{"id":"7789d3c9-6cb2-4c96-9159-b3acb6acfb69","arxiv_id":"2507.12046","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For 104 spectroscopically confirmed quasars at radio flux densities near 3 microjansky, the fraction with AGN-dominated radio emission ranges from about 11 to 58 per cent depending on the star-formation-rate calibration used.","lead":"Using deep MeerKAT radio images, this paper measures how much of the faint radio light from 104 distant quasars comes from black-hole activity versus star formation. It finds the answer depends heavily on which star-formation calibration is used, with a redshift-dependent relation producing more starburst classifications.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Delhaize-SFR error bars omit the IRRC intrinsic scatter, so the claimed Yun-vs-Delhaize contrast in AGN/starburst fractions may be an artifact of over-tight thresholds.","rationale":"The reader's weakest assumption concerns whether the Delhaize et al. (2017) IRRC, calibrated on normal star-forming galaxies, applies to quasar hosts and starbursts out to z~3.4. That is a legitimate external-calibration concern, but by itself it does not threaten the paper's central claim: a demonstration that the derived AGN/SF fractions change with the adopted radio-SFR relation remains valid even if one relation is inappropriate. The more damaging, internal issue is the error budget used to make the comparison quantitative. The paper derives Delhaize-SFR from the IRRC using only measurement errors (Section 4.3) and then counts sources more than 1-sigma_1:1 from equality as AGN-dominated or starbursts. The IRRC intrinsic scatter is not included, even though it dominates the per-source uncertainty for a source assumed to lie on the relation. With sigma_qIR ~0.34 dex, the thresholds in Figures 7–8 and Tables 4–5 are much too tight; many classifications will flip. The qualitative direction (Yun gives more AGN-dominated sources than Delhaize, especially at high redshift) follows from the different normalizations, but the headline numbers, especially the 11–20% versus 22–58% contrast and the 63% high-z starburst fraction, may not be statistically meaningful until scatter is incorporated. The proposed test settles this directly: if the contrast survives the inclusion of scatter, the central claim stands but needs more cautious wording; if it collapses, the central claim is overstated. Either way the appropriate verdict remains CONDITIONAL, so the reader's verdict is unchanged.","tokens_in":53088,"tokens_out":7309,"duration_ms":90001,"concrete_test":"Recompute the Delhaize-SFR uncertainties for the Table D1 sources by adding the Delhaize et al. (2017) intrinsic q_IR scatter (sigma ~0.34 dex; also test 0.2 and 0.5 dex) in quadrature to the propagated radio and redshift errors, then re-derive Tables 4 and 5 with the same >1-sigma_1:1 threshold. If the 3-sigma radio+FIR AGN-dominated fraction for Delhaize/SED3FIT changes from 20% toward the Yun-based 58%, or if any starburst fraction shifts by more than about 10 percentage points, the central claim of a 'crucial' calibration dependence is not supported by the current error treatment.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Abstract; Section 4.4.5, Table 4) is that AGN-dominated fractions 'crucially depend' on the radio-derived SFR calibration, with e.g. 22–58% for Yun et al. (2001) versus 11–20% for Delhaize et al. (2017) among 3-sigma radio+FIR detections. The Delhaize-SFR is obtained by placing every quasar exactly on the Delhaize et al. (2017) IRRC (Equations 6–8). However, Section 4.3 states that the Delhaize-SFR uncertainties include only radio flux-density and redshift measurement errors; the intrinsic scatter of the IRRC is not propagated. Delhaize et al. (2017) report an rms scatter in q_IR of about 0.34 dex, whereas the fit-parameter uncertainties in q_IR(z) are only 0.01–0.03 dex. The 1-sigma_1:1 thresholds used to classify sources as AGN-dominated or possible starbursts in Tables 4–5 are therefore far tighter than the physical scatter of the relation a single source would have if it were star-forming. Repeating the classification with sigma_qIR added in quadrature would likely move many sources across the threshold; the 11–20% versus 22–58% contrast, and the 63% high-redshift starburst fraction in Table 5, could shrink substantially or lose formal significance. This is load-bearing because the paper's headline is a quantitative sensitivity statement, not just a qualitative ranking of calibrations.