{"id":"a5733c20-cefd-4303-a19f-30f4e63af5cd","arxiv_id":"2412.08002","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"HI absorption detected in an untargeted ASKAP-FLASH sample is statistically over-represented towards young, compact (peaked-spectrum) radio AGN, with derived linear sizes below about 2 kpc.","lead":"A study of 33 neutral hydrogen absorbers from an untargeted radio survey finds that nearly two-thirds are young, peaked-spectrum AGN, far more than the 10-20% expected in the general radio population. If the correlation holds, future HI absorption surveys will be strongly biased toward catching radio galaxies in their first few thousand years of activity.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The p=3e-8 binomial excess in Sec. 2 assumes a literature PS fraction for the FLASH parent sample; if the true parent fraction at S_855.5 > 40 mJy is ~50%, the claimed over-representation would disappear.","rationale":"The reader's primary weakest assumption is exactly the unmeasured PS fraction in the FLASH parent sample, and my read identifies the same point as the most load-bearing weakness. I do not see a reason to strengthen the verdict to rejection: the claimed 21/33 excess is large, the comparison baseline is plausibly close to the literature value, and the missing control measurement is directly obtainable from the FLASH Pilot continuum catalogue. The paper is transparent about its preliminary ML classifications and the need for larger samples, and the main statistical signal remains suggestive even for moderately higher parent fractions. A conditional acceptance, with the parent-baseline measurement as a required check, is therefore the right call. My stress-test does not change the reader's verdict.","tokens_in":4337,"tokens_out":8394,"duration_ms":91889,"concrete_test":"Apply the same RADIO SED classifier used for the 33 HI absorbers to the FLASH Pilot continuum parent catalogue (or a flux-stratified random subsample large enough to measure the PS fraction to ~5%) and recompute Pr(N_PS >= 21 | f_PS_parent). Report f_PS_parent as a function of S_855.5, and also the mean flux density of the classified parent sources. If f_PS_parent is <=0.2 the p=3e-8 claim stands; if f_PS_parent is >~0.5 the headline 'over-representation' is not significant.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing assumption is in Section 2: the 21/33 PS fraction is compared with the O'Dea (1998) / Callingham et al. (2017) literature value of 10-20% for 'all radio sources', and this is used to compute Pr(N_PS >= 21) = 3e-8. But the FLASH Pilot continuum parent sample is defined at S_855.5 MHz > 40 mJy, and its PS fraction is never measured, even though the survey detected over 10^4 continuum sources. Selection frequency and flux limit matter: a flux-limited sample at 855 MHz need not have the same fraction of compact peaked-spectrum sources as the samples used to derive the literature baseline. If the true parent PS fraction were only 0.4, the binomial tail would drop from 3e-8 to roughly the per-mille level; if it were above about 0.5, the excess would no longer be statistically significant. The paper's central quantitative claim 'do not resemble the parent population' therefore rests on an unmeasured control population. The authors do not provide any check or ancillary evidence that the FLASH parent PS fraction is close to the literature value, so the strongest headline statement is not yet fully supported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports on the broadband radio spectral properties of 33 sources detected in HI absorption in the ASKAP-FLASH Pilot Surveys, an untargeted search. Using multi-frequency flux densities and the RADIO SED Bayesian fitting framework, the authors classify 21/33 sources as peaked-spectrum (PS) sources. They compare this fraction with a literature value of 10-20% for PS sources among all radio galaxies, compute a binomial probability of 3e-8, and conclude that HI-absorbing galaxies are a specific, young subset of the radio population. They further use the rest-frame peak frequencies to estimate linear sizes of 0.05-2.2 kpc and discuss an apparent 'resonance' between the absorbing screen size and the radio source size. The paper is a short conference proceedings contribution.","tokens_in":4575,"tokens_out":3262,"duration_ms":34908,"significance":"If the central claim is correct, the result is scientifically valuable: it would demonstrate that untargeted HI absorption surveys preferentially select young, compact, peaked-spectrum radio AGN rather than a representative sample of the radio galaxy population. This has consequences for interpreting absorption-selected samples and for planning future all-sky surveys such as the full FLASH survey. The paper also contributes a methodological template