{"id":"d1c6610d-8d1e-4719-9919-5738fe94fdfc","arxiv_id":"2607.03543","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":7,"one_line_summary":"The SFR–1.4 GHz radio correlation is log10(SFR) = 0.790(L′)+1.244(1+z)^0.122−0.033M′ with 0.178 dex scatter, showing significant redshift but weak mass dependence when AGN are treated probabilistically.","lead":"A new Bayesian mixture model applied to deep MeerKAT radio data yields a star-formation-rate to radio-luminosity relation with clear redshift evolution but only weak stellar-mass dependence. The method avoids the usual binary AGN cuts that can bias the calibration used for dust-free cosmic star-formation histories.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the reader's already-flagged spectral-index assumption.","rationale":"The central numerical result (Eq. 14 / abstract) is supported by exhaustive model comparison (WAIC/LOO), an evolving mixture fraction that absorbs Malmquist and mass-selection effects, and explicit volume-limited bootstrap KS tests that recover the same relation. The only modelling choice that can still move the reported redshift coefficient is the spectral-index family used in the k-correction; the paper already quantifies that family and shows the claim survives. Residual limitations (unpublished GRAHSP posteriors, unsampled photo-z) affect reproducibility more than correctness of the published fit. Therefore the reader's CONDITIONAL / HIGH-confidence verdict needs no adjustment.","tokens_in":24271,"tokens_out":427,"duration_ms":4435,"concrete_test":"Once MIGHTEE S-band imaging is public, recompute L_1.4 for the same COSMOS DR1 hosts with measured two-point spectral indices (or a curved-spectrum fit) and re-run the hierarchical mixture model of §3; if the recovered γ_z remains consistent with 0.122 ± 0.019 (or is fully absorbed into the new α distribution) the redshift term is robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest_assumption correctly isolates the most load-bearing modelling choice: conversion of observed flux to rest-frame L_1.4 via a fixed or simply linear-evolving spectral index (Eq. 1 and §4.4). The paper already tests the linear-evolution alternative (η_α = −0.068), recovers a consistent SFR–radio slope, and shows that the redshift term can be re-absorbed into α(z). Volume-limited KS tests (§4.5), model-selection rankings (Table 2), and the weak mass term all remain stable under that alternative. No additional internal inconsistency or untested selection effect rises to the same level of centrality for the strongest claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper derives a mean SFR–1.4 GHz radio luminosity relation for star-formation-dominated galaxies using a Bayesian hierarchical two-component mixture model applied to MIGHTEE COSMOS DR1 radio sources, with SFRs taken from full GRAHSP SED posteriors. Binary SFG/AGN cuts are avoided; instead an evolving logistic mixture fraction (in L1.4 and M*) jointly models the SF sequence and a radio-excess background. The preferred relation is log10(SFR) = 0.790 L' + 1.244 (1+z)^0.122 − 0.033 M' (L' = log10 L1.4 − 23, M' = log10 M* − 10) with intrinsic scatter 0.178 dex. Redshift evolution is required at high significance; stellar-mass dependence is weak. The redshift term can be re-absorbed into a mild spectral-index evolution α(z) = −0.7 − 0.068 z. Volume-limited bootstrap KS tests support that the relation is not driven by Malmquist bias.","tokens_in":24535,"tokens_out":1129,"duration_ms":8667,"significance":"If the result holds, the work supplies a practically usable, dust-agnostic SFR–radio calibration for deep radio surveys (MeerKAT, LOFAR, SKA pathfinders) that does not rely on the circular radio-excess cuts common in IRRC-based studies. The hierarchical mixture approach, full SFR-posterior marginalisation, and explicit model comparison via WAIC/PSIS-LOO-CV are methodological strengths that the community can reuse. The finding that mass dependence is much weaker than in recent IRRC work, while redshift evolution remains, is a concrete, falsifiable claim that can be tested with forthcoming multi-band MIGHTEE data. The paper is therefore of clear interest for galaxy evolution and radio continuum cosmology.","major_comments":[{"comment":"§2.1 and §3: photometric redshifts are adopted for a substantial fraction of the sample but their posteriors are not sampled (only SFR posteriors are). Because both L1.4 (Eq. 1) and the (1+z) term in μ_SF depend on z, residual photo-z scatter or catastrophic outliers can couple into the reported redshift evolution. A quantitative test—e.g. restricting to the spectroscopic/PAU subset, or Monte-Carlo sampling the photo-z PDFs for a representative subsample—should be shown so that the significance of γ_z = 0.122 ± 0.019 can be assessed under realistic redshift uncertainty.","section":null},{"comment":"§4.4 / Eq. (1): the conversion from observed flux to rest-frame L1.4 assumes either fixed α = −0.7 or a simple linear α(z) = −0.7 − 0.068 z. The paper already demonstrates that the redshift term can be re-absorbed