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a spectroscopically complete sample of 104 Type-1 quasars from the COSMOS and XMM-LSS fields of MIGHTEE, with 1.3-GHz radio imaging reaching an rms of about 3 microJy per beam. It extracts radio flux densities, computes radio luminosities, and compares radio-derived SFR estimates (Yun et al. 2001; Delhaize et al. 2017 IRRC) with FIR-derived SFR estimates (Kennicutt 1998b; SED3FIT) to classify sources as AGN-dominated or possible starbursts via a radio-excess method. The central claim is that the AGN-dominated fraction depends strongly on the radio-derived SFR calibration: 22–58 per cent for the Yun relation versus 11–20 per cent for the Delhaize relation among sources detected at 3 sigma in both radio and FIR, and that the possible-starburst fraction rises with redshift.","tokens_in":53415,"tokens_out":7969,"duration_ms":86854,"significance":"If the central quantitative claim holds, the paper is an important cautionary result: the radio-excess technique yields calibration-dependent AGN-versus-star-formation fractions, not intrinsic ones, with direct implications for the interpretation of radio-quiet quasars in deep radio surveys. The paper has clear strengths: transparent measurement and error propagation, retention of negative radio pixel values in statistical analyses, use of multiple independent SFR estimators, spectroscopic completeness to K_s = 21, new SALT redshifts for four objects, and a machine-readable table of radio luminosities and SFRs. The authors also explicitly flag several limitations, including the low resolution of the FIR data and the possible overestimate of FIR-derived SFRs from AGN-heated dust. However, the headline percentages depend on external calibrations and on a classification threshold that, as currently constructed, omits the intrinsic scatter of the infrared-radio correlation; this needs to be addressed before the quantitative claim can be accepted.","major_comments":[{"comment":"The Delhaize-SFR uncertainties are stated in Section 4.3 to include only radio flux-density and redshift measurement errors, and the 1-sigma_1:1 thresholds used for classification in Figures 8–9 and Tables 4–5 are therefore based only on the reported fit-parameter uncertainties of the IRRC. Delhaize et al. (2017) report an intrinsic scatter in q_IR of about 0.34 dex, which is an order of magnitude larger than the quoted parameter uncertainties of 0.01–0.03 dex. Because each quasar is placed exactly on the relation to derive its Delhaize-SFR, a source that is genuinely star-forming but has a typical q_IR offset would be misclassified as AGN-dominated or starburst at the 1-sigma level. I request that the classification be repeated with the intrinsic scatter added in quadrature to the threshold, and that the resulting AGN-dominated and possible-starburst fractions in Table 4 and the redshift trend in Table 5 be reported. Without this, the headline contrast between the Yun et al. (2001) and Delhaize et al. (2017) fractions, which is the central quantitative statement of the paper, may be an artifact of over-tight thresholds.","section":"§4.3, Eqs. (6)–(8), Table 4"},{"comment":"The rise of the possible-starburst fraction to 63 per cent in the highest redshift bin and 67 per cent in the brightest optical-magnitude bin is presented as a main result (conclusion viii). The paper itself argues in Section 4.4.3 that AGN heating of dust on kpc scales can overestimate FIR-derived SFRs precisely in these regimes (Symeonidis 2022; Symeonidis et al. 2022), which would manufacture exactly this trend. Since the FIR luminosities have already had the AGN component subtracted during SED fitting, the effect must come from extended AGN-heated dust not captured by the templates, making it a systematic rather than random uncertainty. The manuscript should either provide a quantitative estimate of the magnitude of this effect (for example, by re-fitting with a wider AGN-heating model or by comparing to a sample without AGN) or explicitly frame the redshift trend as a possible artifact of the SFR calibration, rather than listing it as a standalone finding.","section":"§4.4.3, Table 5"}],"minor_comments":[{"comment":"The whole-sample rows of Table 4 appear corrupted in the manuscript (for example, '0.37σ 1:1 2431 14 / 104 = 2330 13%' and '0 / 104 = 534 0 %'), with overlapping numbers that make the fractions unreadable; these rows should be reformatted.","section":"Table 4"},{"comment":"The statement that the difference between the 22 per cent and 58 per cent AGN-dominated fractions is 'in large part due to the more-tightly-constrained calibration