{"id":"cd90e4ed-216c-4ddf-8bea-2fcd8044eb2a","arxiv_id":"2412.12302","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"C IV absorption in the circumgalactic medium is detected in 72% of star-forming L* galaxies versus 23% of passive galaxies, a dichotomy at roughly 99% confidence.","lead":"This paper reports new Hubble Space Telescope measurements of C IV gas around Milky Way-mass galaxies and combines them with archival data, finding that star-forming galaxies host detectable C IV in their surrounding gas about three times more often than passive galaxies. The result strengthens the view that ongoing star formation leaves a clear chemical and physical imprint on the diffuse gas reservoir around galaxies.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The C IV/sSFR dichotomy is not corrected for the strong radial dependence of C IV; without a radius-matched test, the 72% vs 23% comparison may be confounded by differing Rproj/R200c distributions.","rationale":"The reader's verdict of CONDITIONAL is appropriate, and the reader correctly flags that the headline 'over 99% confidence' depends on an Anderson-Darling test that ignores censoring, while interval-censored tests give p = 0.017 and p = 0.034. That concern is real but secondary: even treating the four excluded star-forming upper limits as non-detections leaves a 64% versus 23% gap. The more load-bearing concern is that the central SF-versus-passive comparison is never adjusted for Rproj/R200c, a quantity the paper itself shows is strongly correlated with C IV detection fraction. Without a radius-matched test, the observed dichotomy could be an artifact of different impact-parameter distributions rather than a genuine sSFR effect. This concern is concrete and testable with the published sample, so it does not require rejection, but it does require a revised analysis before the discovery claim can be accepted. The reader's weakest assumption and my concern are related but distinct, hence 'partial' agreement. I keep the verdict as CONDITIONAL because the requested radial control could either confirm or refute the claim; the paper as written does not yet provide that evidence.","tokens_in":15653,"tokens_out":10795,"duration_ms":107768,"concrete_test":"Restrict the combined 46-sightline sample to a narrow normalized impact-parameter window (e.g., Rproj/R200c = 0.3-0.8) and recompute the star-forming and passive detection fractions above log N_CIV = 13.5. Alternatively, fit a logistic regression of C IV detection (log N_CIV > 13.5) on log sSFR with Rproj/R200c and log Mstar as covariates. If the sSFR coefficient is not significant at p < 0.05 after including Rproj/R200c, the claimed dichotomy is not established. Also report the Rproj/R200c distributions for the two sSFR groups to show whether they overlap.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1 compares C IV detection fractions between star-forming and passive galaxies without controlling for Rproj/R200c, even though the paper's own Figure 2 shows a steep radial decline in C IV detection fraction (from roughly 70% inside 0.5 R200c to below 10% beyond 1 R200c). The combined sample is restricted to log Mstar >= 9.5, but no radial matching, stratifying, or covariate adjustment is presented for the sSFR comparison. If the passive galaxies in the combined sample preferentially lie at larger normalized impact parameters, their lower detection fraction could reflect the radial gradient rather than a true dependence on star formation. This is more load-bearing than the upper-limit censoring issue identified by the reader: recasting the four excluded star-forming upper limits as non-detections lowers the star-forming fraction from 72% to 64%, still leaving a substantial gap, whereas an unaccounted radial confound could in principle remove the gap entirely. The CIViL* sample design controlled Mstar and Rproj matching for target selection, but the final analysis in Section 3.1 does not enforce such matching for the combined sample. Thus the central claim that C IV absorption is tied to recent star formation lacks the necessary control for the strongest known covariate in these data.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"Garza et al. present CIViL*, a new HST/COS program targeting C IV in the CGM of L* galaxies, and combine their 11 new sightlines with archival data from COS-Halos, COS-Dwarfs, and COS-Holes. After restricting to log10 M*/M_sun >= 9.5, they compare C IV detection fractions above log10 N_CIV = 13.5 between star-forming (sSFR > 1e-11 yr^-1) and passive galaxies, finding 72% (21/29) versus 23% (3/13). They claim a dichotomy at over 99% confidence using an Anderson-Darling test, with supporting interval-censored tests at p = 0.017 and p = 0.034. The paper also derives a minimum CGM carbon mass of about 3.03e6 M_sun within 120 kpc and interprets the results as evidence that C IV, like O VI, traces warm gas associated with recent star formation in the halos of L* galaxies.","tokens_in":15827,"tokens_out":6155,"duration_ms":56037,"significance":"If the central claim holds, this is a valuable extension of the well-known O VI dichotomy to C IV in L* galaxies, strengthening the picture that the