{"id":"ed14cda8-4843-4b89-af15-ee12034b9c17","arxiv_id":"2608.03576","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"A 130-galaxy HI-selected sample yields a tight radial acceleration relation with an acceleration scale near 1.5e-10 m/s^2 and a MOND shape parameter around 4, and shows that the bTFR redshift-evolution signal is largely a selection effect.","lead":"Using 130 galaxies detected in neutral hydrogen, this paper finds that the relation between baryonic mass and galaxy dynamics stays tight out to redshift 0.09. It also shows that an apparent evolution in one such relation is largely a selection effect, sharpening tests of dark matter and modified gravity.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Quoted a0 is not robust to the paper's own baryonic-model split: Table 2 gives 1.80±0.08 vs 1.31±0.06 (~5σ) for 'well-behaved' vs 'less well-behaved' subsamples, so the headline 1.50±0.05 needs a systematic term.","rationale":"I read the paper as primarily establishing the RAR and bTFR in a homogeneous HI-selected sample beyond the local Universe. The RAR fitting is well executed: Roxy's MNR method is documented to be unbiased, two independent photometry pipelines are compared, and the intrinsic scatter is consistently non-zero. However, Table 2 exposes a model-level systematic in the headline a0: splitting by photometric consistency moves a0 by ~5σ while leaving the scatter nearly unchanged. This is more load-bearing than the selection-function issue identified by the reader, because it concerns the quoted central value itself rather than the interpretation of a residual trend. The paper's own text acknowledges that a0 is not robust to the baryonic-modelling details, which is an explicit limitation of the headline number. The concern does not invalidate the paper; it means the abstract's a0=(1.50±0.05) should be qualified with a systematic term of order the Table 2 spread, and conclusions that depend on the exact value—especially the delta comparison with SPARC and the a0-anchored evolution fits—should be tested for sensitivity. This reinforces the reader's CONDITIONAL verdict rather than changing it, so the verdict remains UNCHANGED.","tokens_in":35240,"tokens_out":7219,"duration_ms":70152,"concrete_test":"Take the Table 2 subsamples and compute the median redshift and median number of resolved rings for each; then refit the RAR z-evolution model (Eq. 14) separately for the two subsamples with identical priors. If the best-fit a1 differs between subsamples by more than the combined 1σ uncertainty, or if the subsample a0 values correlate with redshift, the a0 offset is a redshift-dependent systematic and the headline a0 and no-evolution RAR claim need a systematic error term. If the two a1 fits agree and the subsample redshifts overlap, the offset is a benign structural effect and the quoted RAR parameters stand.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 4.2 and Table 2, the sample is split at chi2_red≈3 into 61 'well-behaved' galaxies, where Sersic and direct-aperture stellar-mass pipelines agree, and 69 'less well-behaved' galaxies. Fitting Eq. 10 separately gives a0=(1.80±0.08)×10^-10 m/s^2 versus (1.31±0.06)×10^-10 m/s^2, a ~5σ offset, with comparable intrinsic scatters. The paper nevertheless quotes a0=(1.50±0.05)×10^-10 m/s^2 as the fiducial full-sample result and presents it in the abstract without this systematic. The limitation is acknowledged in the text: the authors state that the tightness of the RAR is robust to baryonic modelling 'even when the a0 is not' (Section 4.2). This is load-bearing because the RAR normalization is a headline result, because the delta-family inference in Section 4.3 and the a0-anchored bTFR evolution fits in Section 4.5 use this scale, and because if photometric consistency correlates with redshift or with the number of resolved elements (plausible, given the outer-annulus divergence in Fig. 8), the 5σ a0 offset could masquerade as or mask genuine redshift evolution, affecting the claimed no-evolution RAR and the 3.4σ bTFR comparison. The quoted statistical errors therefore understate the model uncertainty in the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents measurements of the baryonic Tully-Fisher relation (bTFR) and the radial acceleration relation (RAR) for 130 purely HI-selected galaxies from MIGHTEE and LADUMA out to z≈0.09. The authors derive resolved stellar mass profiles via SED fitting using two independent photometry pipelines (direct aperture photometry and Sérsic fits), combine these with 3D Barolo HI rotation curves, and fit the RAR and bTFR with the Roxy/MNR