of the IR-related SFR' should be quantified by explicitly listing the adopted calibration uncertainties for the Kennicutt and SED3FIT SFRs (currently given as '+0.3/−0.5 dex' and '0.1 dex', respectively) and how they translate into the sigma_1:1 values used in the threshold.","section":"§4.3"},{"comment":"The FIR data over the two fields come from different pipelines (super-deblended for COSMOS, HELP for XMM-LSS) with very different detection limits (for example, 1.44 mJy versus 12.5 mJy at 100 micron). The paper should discuss whether the systematic difference in FIR depth could bias the combined-sample fractions in Table 4, especially for the radio-detected subsample.","section":"§2.4"},{"comment":"There are several typographical issues, including 'starbust' instead of 'starburst' in Section 4.3 and Appendix E, and a garbled sentence in the caption of Figure E1 that should be corrected.","section":"Appendix E"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid, well-written data paper with a useful cautionary message. My main technical concern is the missing intrinsic scatter in the Delhaize-SFR uncertainty budget, which directly affects the headline percentages; this is a standard, straightforward check that the authors should be asked to perform. I also think the redshift trend of starbursts is presented too strongly given the paper's own caveats. I would not reject: the central qualitative point, that classification depends on the chosen radio SFR calibration, is likely to survive, but the magnitude of the effect needs to be re-evaluated after including the IRRC scatter. No concerns about circularity; the self-citations are legitimate prior literature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a solid observational first-look, and the core claim holds qualitatively. The 104 spectroscopically confirmed type-1 quasars in MIGHTEE at rms ~3 uJy is a genuinely new sample, and the paper shows clearly that the AGN-dominated versus starburst classification depends on which radio-to-SFR calibration you adopt, Yun et al. (2001) versus Delhaize et al. (2017). That sensitivity is real and worth publishing.\n\nThe paper does several things well. Negative radio pixel values are retained in the statistics. The MIGHTEE noise and confusion estimates are transparent. The redshift compilation is careful, with four SALT follow-up spectra. The machine-readable table of radio luminosities and four SFR estimates lets anyone redo the analysis without trusting the authors' choices. That is reproducible evidence and should be credited.\n\nThe soft spot is the one the stress-test flagged. The Delhaize-SFR error bars propagate only radio flux-density and redshift errors; the intrinsic scatter of the IRRC (about 0.34 dex as reported by Delhaize et al. 2017) is not added. The 1-sigma thresholds in Figures 8 and Tables 4-5 therefore treat a star-forming quasar as if it must land within a few hundredths of a dex of the relation. That makes the 11-20% versus 22-58% contrast tighter than the physics supports. Adding the IRRC scatter in quadrature would almost certainly move sources across the boundaries, and the claimed 63% high-redshift starburst fraction could drop substantially. The paper's other acknowledged limitations (FIR beam blending, AGN-heated dust overestimating FIR-derived SFRs) point the same direction. None of this kills the qualitative point, but the specific percentages in the abstract and Section 4.4.5 are not robust to a defensible treatment of the uncertainty.\n\nThe redshift-trend discussion is honest about being tentative, but the framing occasionally outruns the statistics; that is a revision item rather than a fatal flaw.\n\nWho gets value: anyone working on radio-quiet quasars, radio-excess methods, or the FIR-radio correlation in AGN hosts. It deserves a serious referee and publication after the error-propagation fix and a re-computation of the fractions. I would send it to review, and I would cite the sample even while noting the calibration caveat.","headline":"A genuinely useful first-look MIGHTEE quasar sample; the qualitative calibration-sensitivity result survives, but the headline fractions are computed with over-tight error bars that omit the IRRC intrinsic scatter.","tokens_in":54021,"tokens_out":3011,"would_cite":true,"duration_ms":37190,"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":"For the same 104 quasars, the fraction whose radio emission is blamed on the black hole rather than star formation changes from 22-58 per cent to 11-20 per cent depending on which star-formation calibration