{"id":"3e68a0dd-e4bd-457b-8ef2-1a2d451afe04","arxiv_id":"2412.14970","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Galaxy half-light radii continue to shrink with increasing redshift out to z~12.5, and small high-redshift galaxies are preferentially round, with a merger fraction of about 12%.","lead":"This paper measures the shapes, sizes, and merger activity of 520 galaxies seen when the universe was less than a billion years old, using JWST images. It finds that galaxies were smaller and rounder at earlier times, and that galaxy size kept growing with cosmic time up to redshift 12.5.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Redshift-simulation test misses morphological K-correction: F444W samples rest-frame UV at z>7, so part of the size decline could be a wavelength effect.","rationale":"The reader's weakest assumption focused on photometric redshift and stellar mass reliability, which are valid concerns but are not the most load-bearing for the specific claim that the size evolution is a real evolutionary effect rather than a redshift-observation artifact. The paper's own strongest evidence against a redshift artifact is the AREIA simulation, and that simulation has a critical omission: it does not account for the fact that F444W probes different rest-frame wavelengths at different redshifts. The simulation takes the F444W image of a z~7 galaxy (rest-frame optical) and places it at z~12, where real observations would see rest-frame UV. Because galaxy sizes are known to depend on rest-frame wavelength, the comparison is not apples-to-apples. This gap directly affects the central conclusion. If the K-correction is substantial, the observed size decline could flatten or even vanish after correction, making the claim of continued size evolution to z~12.5 insecure. The lack of spectroscopic redshifts adds uncertainty, but even with perfect redshifts the wavelength issue would remain. The abstract/body discrepancy in the power-law normalization and galaxy count is a separate quality concern but does not change the scientific substance. Therefore, the load-bearing concern is the missing morphological K-correction in the redshift simulation test, and the concrete test described above would settle whether it actually biases the inferred evolution.","tokens_in":32381,"tokens_out":8368,"duration_ms":73206,"concrete_test":"Repeat the AREIA analysis using the F277W or F356W images of the same 285 galaxies as input; at z~7 these bands probe rest-frame ~0.35-0.45 micron, approximating the rest-frame wavelength that F444W samples at z~10-12. Simulate these galaxies to z=12 and remeasure sizes in F444W. If the fractional change in Re acquires a significantly negative gradient (e.g., < -0.02 per unit z) compared to the current 0.008±0.002, the observed size decline is partly a morphological K-correction and the evolutionary interpretation is weakened. If the gradient remains flat, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central evidence that the observed size decline is real rather than a redshift artifact is the AREIA simulation (Section 6.5). Galaxies from 6.5<z<7.5 are re-binned and flux-scaled to z~12, and the measured half-light radius gradient is 0.008±0.002 per unit z, interpreted as flat. However, the simulation preserves the original F444W image, which at z~7 corresponds to rest-frame ~0.55 micron (optical), whereas at z~12 the same band samples rest-frame ~0.34 micron (UV). The simulated galaxy therefore retains an optical morphology at all redshifts, so the test never includes the morphological K-correction that would affect a real z~12 observation. Galaxy sizes are known to decrease at bluer rest-frame wavelengths, so the observed trend from z~7 to z~12 could be partly driven by the shift of F444W from optical to UV, not solely by physical size evolution. The paper claims the simulation includes 'all known cosmological effects' but does not model this wavelength shift. This is a load-bearing gap: without a wavelength-aware simulation, the conclusion that the trend is a genuine evolutionary phase is not fully established.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a structural analysis of 520 (or 521, depending on the manuscript version; see Major Comment 1) galaxy candidates at 6.5 < z < 12.5 from the EPOCHS v1 survey, using GALFIT single-Sersic fits in F444W and Morfometryka non-parametric measurements. The authors derive a size evolution Re = (2.12 ± 0.28)(1+z)^(-0.67 ± 0.06) kpc (full-text version), report a size-mass relation whose normalization evolves such that galaxies of fixed mass are smaller at higher redshift, measure a merger fraction fm = 0.12 ± 0.07, and find an excess of round, compact objects. They use AREIA artificial redshift simulations to argue that the observed size, axis-ratio, and asymmetry trends are real evolutionary effects rather than redshift-dependent observational artifacts.","tokens_in":32658,"tokens_out":6374,"duration_ms":50412,"significance":"If the results hold, they extend the known size evolution of galaxies into the epoch of