framework. The main results are a RAR acceleration scale a0=(1.50±0.05)×10^-10 m/s^2 with intrinsic scatter 0.096±0.006 dex; a delta-family MOND interpolating-function shape δ=4.10(+1.4,-0.68) reported as consistent with Solar System and wide-binary constraints; a bTFR inverse slope s'=0.27±0.01 with orthogonal scatter ≈0.05 dex; and a redshift-evolution analysis in which the bTFR zero-point evolution drops from an 8.7σ preference in the forward fit to 3.4σ in the inverse fit, interpreted as largely arising from HI flux selection.","tokens_in":35578,"tokens_out":7723,"duration_ms":73434,"significance":"If the results hold, this is the first RAR and bTFR measurement from a purely HI-selected sample beyond the local Universe with fully resolved baryonic mass profiles, and the inferred δ value would ameliorate the known tension between SPARC-based RAR fits and Solar System/wide-binary MOND constraints. The paper has notable strengths: it uses two independent photometric pipelines, explicitly examines systematic effects through a well-behaved/less well-behaved sample split, applies an inverse bTFR fit to mitigate selection, and makes use of publicly available fitting and analysis packages. The central demonstration that selection effects and baryonic modeling choices, rather than statistics alone, control apparent redshift evolution of dynamical scaling relations is an important and timely contribution.","major_comments":[{"comment":"The headline value a0=(1.50±0.05)×10^-10 m/s^2 is quoted in the abstract and Section 5(i) without an accompanying systematic term. The authors' own Table 2 shows that the well-behaved and less well-behaved subsamples yield a0=(1.80±0.08) and (1.31±0.06)×10^-10 m/s^2, a ~5σ offset with comparable intrinsic scatters. Because a0 enters the δ-family inference (Section 4.3) and the a0-anchored bTFR evolution fits (Eq. 17), the quoted statistical error understates the model uncertainty in the central claim. The paper should report a0 with a systematic term derived from the pipeline split (for example, half the difference of the subsample values added in quadrature) and should state whether the no-evolution RAR result is robust when the evolution fit is repeated separately for the two subsamples, since photometric consistency could correlate with redshift or with the number of resolved rings.","section":"§4.2, Table 2"},{"comment":"The interpretation that the bTFR evolution is largely selection bias rests on the inverse fit reducing the evolution preference from 8.7σ to 3.4σ. The text correctly says that conditioning on Mbar mitigates but does not fully remove the bias, yet the abstract's 'within ≈2σ of the RAR evolution constraint' and conclusion (vi) present the residual 3.4σ signal as a selection artifact. Because the HI selection function is not modeled, and because redshift correlates with M_HI, M*, and the number of resolved rings (Figs. 14–15), the residual could be genuine evolution. I recommend adding a synthetic-selection or forward-model test that quantifies how much of the measured a1 in the inverse fit is removed under a realistic selection function, or, failing that, explicitly stating in the abstract and conclusions that the data cannot distinguish residual selection from moderate evolution.","section":"§4.6.2, Eq. (17)"},{"comment":"The RAR fit treats all 476 radial bins as independent data points. Rotation-curve rings are spaced at half the beam and the baryonic profiles are smoothed, so adjacent points within a galaxy are likely correlated. This can bias the quoted intrinsic scatter sigma_int=0.096±0.006 dex and the formal uncertainties on a0. The paper should either model within-galaxy covariance (for example through resampling or a per-galaxy random effect) or demonstrate, such as by fitting a single point per galaxy, that the conclusions are unchanged. This is particularly relevant to the claim that the intrinsic scatter is clearly non-zero statistically.","section":"§4.1, Eq. (10)"}],"minor_comments":[{"comment":"Section 4.1 states the two photometric pipelines agree on a0 at the 0.1σ level, while Section 5(ii) says 0.3σ; the numbers should be made consistent.","section":"§4.1 vs §5(ii)"},{"comment":"The COSMOS reference intrinsic scatter is listed as 0.045±0.002 dex in Table 2 but quoted as 0.045±0.022 dex in Section 4.1; the table entry appears to be a typo.","section":"Table 2"},{"comment":"The statement that excluding galaxies with foreground stellar contaminants 'does not bias our Hi-selected sample in any way' is too strong; masking decisions can in principle correlate with galaxy properties. Please