for combining broadband SED classification with HI absorption spectroscopy. The sample itself, drawn from an untargeted survey with a clearly described linefinder and significance threshold, is a useful resource. However, the headline statistical claim currently rests on an unmeasured control population, and several secondary inferences are partly circular or rely on preliminary classifications, so the strength of the conclusions exceeds what the present analysis supports.","major_comments":[{"comment":"The binomial calculation Pr(N_PS >= 21) = 3e-8 assumes that the peaked-spectrum fraction in the parent FLASH continuum sample at S_855.5 MHz > 40 mJy is 20%, taken from O'Dea (1998) and Callingham et al. (2017). The paper does not measure this fraction in the FLASH parent sample, even though the pilot surveys detected over 10^4 continuum sources. The literature baseline was derived at different frequencies, flux limits, and from heterogeneous samples, so it is not established to apply to the 855.5 MHz, 40 mJy flux-limited parent population. If the true parent PS fraction were substantially higher (e.g., ~50%), the claimed excess would disappear. The central statement that these absorbers 'do not resemble the parent population' therefore requires either a direct measurement of the parent PS fraction in the FLASH continuum catalogue or a substantially hedged interpretation. As written, the headline over-representation claim is not yet fully supported.","section":"Section 2"},{"comment":"The linear-size estimates are derived from the rest-frame peak frequency of each source using the Jeyakumar (2016) relation. These peak frequencies are the same measurements used to classify the sources as peaked-spectrum in Section 2, so the derived linear sizes are not an independent check of the claim that HI absorbers are preferentially compact; a source classified as PS will, by construction, tend to have a small derived size. The statement in Section 3 that the sample 'supports the idea of a resonance between the size of the absorbing screen and the radio source' is therefore partly circular. To make this argument convincing, the authors should either compare the derived sizes with a control sample of non-absorbing PS sources from the same survey or clearly state that the size distribution is a consequence of the selection and does not by itself confirm the resonance interpretation.","section":"Section 3"},{"comment":"The right panel of Figure 1 and the associated discussion separate sources into associated and intervening line-of-sight classifications based on 'preliminary' machine learning from Curran et al. (2016). The authors themselves flag these classifications as preliminary, yet the interpretation of the linear-size/linewidth relation, and the upper limits placed on intervening systems, depend on this split. Given that the sample spans only 33 sources and the classifications are uncertain, the conclusion that 'the width of the absorption line profile is not strongly correlated with the linear size' is not robustly supported. The analysis would benefit from a sensitivity check that treats the uncertain classifications as a source of systematic error, or from explicit statements of how misclassification would change the plotted trends.","section":"Section 3 / Figure 1"}],"minor_comments":[{"comment":"The running title 'Neutral Hydrogen in an around galaxies in the SKA era' appears to contain a typo; it should likely read 'in and around galaxies'.","section":"Header"},{"comment":"The text states '3,000 deg2 of sky'; the unit should be typeset as deg^2 or 'square degrees' for consistency.","section":"Section 1"},{"comment":"The description of the Bayesian linefinder says it 'identified highly significant absorption lines in each 288 MHz-wide spectrum'; it would be clearer to state that it identified lines in the spectra, not in each spectrum in a way that implies a line in every spectrum.","section":"Section 1"},{"comment":"The symbol 'S855.5 MHz' should be typeset with the frequency as a subscript (S_855.5 MHz) to avoid confusion with a multiplication.","section":"Section 2"},{"comment":"The phrase 'restframe peak' is used in the text and Figure 1; this should be 'rest-frame peak' for consistency with standard terminology.","section":"Section 3"},{"comment":"The sentence 'the location of the spectral peaks in these sources suggest that they all have linear sizes less than 2 kpc' has a subject-verb agreement issue ('location ... suggest' should be 'suggests').","section":"Section 3"}],"recommendation":"major_revision","confidential_remarks":"The paper is a compact conference proceedings contribution, not a full journal article, so the bar for completeness is lower. However, the central quantitative claim—the 3e-8 binomial excess—depends on an external literature