reionization and support the interpretation that the compactness of z > 6.5 galaxies is a genuine evolutionary phase. The paper's strengths include the use of multiple independent size measures (Sersic, Petrosian, Kron), tests of PSF systematics (Appendix C), recovery simulations of Sersic profiles (Appendix E), and a uniform reduction across six JWST fields. The paper also makes comparisons with simulations and previous work. The main technical risk is the incompleteness of the artificial redshift simulation in reproducing rest-frame wavelength changes, and the internal inconsistency in the abstract's headline numbers.","major_comments":[{"comment":"The arXiv version of the abstract quotes 521 galaxies and Re = (2.74 ± 0.49)(1+z)^(-0.79 ± 0.08) kpc, while the full-text abstract and the conclusions in Section 8 quote 520 galaxies and Re = (2.12 ± 0.28)(1+z)^(-0.67 ± 0.06) kpc. These are the two central quantitative claims of the paper, and the discrepancy must be reconciled before publication; it also raises the question of which version corresponds to the submitted analysis.","section":"Abstract and Section 8"},{"comment":"The AREIA simulation re-bins and flux-scales the F444W image and convolves it with the F444W PSF, but it does not change the rest-frame wavelength of the light being observed. At z ~ 7 F444W samples rest-frame ~0.55 micron, while at the simulated z ~ 12 it would sample ~0.34 micron in a real observation. Since galaxy size is wavelength-dependent, the simulation cannot test whether part of the observed size decline from z ~ 7 to z ~ 12 is a morphological K-correction. The claim in Section 6.5 that the simulation shows the trends 'are due to real evolutionary effects and not redshift effects' is therefore stronger than the simulation supports. Please either include wavelength-dependent mock observations (e.g., using multi-band imaging to assign rest-frame wavelength-dependent morphologies) or explicitly quantify the expected K-correction and revise the claim accordingly.","section":"Section 6.5 and Figures 13/Table 3"},{"comment":"The photometric redshift uncertainty, including the admitted z ~ 10 scattering between F115W and F150W, is not propagated into the size-redshift power law or the size-mass relation. The size-mass relation (Table 1) assumes a fixed gradient from the 6.5 < z < 7.5 bin, so the claimed evolution is entirely in the normalization; this assumption should be checked (e.g., by allowing the slope to vary in stacked bins) and the effect of photo-z scatter on the bin medians should be quantified or at least discussed as a systematic.","section":"Section 3.2 and Section 6.1"},{"comment":"73 of the 520 galaxies (14%) have their Sersic index fixed to n = 0.05 after failing the free-fit criteria. While Appendix A shows similar sizes for n = 0.05 and n = 1, the possibility that these objects are systematically different (e.g., higher-n compact objects) could bias the median sizes and the size evolution. Please show that the main results are robust to excluding these 73 objects.","section":"Section 5.3 and Appendix A"}],"minor_comments":[{"comment":"The paper states that F444W is used to 'best probe the rest frame optical', but at z > 9 F444W corresponds to rest-frame UV (~340 nm at z = 12). The wording should be adjusted, and the caveat that the rest-frame wavelength varies across the sample should be stated wherever morphological K-correction is discussed.","section":"Section 4.1"},{"comment":"The phrase 'all known cosmological effects' in Section 6.5 is an overstatement; the simulation includes geometric rebinning, surface brightness dimming, and noise, but not wavelength-dependent morphology or color-dependent PSF variations.","section":"Section 6.5"},{"comment":"The header rows use approximate redshifts (7, 8, 9, 10.5, 12) rather than the bin definitions in the text; please make the bins consistent for reproducibility.","section":"Table 2"},{"comment":"The cutout mosaics would benefit from a scale bar in physical kpc at each redshift; currently the PSF FWHM is shown but not a kpc scale, making visual size comparisons difficult.","section":"Figure 2"},{"comment":"The NMAD values in Table 4 are useful, but the sample used (235 galaxies) is a subset; please state whether the F410M trends are verified on the same redshift bins and with the same selection.","section":"Appendix B"},{"comment":"Several entries have only arXiv IDs (e.g., Adams et al. 2024) and should be updated with accepted journal references where available.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The abstract inconsistency is likely a version-control remnant, but it must be fixed before the paper can be considered. The main technical risk is the missing morphological K-correction in the AREIA simulation; this is fixable by either adding a wavelength-aware simulation or toning down the conclusion and adding a systematic budget. The paper is otherwise well-executed and deserves another round after these points are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a useful paper and probably right on the main trend. The new content is the extension of measured rest-optical (or near-UV at the top end) sizes and size-mass relations to z~12.5 with 520 galaxies from a homogeneous reduction, plus the high-b/a excess at low masses. The earlier JWST work stops around z~8, and this sample is several times larger than previous structural studies at z>8, so the headline measurement is genuinely new.