warm, highly ionized CGM is coupled to recent star formation. The paper's strengths include new public HST/COS data that substantially increase C IV coverage in this regime, the use of Wilson binomial confidence intervals, and a genuinely useful appendix of censored-data tests and threshold variations. The main limitations are the small sample size (46 sightlines after the mass cut, including only 13 passive galaxies) and the fact that the central star-forming/passive comparison has not been controlled for the strong radial gradient in C IV detection fraction that the paper itself demonstrates.","major_comments":[{"comment":"The central detection-fraction comparison between star-forming and passive galaxies does not control for Rproj/R200c. The paper's own Figure 2 (top panel) and Table A.1 show a steep radial decline in C IV detection fraction, from roughly 70% inside 0.5 R200c to below 10% beyond 1 R200c. Because the combined sample is assembled from surveys with different selection functions and the star-forming and passive subsamples are not shown to have matching Rproj/R200c distributions, the observed 72% versus 23% difference could partly or wholly reflect radial selection rather than a physical dependence on sSFR. A stratified comparison within radial bins, or a regression of detection status on both sSFR and Rproj/R200c, is needed to support the dichotomy claim. This is the most load-bearing issue in the paper.","section":"Section 3.1 and Figure 2"},{"comment":"The headline claim of 'over 99% confidence' rests on an Anderson-Darling test that treats upper and lower limits as detections, a point the paper itself acknowledges. The interval-censored tests in Appendix D give p = 0.017 (one-sided log-rank) and p = 0.034 (two-sided k-sample test), corresponding to roughly 2.1-2.4 sigma. The abstract and summary should either present the censored-data significance as the primary evidence or explicitly state that the >99% value comes from a test that ignores the censoring structure. Reporting only the limit-naive Anderson-Darling p-value overstates the statistical confidence in the result.","section":"Section 3.1, Appendix D, Abstract, and Section 5"},{"comment":"The treatment of upper limits above the detection threshold is asymmetric: four star-forming upper limits are excluded from the denominator while no passive ones are. If those four upper limits are reclassified as non-detections, the star-forming detection fraction drops from 21/29 = 72% to 21/33 = 64%, still higher than the passive 23% but with a larger uncertainty. The paper should present this sensitivity explicitly in the main text and clearly define, for each statistical test, whether the excluded upper limits are counted as detections, non-detections, or omitted.","section":"Section 3.1 and Figure 3 caption"}],"minor_comments":[{"comment":"The sample size is reported inconsistently: the abstract says 46 observations, Section 2.4 says the combined sample has 65 observations before cuts, and Section 5 says the final sample has 45 lines of sight. These numbers should be reconciled with a single consistent accounting of the mass cut and duplicates.","section":"Title and Section 2.4"},{"comment":"In the equal-number radial bins, the last bin is labeled '1.0-2.0' in the first column but the row appears to be missing a closing parenthesis in the confidence interval (the entry reads '0.77' rather than '(0.25, 0.75)'). Also, the 8-bin table lists a bin '1.75-1.0' that should presumably be '1.75-2.0'.","section":"Table A.1"},{"comment":"The 'grey area' classification of five galaxies uses a combination of sSFR, optical spectra, and morphology, but the criteria are not fully quantitative. It would be helpful to state explicitly how many galaxies were reclassified relative to their sSFR bins and whether the main result is robust to moving those borderline objects to the opposite class.","section":"Appendix B"},{"comment":"The carbon mass estimate is based on the Bordoloi et al. (2014) formula with an assumed ionization fraction f_CIV = 0.3. The paper should clarify whether this is intended as a strict lower limit given the possibility of lower f_CIV values, and the role of saturated lower limits in the mean column densities should be explicitly discussed.","section":"Section 3.2 and Equation (1)"},{"comment":"The statement that C IV and O VI 'statistically mirror' the dichotomy is descriptive rather than statistical. No formal comparison of the C IV and O VI samples is presented, so the language should be softened accordingly.","section":"Figure 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and the underlying dataset is valuable. The main issue is the lack of a radius-controlled comparison for the central dichotomy claim; this is fixable with a stratified analysis or a regression that includes Rproj/R200c. I would be comfortable with acceptance after that analysis is added and the censoring-based significance is reported accurately. The count inconsistencies and small typographical issues should also be cleaned up."