, stricter redshift quality cuts, RFI masking, and the combination of two independent data sets.","pith_inferences":["If the slope is truly mass-independent, stacking in finer mass bins at higher redshift should continue to find parallel relations; a future measurement at z>0.5 with the same method would provide a direct test.","The claimed bottleneck in HI→H2 conversion could be tested by comparing resolved HI and CO maps at matched physical scales once SKA-era telescopes reach z~0.4.","The catalog-driven shift in slope and normalization implies that similar stacking results from other fields may be systematically sensitive to spectroscopic incompleteness at the high-mass end; independent spectroscopic campaigns would settle this.","Extending the same combined-stacking technique to other deep extragalactic fields would check whether the COSMOS-specific cosmic variance affects the normalization."],"forward_implications":["HI content at fixed stellar mass evolves as (1+z)^~2 between z=0 and z=1, meaning galaxies at z≈0.36 are intermediate between local and cosmic-noon values.","The slope of the MHI–M* relation is consistent across redshift, so stellar mass does not modulate HI gain/loss mechanisms over the last 8 Gyr.","Atomic gas grows or depletes more slowly than molecular gas and star formation (index ~1.99 vs ~3.6), suggesting a bottleneck in the HI→H2 conversion.","The updated MIGHTEE-only result supersedes the earlier S22 relation because the change is driven by the spectroscopic catalog, not by RFI masking or photometry.","This provides the strongest anchor at z~0.36 for calibrating galaxy simulations and semi-empirical models."],"supporting_citations":[{"why":"Supplies the stacking pipeline and the earlier MIGHTEE-only relation that this work updates and supersedes.","marker":"S22"},{"why":"Provides the z≈1 reference relation used to infer redshift evolution of HI content.","marker":"C22"},{"why":"Provides the z≈0 ALFALFA-based relation used as the local anchor for the evolutionary fit.","marker":"G21"},{"why":"GMRT stacking relation at z≈0.35 used as a comparison and tension point.","marker":"A. Bera et al. 2023a"},{"why":"Presents the MIGHTEE-HI Early Science cubes and noise properties used for the wide survey component.","marker":"N. Maddox et al. 2021"},{"why":"COSMOS2020 photometric catalog from which stellar masses, SFRs, and photometric redshifts are drawn.","marker":"J. R. Weaver et al. 2022"},{"why":"Updated merged spectroscopic catalog whose redshift quality cuts are the main driver of the new result.","marker":"A. Khostovan et al. 2025"},{"why":"Simulations that set the ~10% source-confusion correction applied to the stacked HI masses.","marker":"E. C. Elson et al. 2016"}],"fun_headline_variants":["HI–mass relation at z=0.36 from 6,598 stacked galaxies","Combining MIGHTEE+CHILES sharpens HI–mass slope at z≈0.36","HI–M* slope flat from z=0 to 1; normalization grows as (1+z)^2","Stacking 6,598 galaxies reveals HI richness at z=0.36 between local and z=1"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The updated merged spectroscopic catalog and the new redshift quality cuts fully remove the systematic that shifted the MIGHTEE-only stacking result relative to S22, so the corrected catalog is the final word on where each galaxy's line sits in the stack.","fun_headline_variants_meta":{"raw":{"variants":["HI–mass relation at z=0.36 from 6,598 stacked galaxies","Combining MIGHTEE+CHILES sharpens HI–mass slope at z≈0.36","HI–M* slope flat from z=0 to 1; normalization grows as (1+z)^2","Stacking 6,598 galaxies reveals HI richness at z=0.36 between local and z=1"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000484,"raw_usage":{"total_tokens":2425,"prompt_tokens":1017,"completion_tokens":1408,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":1304}},"tokens_in":633,"tokens_out":1408,"duration_ms":12427,"temperature":1.0,"reasoning_tokens":1304,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T20:11:57.439743+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the identical stacking pipeline on the same data cubes but with spectroscopic redshifts from an independent, complete survey in COSMOS, and compare the highest-stellar-mass bin: if the recovered HI mass moves by more than ~0.2 dex, the claimed slope of 0.32 and the constant-slope conclusion would not hold.","supporting_citations":[{"cited_title":"C., Blyth, S","cited_arxiv_id":null,"evidence_quote":"Simulations that set the ~10% source-confusion correction applied to the stacked HI masses."}],"review_version":1}