fraction that may not describe the FLASH parent continuum sample. This is a load-bearing issue, not a cosmetic one. If the authors can measure or approximate the PS fraction in the FLASH parent catalogue, or if they can convincingly argue that the literature baseline applies (e.g., by quantifying how the PS fraction varies with frequency and flux density), the paper could be accepted after a moderate revision. The secondary points about circularity in the linear-size estimates and the preliminary machine-learning classifications also need to be addressed, but they are less central than the parent-population issue. I have no concerns about the novelty or the fit with the symposium scope; the topic is timely and the data are unique."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know about this paper is that it reports a real observational result — 21 of 33 HI absorbers from the untargeted FLASH pilot are classified as peaked-spectrum sources — and that the binomial probability of 3e-8 they quote is correct given their assumptions. But the assumption doing the work is that the parent FLASH continuum sample, defined at S_855.5 > 40 mJy, has the same 10–20% PS fraction as the literature samples. They never measure that fraction, even though the survey detected over 10^4 continuum sources. The stress-test note is right: if the true parent fraction were 40%, the p-value drops to per-mille; if it were above ~50%, the excess vanishes. That is a genuine soft spot in the paper's central claim, and it is not a minor one.\n\nWhat the paper does well: it is the first statistical statement of this effect in an untargeted sample, the SED fitting via RADIO SED is a legitimate tool, and the authors are transparent about their caveats — they flag the preliminary machine-learning classifications, the small sample, and the need for larger samples. The section on linear sizes is appropriately hedged, and the 'resonance' interpretation is presented as a suggestion, not a conclusion. For a proceedings paper, the level of honesty is good.\n\nThe soft spots are, in order: (1) the unmeasured parent baseline, which is the load-bearing assumption; (2) the derived linear sizes use the same peak-frequency measurements that define sources as young and compact, so there is some circularity in the size distribution claim; (3) the associated/intervening classification, which affects the size upper limits, is preliminary for most sources. None of these are fatal to the observation that something interesting is going on, but they do mean the headline 'do not resemble the parent population' is not yet fully supported.\n\nWho gets value from this: the AGN/HI absorption community, especially people planning FLASH and SKA surveys. It is a useful pointer that untargeted HI searches may be biased toward young, compact sources, but it needs a proper measurement of the parent PS fraction before that becomes a solid conclusion.\n\nMy recommendation: this deserves a serious referee, not a desk reject, because the claim is important and the data are real. But the referee should ask for the parent-sample baseline or a clear qualification that the 3e-8 is conditional on an assumed literature fraction. I would accept it with that condition.","headline":"A short, honest proceedings paper with a striking but load-bearing statistical claim: the 21/33 PS excess among HI absorbers is only as strong as the unmeasured parent-sample PS fraction.","tokens_in":5117,"tokens_out":1378,"would_cite":false,"duration_ms":16703,"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":"An untargeted HI absorption survey finds that 21 of its 33 detections are young, peaked-spectrum radio AGN, with chance odds of one in 33 million.","keywords":["radio AGN","peaked-spectrum sources","HI 21-cm absorption","ASKAP-FLASH survey","neutral hydrogen","young radio galaxies","broadband SED classification","intermediate-redshift gas"],"falsifier":"Compute the peaked-spectrum fraction for the full set of continuum sources in the same pilot fields above the survey's $S_{855.5\\,\\mathrm{MHz}} > 40$ mJy limit; if that parent fraction is already close to 60%, the 21-out-of-33 result is not an HI-specific over-representation. Alternatively, high-resolution radio imaging that resolves most of the 21 peaked sources into extended kiloparsec-scale structures would falsify the compact-source, sub-kiloparsec size interpretation.","tokens_in":4137,"feed_emoji":"📡","tokens_out":14071,"duration_ms":125584,"temperature":0.7,"pith_summary":"Using the 33 neutral-hydrogen (HI) 21-cm absorption systems found by the untargeted FLASH pilot surveys, this paper finds that 21 of them have broadband radio spectra which peak at low frequencies, the signature of young radio AGN triggered within roughly the last thousand years. Since peaked-spectrum