soften the claim or provide justification.","section":"§2.4"},{"comment":"The text reports the vertical intrinsic scatter as 0.054±0.011 dex, while Table 3 gives 0.054±0.006 dex; also, the 'direct-equivalent zero-point I=2.39±0.33' is actually -I/s' in Eq. (13), so the notation should be clarified to avoid confusion with the fitted intercept.","section":"§4.4, Table 3"},{"comment":"The caption says a0=1.50×10^-10 for both panels, but the Sérsic-based fit gives a0=(1.48±0.04)×10^-10; the caption should indicate that the displayed curve is the fiducial fit rather than the Sérsic fit value.","section":"Fig. 7 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is appropriate for MNRAS and the central statistical analysis is generally sound. The main issue is that the headline a0 is presented without the systematic uncertainty that the authors themselves quantify in Table 2, and this systematic propagates into the delta and evolution claims. The bTFR selection discussion is already cautious, but a synthetic-selection test would materially strengthen the central interpretation. I do not see grounds for rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should look at this one. It is the first RAR/bTFR measurement on a large, homogeneous HI-selected sample beyond z≈0: 130 galaxies from MIGHTEE and LADUMA, with resolved rotation curves and resolved baryonic mass profiles from SED fitting. That alone is a real step forward. The paper does several things well. Two independent photometry pipelines give a0 consistent at the 0.1-0.3 sigma level, which is genuinely reassuring. The intrinsic scatter is statically nonzero and in line with local values. The delta-family fit yielding delta~4.1, consistent with Solar System and wide-binary constraints, is an interesting and honest result, though the posterior is broad. And the forward-vs-inverse bTFR comparison is a clean demonstration of selection bias, with the forward 8.7-sigma evolution dropping to 3.4-sigma in the inverse fit.\n\nThe soft spots are real but mostly acknowledged in the text. The big one is Table 2: splitting by photometric consistency gives a0 = 1.80±0.08 for the 61 'well-behaved' galaxies and 1.31±0.06 for the 69 'less well-behaved' ones, a 5-sigma offset. The paper says the RAR's tightness is robust even when a0 is not, which is fair, but the abstract quotes a0 = 1.50±0.05 without a systematic term. That is misleading. The reader's stress-test is right: this systematic should be in the headline or the claim softened. It matters because the delta-family inference and the a0-anchored bTFR evolution fits inherit this scale.\n\nThe second soft spot is the selection function. The paper conditions on Mbar to mitigate HI flux selection and is careful to say this 'mitigates but does not fully remove' the bias. But the residual 3.4-sigma inverse-fit evolution could still be genuine, or could be a residual of unmodeled selection on Vout, rotation-curve extent, or data quality. I don't think this kills the paper—the interpretation as selection is plausible and the evidence for it is decent—but it is not a closed case.\n\nMinor: data availability says 'available on request' and no code is released. That is weak for a paper this method-heavy.\n\nVerdict: the central RAR measurement is likely sound; the evolution conclusions are plausible but not definitive. I would send this to a serious referee, with a request that the authors add a systematic term to a0, model the selection function more explicitly, and release the data products. It deserves to be in the literature after revision.","headline":"The largest homogeneous HI-selected RAR/bTFR sample beyond z=0, with a clean central measurement but a known 5-sigma a0 systematic that the abstract understates; worth serious review.","tokens_in":36341,"tokens_out":2042,"would_cite":true,"duration_ms":20626,"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":"With 130 HI-selected galaxies out to z≈0.09, the radial acceleration relation remains tight, and the apparent baryonic Tully-Fisher zero-point evolution is mostly selection bias.","keywords":["radial acceleration relation","baryonic Tully-Fisher relation","MOND interpolating function","HI-selected galaxies","MIGHTEE-HI","LADUMA","redshift evolution","selection bias"],"falsifier":"A definitive check is to repeat the bTFR analysis on a sample selected by stellar mass or with a fully characterised HI selection function, or to split the present sample into redshift bins with matched HI-mass distributions and test whether the inverse-fit zero-point shift persists; a surviving 3.4σ signal under complete