is used.","keywords":["quasars","radio continuum","star formation","AGN","MIGHTEE","MeerKAT","infrared-radio correlation","radio-excess method"],"falsifier":"Take the 45 quasars in the sub-sample with SNR_FIR $> 3$ and 1.3-GHz flux $> 3\\sigma$ and observe them with arcsecond-resolution radio imaging (e.g., VLBI) and sub-arcsecond far-infrared imaging (e.g., ALMA). If the sources classed as 'possible starbursts' exhibit compact AGN radio cores and FIR emission concentrated on the quasar rather than extended disk emission, the paper's AGN-heated-dust explanation is supported; if they show resolved star-forming disks and no radio core, the starburst classification is real and the redshift trend is physical.","tokens_in":52893,"feed_emoji":"📡","tokens_out":10384,"duration_ms":107284,"temperature":0.7,"pith_summary":"This paper studies 104 spectroscopically confirmed, unobscured quasars in the COSMOS and XMM-LSS fields observed at 1.3 GHz by MIGHTEE, and asks how much of their radio light comes from the central black hole (AGN) versus star formation. The central claim is that the answer is not a stable property of the sources: the fraction classed as 'AGN-dominated' ranges from 22-58 per cent when the radio luminosity is converted to a star-formation rate with the redshift-independent relation of Yun et al. (2001), but only 11-20 per cent when a redshift-dependent infrared-radio correlation (Delhaize et al. 2017) is used instead. With the redshift-dependent relation, a larger share of quasars appear as 'possible starbursts', and that starburst fraction increases with redshift, reaching 63 per cent at the highest redshifts. This matters because the radio-excess method is a standard tool for separating AGN and star formation in faint radio sources, and the paper shows its output depends on which empirical calibration is chosen, prompting researchers to revisit the underlying assumptions.","feed_headline":"Same quasars, AGN share swings 11-58% with SFR recipe","feed_subtitle":"Deep MeerKAT data show the radio-excess method's answer depends on the chosen star-formation calibration.","key_machinery":"The machinery is the radio-excess method: each quasar's radio luminosity is converted into a star-formation rate (SFR) under the assumption that all the radio emission is from star formation, and this is compared with an independent SFR derived from infrared light. The comparisons use two radio-based calibrations, the redshift-independent relation of Yun et al. (2001) and the redshift-dependent infrared-radio correlation of Delhaize et al. (2017), and two infrared-based estimates, the Kennicutt (1998b) conversion and SED3FIT (Berta et al. 2013). Where the radio-derived SFR exceeds the infrared-derived one by more than $1\\sigma$, the paper labels the source 'AGN-dominated'; where it falls short by more than $1\\sigma$, the source is a 'possible starburst'. The paper shows that the choice of the radio-based calibration alone changes the AGN-dominated fraction by more than a factor of two.","core_discovery":"For a sample of 104 Type-1 quasars with deep MeerKAT radio data, the paper finds that the fraction of sources whose radio emission is dominated by the active galactic nucleus depends crucially on the star-formation-rate (SFR) estimate derived from the radio luminosity. Considering only quasars detected at $>3\\sigma$ in both radio and far-infrared, the AGN-dominated fraction is 22 per cent when the radio-derived SFR is based on the Yun et al. (2001) relation and compared with the Kennicutt (1998b) SFR, and 58 per cent when compared with the SED3FIT SFR; using the redshift-dependent infrared-radio correlation of Delhaize et al. (2017) instead lowers this fraction to 11-20 per cent. The paper also reports that the fraction of 'possible starbursts' rises with redshift, from 31-38 per cent in the lower bins to 63 per cent in the highest-redshift bin, and interprets this trend most plausibly as AGN-heated dust inflating the far-infrared-derived SFRs, so the AGN-dominated fractions should be treated as lower limits.","pith_inferences":["A natural but unstated consequence is that AGN fractions from surveys that adopt different radio-to-SFR calibrations are not directly comparable; a community-standard, redshift- and stellar-mass-dependent infrared-radio correlation would be needed for a single set of numbers.","If the AGN-heated-dust explanation is correct, the apparent rise of starbursts with redshift is mostly a calibration artifact, and the true AGN contribution to radio emission at high redshift is likely larger than the Delhaize-based