sources are usually estimated to make up only 10-20% of the radio population, the observed fraction is statistically extreme: under the 20% baseline the probability of seeing 21 or more is about 3 in 100 million. The paper concludes that untargeted HI absorption does not see a representative sample of radio galaxies; it preferentially detects compact, young sources with estimated linear sizes between 0.05 and 2.2 kpc. If true, 21-cm absorption becomes a discovery tool for young AGN at intermediate redshifts, and the neutral gas must be arranged in screens whose sizes are matched to the compact radio emission.","feed_headline":"Untargeted HI survey finds 21 of 33 absorbers are young radio AGN","feed_subtitle":"The chance odds are 3 in 100 million, so 21-cm absorption appears to pick out compact, just-triggered radio jets.","key_machinery":"The argument is carried by the radio broadband spectral energy distribution of each AGN, assembled from flux-density measurements between 80 MHz and 100 GHz and modelled with RADIO SED, a Bayesian fitting framework designed to identify young, peaked-spectrum sources. The peaked-spectrum classification is the load-bearing step: a turnover in flux density over several decades of frequency marks a source as young (typically within about $10^3$ years of triggering) rather than old or blazar-variable. The paper then maps each rest-frame peak frequency to an estimated linear size through the empirical correlation between turnover frequency and source size, and uses the HI line's position relative to the SED peak to set lower or upper limits on that size depending on whether the absorber is classified as associated or intervening.","core_discovery":"The paper's central claim is that the galaxies against which neutral hydrogen is detected in an untargeted survey are a specific subset of the radio-AGN population: young, compact, peaked-spectrum sources. Among the 33 strongest HI absorbers from the FLASH pilot surveys, 21 statistically favour a peaked broadband model, whereas the literature fraction of peaked-spectrum sources is 10-20%; with a 20% assumed fraction the binomial probability of 21 or more peaked detections is $3\\times 10^{-8}$, and it falls further under the lower estimate. The authors further note that the rest-frame spectral peaks tend to sit just below the 1.4 GHz frequency of the HI line, and they convert peak frequencies into linear sizes of 0.05-2.2 kpc (mean $0.6\\pm0.5$ kpc) using the empirical peak-size relation. These small sizes match the compact-source preference seen in targeted HI samples and support the idea of a size resonance between the absorbing gas screen and the young radio source.","pith_inferences":["A direct test of the paper's baseline assumption would be to run the same SED classification on all continuum sources in the pilot fields above the survey's flux limit; if the parent peaked-spectrum fraction is already high, the apparent preference for young sources among HI absorbers would shrink or vanish.","The resonance interpretation predicts that high-resolution VLBI imaging of the 21 peaked sources will show compact double or compact-symmetric-object morphologies on sub-kiloparsec scales, and that HI absorption detection rates should peak for radio sources whose linear size matches the gas-screen scale.","If the effect is general rather than a FLASH selection effect, other untargeted 21-cm absorption surveys should show the same excess of peaked-spectrum sources once they accumulate comparable numbers of detections.","Since the sample mixes associated and intervening systems, the linear-size distribution is only firmly interpretable for the associated subset; secure optical redshifts will be required before the sizes can be compared with models of jet-gas interactions."],"forward_implications":["The full FLASH survey should confirm the effect: as the absorber sample grows to hundreds or thousands, the peaked-spectrum fraction should remain well above the 10-20% typical of the general radio population.","Untargeted 21-cm absorption searches can be used as a discovery channel for young, compact radio AGN at $0.4 < z < 1$, a redshift range where both Ly-alpha emission and 21-cm emission are hard or impossible to observe from the ground.","For associated absorbers, the HI line's position in the SED sets a lower limit on the rest-frame peak frequency and therefore an upper limit on the radio source's linear size, giving size information without VLBI imaging.","The lack of a strong correlation between HI line width and estimated linear size means line width is not governed simply by source size, so future samples with secure spectroscopic redshifts will be needed to separate geometry from gas kinematics.","The mean estimated linear size of about 0.6 kpc matches the compact-source preference