selection modelling would indicate genuine evolution, while a vanishing signal would confirm that the forward 8.7σ is selection bias.","tokens_in":34985,"feed_emoji":"🌀","tokens_out":8319,"duration_ms":69649,"temperature":0.7,"pith_summary":"The paper sets out to measure how closely galaxy rotation tracks the mass of visible baryons using 130 galaxies selected purely by their neutral-hydrogen emission, observed with MeerKAT in the MIGHTEE and LADUMA surveys out to z≈0.09. It claims the radial acceleration relation (RAR) holds out to this redshift with an acceleration scale $a_0=(1.50\\pm0.05)\\times10^{-10}\\,\\mathrm{m\\,s^{-2}}$ and an intrinsic scatter of $0.096\\pm0.006$ dex, matching local results. It further claims the MOND interpolating-function shape parameter is $\\delta=4.10^{+1.4}_{-0.68}$, consistent with Solar System gravitational constraints and a null Wide Binary Test, and that the apparent redshift evolution of the baryonic Tully-Fisher (bTFR) zero-point is mostly a selection artefact: the forward fit shows an $8.7\\sigma$ trend, which drops to $3.4\\sigma$ when the fit conditions on baryonic mass. If true, these results show that the tight baryon-dynamics coupling extends beyond the local Universe and that reported evolution of scaling relations needs re-examination for selection bias.","feed_headline":"RAR scatter stays near 0.1 dex out to z≈0.09","feed_subtitle":"130 HI-selected galaxies confirm the baryon-dynamics link; apparent Tully-Fisher evolution fades in inverse fits.","key_machinery":"The central object is the radial acceleration relation $g_{\\rm obs}=\\mathcal{F}(g_{\\rm bar})$, linking the observed centripetal acceleration to the acceleration predicted from the baryonic mass distribution alone, together with its deep-MOND limit $V_f^4\\propto G M_{\\rm bar} a_0$, which yields the baryonic Tully-Fisher relation. The analysis is carried by three devices: resolved SED fitting that lets the stellar mass-to-light ratio vary with radius, 3D Barolo tilted-ring fits to the HI data cubes that give rotation curves and gas surface densities, and the Roxy marginalised normal regression that handles uncertainties in both coordinates and intrinsic scatter. For the redshift-evolution claim, the decisive step is fitting the bTFR in the inverse direction, conditioning on $M_{\\rm bar}$ rather than $V_{\\rm out}$, because the HI flux selection acts on $M_{\\rm HI}$ and hence on $M_{\\rm bar}$.","core_discovery":"Using 130 HI-selected galaxies with resolved HI rotation curves and resolved baryonic mass profiles, the paper measures the radial acceleration relation and the baryonic Tully-Fisher relation out to z≈0.09. The RAR is tight: $a_0=(1.50\\pm0.05)\\times10^{-10}\\,\\mathrm{m\\,s^{-2}}$ and intrinsic scatter $0.096\\pm0.006$ dex with direct-aperture photometry, and the acceleration scale is robust to the photometric pipeline. The fiducial inverse bTFR fit gives a slope of $0.27\\pm0.01$ (forward-equivalent $3.72\\pm0.16$) with an orthogonal scatter of about $0.05$ dex. Fitting the $\\delta$-family of MOND interpolating functions returns $\\delta=4.10^{+1.4}_{-0.68}$, a steep transition that is consistent with Cassini quadrupole and wide-binary null results, unlike the SPARC-preferred $\\delta\\approx1$. The paper finds no significant redshift evolution of the RAR acceleration scale ($a_1=-1.6\\pm2.3\\times10^{-10}\\,\\mathrm{m\\,s^{-2}}$), while the bTFR zero-point shows an apparent evolution that depends strongly on fit direction: $8.7\\sigma$ in the forward fit, reduced to $3.4\\sigma$ in the inverse fit conditioning on $M_{\\rm bar}$, within about $2\\sigma$ of the RAR constraint. The paper interprets this as the signature of HI flux selection biasing the forward fit.","pith_inferences":["Not claimed in the paper, but following from its results: if $\\delta\\approx4.1$ is the true MOND transition shape, the Solar System quadrupole should sit near its current upper bound and wide-binary MOND signals should remain undetected, sharpening the MOND programme's testable predictions.","The paper's selection-bias argument generalises to other flux-limited HI samples: scaling relations that place a selection-correlated mass on the ordinate should show spurious redshift evolution, so inverse fits or relations with that quantity on the abscissa are safer diagnostics. Re-running the analysis on optical emission-line samples with an explicit HI-equivalent selection model would test th","If a future sample with a fully characterised selection function still recovers the $3.4\\sigma$ inverse-fit signal, that would point to a mild genuine evolution