into α(z) and that the slope remains consistent, which is good. However, any spectral curvature or population-dependent spectral shape not captured by this one-parameter form will still be absorbed into the reported (1+z) term. Given that multi-frequency MIGHTEE S-band data are imminent, the manuscript should state more explicitly that the present redshift coefficient is degenerate with spectral-index assumptions and should be re-calibrated once measured α are available.","section":null}],"minor_comments":[{"comment":"Table numbering is inconsistent: the main posterior table is labelled Table 3 in the text but appears as Table 5 in the caption; the second/third-best models are Table 4. Please renumber for consistency.","section":null},{"comment":"Fig. 2 right panel: the quantity log10(ΔSFR/SFR) is defined in the caption but the axis label is abbreviated; a short explicit definition in the figure itself would help.","section":null},{"comment":"§3.1: the cut log10(ΔSFR/SFR) > 1 removes only 1.8 % of sources; a one-sentence statement that results are insensitive to the precise threshold would strengthen the claim of robustness.","section":null},{"comment":"Eq. (4) and surrounding text: the decision to drop the constant intercept ζ when both mass and redshift terms are present is sensible, but a brief note that the intercept is absorbed into the normalisation of the (1+z) term would clarify the parameterisation for readers who wish to implement the relation.","section":null},{"comment":"References: a few in-prep works (Hale et al. submitted; Stylianou et al.; Jackson et al.) are cited for catalogue construction; ensure that the public data products needed to reproduce the sample selection are clearly identified or will be released with the paper.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The methodological advance is real and the central claim is defensible; the two major points are load-bearing but fixable with additional tests already within the authors’ reach. I would not require a full re-analysis with free α0 or multi-frequency data before acceptance. Fit for MNRAS is good."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The real advance here is methodological: they stop doing binary SFG/AGN cuts and instead fit a hierarchical Bayesian mixture that keeps the full GRAHSP SFR posteriors and lets the SF fraction evolve with L and M*. That produces a concrete relation with slope 0.79, intrinsic scatter 0.18 dex, a statistically required mild redshift term, and only a weak mass term (~0.13 dex across four decades in M*). That combination is new relative to Delhaize, Delvecchio, Smith, Gürkan et al., and it is the part worth citing.\n\nThey do the work carefully. Model selection uses WAIC and PSIS-LOO-CV; the top models agree on redshift evolution plus L+M* dependence in the mixture fraction. Volume-limited bootstrap KS tests show the relation is not just Malmquist bias. They explicitly re-run with α(z) = −0.7 − 0.068z and recover a consistent slope, so the redshift term can be re-absorbed into spectral-index evolution if that is the physical story. The circularity burden of earlier IRRC-style cuts is genuinely lower.\n\nSoft spots are real but secondary. Photometric-redshift posteriors are not sampled, the SED products are still in prep, and the spectral-index family remains simple (fixed or linear in z). Curvature or population-dependent spectra would still be folded into the redshift coefficient. None of that overturns the main claim or the ranking of models. The weak mass dependence is the result that will get the most push-back; their argument that previous strong mass trends were partly selection artefacts is plausible given the method, but it will need independent checks on other fields.\n\nThis is for people who actually use radio continuum as an SFR tracer or who care about AGN contamination in deep surveys. It deserves a serious referee. I would cite the numerical relation and the mixture approach, and I would bring it to reading group.","headline":"Cleaner, less-circular SFR–radio calibration with a hierarchical mixture; redshift term is real, mass term is weak, and the spectral-index assumption is already stress-tested.","tokens_in":25159,"tokens_out":502,"would_cite":true,"duration_ms":5099,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A Bayesian mixture model measures the SFR–radio relation without cutting out AGN, finding clear redshift evolution but only weak stellar-mass dependence.","keywords":["star formation rate","radio continuum","SFR–radio correlation","active galactic nuclei","Bayesian mixture model","spectral index evolution","galaxy evolution","MIGHTEE"],"falsifier":"Matched-resolution multi-frequency radio imaging of the same MIGHTEE fields that measures the true spectral-index distribution of SF-dominated galaxies as a function of redshift; if the measured indices show no systematic steepening of order Δα ≈ −0.07 per unit redshift, the physical explanation offered for the redshift term fails.","tokens_in":25193,"feed_emoji":"📡","tokens_out":992,"duration_ms":7671,"temperature":0.7,"pith_summary":"Radio