seen in targeted HI samples, which the paper interprets as evidence that the size match between the gas screen and the radio source, rather than target selection, drives absorption detectability."],"supporting_citations":[{"why":"Supplies the 20% upper-end fraction of peaked-spectrum sources used as the baseline for the binomial probability.","marker":"O'Dea 1998"},{"why":"Supplies the lower 10% population fraction; using it makes the reported excess more significant.","marker":"Callingham et al. 2017"},{"why":"Describes the RADIO SED Bayesian framework used to classify each source's broadband spectrum as peaked or not.","marker":"Kerrison et al. 2024"},{"why":"Provides the Bayesian linefinder that selected the 33 strong HI absorption lines from the survey spectra.","marker":"Allison et al. 2012"},{"why":"Defines the two FLASH pilot surveys and the continuum sample above the 40 mJy, 855.5 MHz limit from which the absorbers were drawn.","marker":"Yoon et al. 2024"},{"why":"Describes the full FLASH survey whose pilot phase produced the detections analyzed here.","marker":"Allison et al. 2022"},{"why":"Provides the empirical correlation between rest-frame spectral peak and linear size used to estimate source sizes.","marker":"Jeyakumar 2016"},{"why":"Supplies the preliminary machine-learning classification of absorption lines as associated or intervening, used to split the sample.","marker":"Curran et al. 2016"},{"why":"Introduces the size-resonance idea between the absorbing screen and the radio source invoked to explain the young-source preference.","marker":"Curran et al. 2013"}],"fun_headline_variants":["HI absorption picks out young, compact radio AGN","Neutral gas traces recently triggered radio jets","21-cm gas reveals just-born radio sources","Untargeted survey: HI absorbers are young AGN"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the 10-20% literature fraction of peaked-spectrum sources is the correct baseline for the specific flux-limited, 855.5 MHz parent sample from which these absorbers were drawn; if compact sources are already over-represented in that parent sample, the reported excess disappears.","fun_headline_variants_meta":{"raw":{"variants":["HI absorption picks out young, compact radio AGN","Neutral gas traces recently triggered radio jets","21-cm gas reveals just-born radio sources","Untargeted survey: HI absorbers are young AGN"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1305,"prompt_tokens":857,"completion_tokens":448,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":473,"completion_tokens_details":{"reasoning_tokens":386}},"tokens_in":473,"tokens_out":448,"duration_ms":5224,"temperature":1.0,"reasoning_tokens":386,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:19:03.938904+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the peaked-spectrum fraction for the full set of continuum sources in the same pilot fields above the survey's $S_{855.5\\,\\mathrm{MHz}} > 40$ mJy limit; if that parent fraction is already close to 60%, the 21-out-of-33 result is not an HI-specific over-representation. Alternatively, high-resolution radio imaging that resolves most of the 21 peaked sources into extended kiloparsec-scale structures would falsify the compact-source, sub-kiloparsec size interpretation.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the 20% upper-end fraction of peaked-spectrum sources used as the baseline for the binomial probability."},{"cited_title":"R., Ekers, R","cited_arxiv_id":null,"evidence_quote":"Supplies the lower 10% population fraction; using it makes the reported excess more significant."},{"cited_title":"F., Allison, J","cited_arxiv_id":null,"evidence_quote":"Describes the RADIO SED Bayesian framework used to classify each source's broadband spectrum as peaked or not."},{"cited_title":"R., Sadler, E","cited_arxiv_id":null,"evidence_quote":"Provides the Bayesian linefinder that selected the 33 strong HI absorption lines from the survey spectra."},{"cited_title":"R., Sadler, E","cited_arxiv_id":null,"evidence_quote":"Describes the full FLASH survey whose pilot phase produced the detections analyzed here."},{"cited_title":"2016, Monthly Notices of the Royal Astronomical Society , 458(4), 3786--3794","cited_arxiv_id":null,"evidence_quote":"Provides the empirical correlation between rest-frame spectral peak and linear size used to estimate source sizes."},{"cited_title":"J., Duchesne, S","cited_arxiv_id":null,"evidence_quote":"Supplies the preliminary machine-learning classification of absorption lines as associated or intervening, used to split the sample."},{"cited_title":"J., Allison, J","cited_arxiv_id":null,"evidence_quote":"Introduces the size-resonance idea between the absorbing screen and the radio source invoked to explain the young-source preference."}],"review_version":1}