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful dataset and a probably-true result, but the paper oversells its own statistics and never addresses the radial gradient it displays in Figure 2. The C IV dichotomy between star-forming and passive L* galaxies is likely real — the Wilson intervals (72% vs 23%) and the interval-censored tests (p = 0.017, 0.034) all point the same way — but the “>99% confidence” headline comes from an Anderson-Darling test that pretends upper limits are detections. That is the wrong test for a headline claim. The censored-data tests give ~2–2.5 sigma, which is suggestive, not definitive.\n\nThe real issue is the radial confound. Their own Figure 2 shows C IV detection fraction dropping from ~70% inside 0.5 R200c to below 10% beyond 1 R200c. Section 3.1 compares star-forming and passive galaxies with no control for Rproj/R200c. If the passive galaxies in the combined sample sit at larger normalized radii — which is plausible given the mixed archival samples — the gap between 72% and 23% could shrink or disappear. The paper needs a radius-matched comparison, or at least a regression that includes Rproj/R200c as a covariate. Without that, the central claim is not as clean as the abstract suggests.\n\nAlso, the four star-forming upper limits above threshold are excluded while no passive ones are; recasting them as non-detections changes the star-forming fraction from 72% to 64%, which weakens but does not erase the gap. That is a smaller problem than the radial confound but still worth a caveat.\n\nThe new survey itself is valuable: 11 new C IV sightlines, carefully reduced, with clear upper/lower limits, and the combination with archival data more than doubles available C IV constraints around L* galaxies. That alone deserves publication. The carbon mass estimate is a reasonable order-of-magnitude exercise but inherits the assumed f_CIV = 0.3 and saturated-column uncertainties, so treat it as secondary.\n\nWho should read this: anyone working on the CGM of L* galaxies or ionized gas in galaxy halos. It is a solid data paper with an interpretable result, and the flaws are addressable. I would send it to a serious referee, with the clear instruction that radial control and honest confidence statements are required before acceptance.","headline":"A genuinely useful C IV dataset and a probably-true dichotomy, but the headline confidence is overstated and the analysis never controls for the radial gradient shown in its own Figure 2.","tokens_in":16492,"tokens_out":2485,"would_cite":true,"duration_ms":22609,"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":"C IV absorption in galaxy halos splits sharply by star formation state.","keywords":["circumgalactic medium","C IV absorption","L* galaxies","star formation","HST/COS","quasar absorption spectroscopy","galaxy halos","O VI dichotomy"],"falsifier":"Deep COS spectra of the four star-forming sightlines whose current upper limits exceed $\\log N_{\\rm CIV}/{\\rm cm}^{-2} = 13.5$ would settle whether the 72% fraction is real: if all four turn out to be non-detections, the star-forming fraction drops to 64% and the gap to the passive 23% shrinks though it does not vanish.","tokens_in":15377,"feed_emoji":"🌌","tokens_out":4760,"duration_ms":40241,"temperature":0.7,"pith_summary":"This paper reports that C IV absorption in the gas halos of typical-mass (L*) galaxies splits sharply according to whether the galaxy is actively forming stars. Combining 11 new HST/COS sightlines with archival measurements, the authors find C IV above a column density of $10^{13.5}$ cm$^{-2}$ in 72% of star-forming galaxies but only 23% of passive galaxies, a difference they place above 99% confidence. The result matters because C IV sits between well-studied low-ionization and high-ionization tracers, and if the dichotomy is real it means this intermediate-ionization gas is tied to recent or ongoing star formation, matching the known O VI dichotomy. The paper also derives a minimum carbon mass in the CGM of roughly $3 \\times 10^6$ $M_\\odot$ within 120 kpc, indicating a substantial carbon reservoir in these halos.","feed_headline":"Star-forming galaxies show C IV in their halos 3x more often","feed_subtitle":"New HST survey finds C IV absorption in 72% of star-forming L* halos versus 23% of passive ones.","key_machinery":"The load-bearing object is the C IV doublet ($\\lambda\\lambda1548, 1550$ \\AA) seen in absorption against background quasars, which reports the column density of triply ionized carbon along each sightline. The argument runs on detection fractions above a fixed threshold ($\\log N_{\\rm CIV}/{\\rm cm}^{-2} = 13.5$), computed with 2$\\sigma$ Wilson binomial confidence intervals, and on comparing the star-forming and passive samples with Anderson-Darling and interval-censored survival tests. The survey design adds NUV COS coverage to galaxies with existing O VI data, so each C IV measurement sits on a galaxy with known stellar mass, impact parameter, and specific star formation rate.","core_discovery":"The central discovery claim is that the circumgalactic medium of L* galaxies contains a C IV dichotomy: star-forming galaxies (sSFR $> 10^{-11}$ yr$^{-1}$) show C IV absorption above $\\log N_{\\rm CIV}/{\\rm cm}^{-2} = 13.5$ in $72^{+14}_{-18}\\%$ [21/29] of sightlines, while passive galaxies show it in only $23^{+27}_{-15}\\%$ [3/13]. The authors reject the null hypothesis that the two column-density distributions come from the same parent population, at $>99.5\\%$ confidence by an Anderson-Darling test ($p = 0.0016$) and at $>2\\sigma$ by interval-censored survival tests ($p = 0.017$ and $p = 0.034$). They interpret this as C IV behaving like O VI rather than like low-ionization gas: it traces warm, metal-enriched gas present when galaxies are actively forming stars and largely absent around passive galaxies. A minimum carbon mass of about $3 \\times 10^6$ $M_\\odot$ out to 120 kpc follows from the mean column densities in three radial bins.","pith_inferences":["If the dichotomy holds in a larger sample, C IV could serve as a practical alternative or complement to O VI for mapping warm gas around galaxies, since C IV is visible in the NUV where COS is efficient.","The sharpness of the split suggests a physical connection between star-formation-driven outflows and the presence of warm ionized carbon; simulations of galaxy halos should be tested against the 72% versus 23% numbers rather than only against O VI.","Five galaxies near the sSFR cutoff were classified by eye from spectra and morphology, so a blind or automated classification on a larger sample would test whether the dichotomy is as sharp as it appears.","Extending the same C IV measurement to higher redshift or to lower-mass galaxies could reveal whether this dichotomy is a universal feature of galaxy halos or specific to L* galaxies at $z \\lesssim 0.25$."],"forward_implications":["C IV joins O VI as a tracer of star-formation-linked gas in galactic halos, so intermediate-ionization carbon can be used to track feedback and accretion in the circumgalactic medium.","The minimum carbon mass of about $3 \\times 10^6$ $M_\\odot$ within 120 kpc means the CGM of L* galaxies holds a carbon reservoir comparable to the galaxies' own interstellar medium, not a negligible halo component.","The lower-limit CGM gas mass and the depletion time of roughly 1.9 Gyr imply that galaxies need ongoing gas resupply to sustain star formation, since the CGM alone cannot fuel them for more than a couple of gigayears.","The dichotomy gives a new observational handle on galaxy transition from the blue cloud to the red sequence: the warm C IV-bearing gas disappears as star formation shuts off."],"supporting_citations":[{"why":"Supplies the O VI dichotomy in the CGM of L* galaxies that this paper's C IV result mirrors, and provides the comparison figure.","marker":"Tumlinson et al. (2011)"},{"why":"Source of COS-Halos galaxies and archival C IV and O VI measurements plus galaxy properties used in the combined sample.","marker":"Werk et al. (2013)"},{"why":"Provides COS-Dwarfs C IV observations and the carbon-mass estimation method used in Section 3.2.","marker":"Bordoloi et al. (2014)"},{"why":"Adds COS-Holes C IV observations to the combined sample and reports evidence on C IV versus supermassive black hole mass.","marker":"Garza et al. (2024)"},{"why":"Supplies the mass-controlled O VI dichotomy analysis and motivates the sSFR and stellar-mass cuts used here.","marker":"Tchernyshyov et al. (2023)"},{"why":"Contributes COS-GASS galaxies included in the CIViL* sample and their CGM absorption measurements.","marker":"Borthakur et al. (2015)"}],"fun_headline_variants":["C IV in halos: star-forming galaxies show 3x more absorption","HST/COS finds C IV dichotomy between active and passive L* galaxies","72% vs 23%: C IV detection split by star formation in L* halos","Warm carbon gas 3x more prevalent around star-forming galaxies","New survey: C IV halo gas tracks star formation in L* galaxies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison assumes that excluding upper limits that sit above the detection threshold does not bias the result: four star-forming sightlines are thrown out while no passive ones are, and if those four were true non-detections the star-forming fraction would fall from 72% to 64%.","fun_headline_variants_meta":{"raw":{"variants":["C IV in halos: star-forming galaxies show 3x more absorption","HST/COS finds C IV dichotomy between active and passive L* galaxies","72% vs 23%: C IV detection split by star formation in L* halos","Warm carbon gas 3x more prevalent around star-forming galaxies","New survey: C IV halo gas tracks star formation in L* galaxies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000258,"raw_usage":{"total_tokens":1651,"prompt_tokens":1084,"completion_tokens":567,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":700,"completion_tokens_details":{"reasoning_tokens":465}},"tokens_in":700,"tokens_out":567,"duration_ms":5268,"temperature":1.0,"reasoning_tokens":465,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T14:14:08.418340+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Deep COS spectra of the four star-forming sightlines whose current upper limits exceed $\\log N_{\\rm CIV}/{\\rm cm}^{-2} = 13.5$ would settle whether the 72% fraction is real: if all four turn out to be non-detections, the star-forming fraction drops to 64% and the gap to the passive 23% shrinks though it does not vanish.","supporting_citations":[],"review_version":1}