of the baryon-dynamics coupling, possibly tied to gas-fraction evolution rather than to MOND's acceleration scale.","The paper's comparison with SPARC anchoring suggests sample-specific systematics dominate once a z≈0 anchor is added; extending the homogeneous HI-selected analysis to a longer redshift baseline with one consistent baryonic model is the direct way to separate those systematics from cosmology."],"forward_implications":["The RAR's intrinsic scatter near 0.1 dex and acceleration scale matching local values imply the baryon-dynamics coupling is already in place by z≈0.09, not just at z=0.","A MOND interpolating-function shape of approximately 4.1 would reconcile galaxy-scale RAR data with Solar System and wide-binary constraints under the MOND interpretation, easing the reported tension with SPARC.","Apparent bTFR zero-point evolution measured with forward fits in HI flux-limited samples should be treated with caution; inverse fits conditioning on $M_{\\rm bar}$ are a necessary safeguard.","Future claims of bTFR redshift evolution from HI-selected samples must model the selection function, since conditioning on $M_{\\rm bar}$ mitigates but does not fully remove the bias.","Under the fiducial inverse fit, the RAR and bTFR remain mutually consistent within about $2\\sigma$ over the redshift range probed, so genuine evolution is neither required nor excluded by the data."],"supporting_citations":[{"why":"Defines the radial acceleration relation and provides the local SPARC baseline against which the new measurement is compared.","marker":"McGaugh et al. (2016)"},{"why":"Supplies the local SPARC baryonic Tully-Fisher relation using Vflat that anchors the z≈0 end and the comparison zero-point.","marker":"Lelli et al. (2016b)"},{"why":"Provides the earlier MIGHTEE bTFR measurement at similar redshifts using the same outer rotation velocity Vout.","marker":"Ponomareva et al. (2021)"},{"why":"Establishes the resolved SED mass-to-light-ratio methodology and the previous 19-galaxy COSMOS RAR result this work extends.","marker":"Vărăşteanu et al. (2025)"},{"why":"Supplies the Roxy marginalised normal regression fitting method used for all RAR and bTFR fits.","marker":"Bartlett & Desmond (2023)"},{"why":"Defines the δ-family of MOND interpolating functions used to infer the transition shape.","marker":"Famaey & McGaugh (2012)"},{"why":"Updates the Solar System quadrupole constraint that requires a steep MOND transition, i.e. high δ.","marker":"Park et al. (2026)"},{"why":"The null Wide Binary Test that, together with the quadrupole, requires δ≳2–3.","marker":"Banik et al. (2024)"}],"fun_headline_variants":["130 HI galaxies test baryon-dynamics link to z≈0.09","RAR stays tight to z≈0.09 with scatter near 0.1 dex","bTFR evolution fades in inverse fits of 130 HI-selected galaxies","MOND steep δ≈4.1 favored by 130 HI galaxy dynamics","Selection effects explain apparent bTFR evolution in HI galaxies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that conditioning the bTFR fit on baryonic mass leaves only a small residual selection bias, so the surviving 3.4σ inverse-fit signal is not genuine evolution; if the HI flux selection also correlates with rotation velocity, rotation-curve extent, or data quality that tracks redshift, the residual signal could be real.","fun_headline_variants_meta":{"raw":{"variants":["130 HI galaxies test baryon-dynamics link to z≈0.09","RAR stays tight to z≈0.09 with scatter near 0.1 dex","bTFR evolution fades in inverse fits of 130 HI-selected galaxies","MOND steep δ≈4.1 favored by 130 HI galaxy dynamics","Selection effects explain apparent bTFR evolution in HI galaxies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000724,"raw_usage":{"total_tokens":3418,"prompt_tokens":1289,"completion_tokens":2129,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":905,"completion_tokens_details":{"reasoning_tokens":2028}},"tokens_in":905,"tokens_out":2129,"duration_ms":13414,"temperature":1.0,"reasoning_tokens":2028,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:48:57.050418+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A definitive check is to repeat the bTFR analysis on a sample selected by stellar mass or with a fully characterised HI selection function, or to split the present sample into redshift bins with matched HI-mass distributions and test whether the inverse-fit zero-point shift persists; a surviving 3.4σ signal under complete selection modelling would indicate genuine evolution, while a vanishing signal would confirm that the forward 8.7σ is selection bias.","supporting_citations":[],"review_version":2}