luminosity is a dust-free tracer of star formation, but earlier calibrations of the SFR–radio correlation had to throw away sources that looked like AGN using hard cuts. Those cuts can bias the remaining sample. This paper replaces the cuts with a hierarchical Bayesian mixture model that treats every radio-detected galaxy as a probabilistic mixture of a star-formation-dominated sequence and a non-SF background. The model is fit to the deep MIGHTEE COSMOS catalogue while folding in the full posterior SFR distributions from multi-wavelength SED fitting. The result is a mean relation whose slope is sub-linear, whose intercept rises gently with redshift, and whose stellar-mass term is only ~0.13 dex across four decades in mass—far weaker than previous claims. The authors show that a mild steepening of the radio spectral index with redshift can fully account for the observed redshift term, and that the same relation holds inside volume-limited subsets free of Malmquist bias.","feed_headline":"SFR–radio link needs redshift, not mass cuts","feed_subtitle":"Mixture model on MIGHTEE data finds weak mass term and mild redshift evolution that spectral-index change can explain","key_machinery":"A two-component hierarchical Bayesian mixture model whose SF mean is linear in log radio luminosity, additive in a power-law redshift term and a linear mass term, and whose SF fraction is a logistic function of luminosity and stellar mass; the model is sampled with the full per-source SFR posterior.","core_discovery":"The mean star-formation-dominated SFR–radio correlation is log10(SFR/M⊙ yr−1) = 0.790 \times (log10(L1.4/W Hz−1) − 23) + 1.244 \times (1 + z)^0.122 − 0.033 \times (log10(M*/M⊙) − 10), with an intrinsic scatter of 0.178 dex. Redshift evolution is required at high significance; stellar-mass dependence is statistically present but practically negligible.","pith_inferences":["If the same mixture approach is applied to other SFR tracers (Hα, UV, IR), residual mass or redshift trends currently attributed to dust or IMF variations may likewise shrink once AGN contamination is treated probabilistically.","The logistic SF-fraction surface recovered here is itself a selection-function map; it can be used as a prior when stacking or luminosity-function modelling of the same field.","Once S-band data arrive, the spectral-index explanation can be turned into a joint fit of α(z) and the SFR–radio plane rather than a post-hoc test."],"forward_implications":["Future radio surveys can convert luminosity to SFR with a relation whose mass term can be dropped at little cost and whose redshift term is now quantified.","Binary AGN/SFG cuts are no longer required for calibrating the relation; probabilistic membership is sufficient and less biased.","Claims of strong stellar-mass dependence in the IRRC or SFR–radio plane should be re-examined under mixture modelling that keeps AGN in the sample.","A mild spectral-index evolution of order α(z) = −0.7 − 0.068z is already enough to erase the need for an explicit redshift term in the SFR–radio plane."],"fun_headline_variants":["MIGHTEE mix model: SFR-radio needs mild redshift, weak mass","Bayesian fit finds z-evolution, negligible mass in SFR-L1.4","No AGN cuts: mixture model pins redshift-driven SFR-radio link","SFR-radio correlation evolves with z, not stellar mass cuts","Statistically driven SFR-L1.4 relation shows mild redshift rise"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The conversion from observed radio flux to rest-frame 1.4 GHz luminosity assumes a spectral index that is either fixed at −0.7 or allowed only a simple linear change with redshift; any real curvature or population dependence not captured by that form is absorbed into the reported redshift evolution.","fun_headline_variants_meta":{"raw":{"variants":["MIGHTEE mix model: SFR-radio needs mild redshift, weak mass","Bayesian fit finds z-evolution, negligible mass in SFR-L1.4","No AGN cuts: mixture model pins redshift-driven SFR-radio link","SFR-radio correlation evolves with z, not stellar mass cuts","Statistically driven SFR-L1.4 relation shows mild redshift rise"]},"model":"grok-4.5","effort":"low","cost_usd":0.006216,"raw_usage":{"total_tokens":1710,"prompt_tokens":963,"num_sources_used":0,"completion_tokens":86,"cost_in_usd_ticks":62160000,"prompt_tokens_details":{"text_tokens":963,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":661,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":963,"tokens_out":86,"duration_ms":6048,"temperature":1.0,"reasoning_tokens":661,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T01:43:38.442598+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Matched-resolution multi-frequency radio imaging of the same MIGHTEE fields that measures the true spectral-index distribution of SF-dominated galaxies as a function of redshift; if the measured indices show no systematic steepening of order Δα ≈ −0.07 per unit redshift, the physical explanation offered for the redshift term fails.","supporting_citations":[],"review_version":1}