fractions suggest.","The same two-calibration comparison could be applied to X-ray-selected AGN or submillimetre galaxies to see whether the calibration sensitivity is generic to the radio-excess method rather than specific to these quasars."],"forward_implications":["The radio-excess method produces calibration-dependent fractions: the same sample yields 22-58 per cent AGN-dominated with the Yun et al. (2001) radio-SFR relation versus 11-20 per cent with the Delhaize et al. (2017) redshift-dependent infrared-radio correlation.","Using the redshift-dependent correlation shifts the population toward 'possible starbursts', whose fraction climbs from roughly a third to 63 per cent in the highest-redshift bin ($z \\approx 2.2$-$3.4$).","The paper concludes that at high redshift, AGN-heated dust likely inflates far-infrared SFRs, so the quoted AGN-dominated fractions are lower limits rather than true values.","Because the results depend so strongly on calibration, future samples need larger numbers and stellar-mass information (as the paper plans) before the AGN-versus-starburst balance of the quasar population can be pinned down."],"supporting_citations":[{"why":"Supplies the redshift-independent radio-SFR relation (Eq. 4) that yields the higher AGN-dominated fractions (22-58 per cent).","marker":"Yun et al. (2001)"},{"why":"Supplies the redshift-dependent infrared-radio correlation (Eq. 6) that, when used to derive SFRs, lowers the AGN fraction to 11-20 per cent and raises the starburst fractions.","marker":"Delhaize et al. (2017)"},{"why":"Provides the FIR-luminosity-to-SFR conversion (Eq. 5) used as one of the two infrared-based SFR arms in the comparison.","marker":"Kennicutt (1998b)"},{"why":"Provides SED3FIT, the SED-fitting code whose more tightly constrained SFRs produce the highest AGN-dominated fraction (58 per cent) and the lowest starburst fractions.","marker":"Berta et al. (2013)"},{"why":"Provides the two-component FIR SED fits and SNR_FIR values for COSMOS sources, used to build the FIR-based SFR and detection criteria.","marker":"Jin et al. (2018)"},{"why":"Documents how accretion-related processes can make quasar hosts mimic normal star-forming galaxies, supporting the paper's treatment of AGN fractions as lower limits.","marker":"Wong et al. (2016)"}],"fun_headline_variants":["Quasar radio AGN share swings 11-58% with SFR recipe","MIGHTEE quasars: SFR method flips AGN fraction","Radio-loud quasar fraction depends on SFR calibration","How star-formation recipes alter quasar radio results","AGN vs starburst: SFR choice decides for quasars"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Delhaize et al. (2017) infrared-radio correlation, fitted to normal star-forming galaxies, accurately describes quasar host galaxies (including starbursts) out to $z \\approx 3.4$; if that correlation is wrong for quasars, the inferred AGN and starburst fractions—and their redshift trend—change.","fun_headline_variants_meta":{"raw":{"variants":["Quasar radio AGN share swings 11-58% with SFR recipe","MIGHTEE quasars: SFR method flips AGN fraction","Radio-loud quasar fraction depends on SFR calibration","How star-formation recipes alter quasar radio results","AGN vs starburst: SFR choice decides for quasars"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000736,"raw_usage":{"total_tokens":3372,"prompt_tokens":1111,"completion_tokens":2261,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":727,"completion_tokens_details":{"reasoning_tokens":2170}},"tokens_in":727,"tokens_out":2261,"duration_ms":17949,"temperature":1.0,"reasoning_tokens":2170,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:55:17.542932+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the 45 quasars in the sub-sample with SNR_FIR $> 3$ and 1.3-GHz flux $> 3\\sigma$ and observe them with arcsecond-resolution radio imaging (e.g., VLBI) and sub-arcsecond far-infrared imaging (e.g., ALMA). If the sources classed as 'possible starbursts' exhibit compact AGN radio cores and FIR emission concentrated on the quasar rather than extended disk emission, the paper's AGN-heated-dust explanation is supported; if they show resolved star-forming disks and no radio core, the starburst classification is real and the redshift trend is physical.","supporting_citations":[{"cited_title":"(2017) (Equations 6–8), respectively","cited_arxiv_id":null,"evidence_quote":"Supplies the redshift-dependent infrared-radio correlation (Eq. 6) that, when used to derive SFRs, lowers the AGN fraction to 11-20 per cent and raises the starburst fractions."}],"review_version":1}