\n\nWhat is done well: careful sample selection and fitting pipeline; multiple independent size estimators (GALFIT R_e, Petrosian/Kron radii, R50) all show the same decline; F444W/F410M comparison; PSF model tests; simulated Sersic recovery tests in Appendix E; the merger criterion is from external CAS calibration. These make the main claims credible.\n\nSoft spots, in order. First, the abstract in the arXiv version says 521 galaxies and R_e = 2.74±0.49 (1+z)^-0.79±0.08 kpc, while the paper body abstract and conclusions say 520 and R_e = 2.12±0.28 (1+z)^-0.67±0.06. That is a direct internal inconsistency on two headline numbers; it has to be fixed before publication. Second, the AREIA simulation keeps the original F444W image and just re-bins and dims it, so it never changes rest-frame wavelength. At z~7 F444W is ~0.55 micron; at z~12 it is ~0.34 micron. Since galaxy sizes depend on wavelength, part of the observed decline could be morphological K-correction, and the claim that the simulation includes 'all known cosmological effects' is not justified. I would not call the central result dead—the independent size measures and the size-mass behavior at fixed mass support a real trend—but the simulation section should either be re-run wavelength-aware or explicitly recalibrated as a dimming/resolution test only. Third, photo-z and SED masses carry a real burden at z~10; the paper is honest about this, and the redshift bin merging helps, but a systematic photo-z error could tilt the high-z size values. Finally, the round-object excess at small sizes is at least partly a fitting-bias suspect; Appendix E addresses it but with simulated Sersic profiles rather than realistic clumpy morphologies, so the caveat should stay.\n\nAudience: observers and theorists working on early galaxy structure, size-mass relations, and merger fractions. It deserves a serious referee round; with the abstract fixed and the simulation claim tightened, it would be a solid contribution.","headline":"Solid extension of size evolution to z~12.5 with a larger sample, but the abstract/body numbers disagree and the redshift-simulation test misses the wavelength shift it claims to cover.","tokens_in":33324,"tokens_out":3117,"would_cite":true,"duration_ms":26120,"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":"This paper shows that the known decline of galaxy half-light radius with redshift continues to $z \\sim 12.5$, with $R_e = (2.12 \\pm 0.28)\\,(1+z)^{-0.67 \\pm 0.06}$ kpc, and that at fixed stellar mass galaxies were more compact earlier…","keywords":["galaxy structure","high-redshift galaxies","epoch of reionization","Sersic profile","half-light radius","size-mass relation","galaxy mergers","JWST NIRCam"],"falsifier":"If a substantial fraction of the $z \\sim 9$-$12$ candidates were reassigned to neighboring redshifts by JWST/NIRSpec spectroscopy, and the size-redshift or size-mass trends flattened after reassignment, the central claim would fail. A direct test would take the 520 galaxies, measure spectroscopic redshifts for as many as possible, and re-derive the $R_e(z)$ power law and the size-mass normalization with the secure redshifts only; if the smaller sizes at $z \\sim 10$-$12$ disappear, the evolutionary interpretation is wrong.","tokens_in":32225,"feed_emoji":"🔭","tokens_out":6245,"duration_ms":47221,"temperature":0.7,"pith_summary":"This paper analyzes the sizes and shapes of 520 galaxies seen when the universe was less than a billion years old, between redshifts 6.5 and 12.5. It argues that galaxies did not stop shrinking at the redshifts probed by earlier surveys: the known decline of half-light radius continues, following $R_e = (2.12 \\pm 0.28)\\,(1+z)^{-0.67 \\pm 0.06}$ kpc, and at a fixed stellar mass galaxies were noticeably more compact earlier, growing by about 53% in roughly 400 Myr. The paper also reports that the fraction of galaxies identifiable as mergers stays near $f_m \\sim 0.12$ with little redshift dependence, and that the smallest galaxies tend to be rounder. The reason these trends matter is that they bear on whether compact galaxies in the early universe are a real evolutionary phase or an artifact of observing faint, distant sources. Artificial redshift experiments are used to argue that the trends are real.","feed_headline":"JWST shows galaxies shrank continuously to z~12.5","feed_subtitle":"A 520-galaxy sample proves compactness at cosmic dawn is real evolution, not a distance illusion.","key_machinery":"The argument is carried by single-component S\\'ersic profile fits made with GALFIT in the JWST NIRCam F444W band, chosen to sample rest-frame optical light and to minimize morphological K-correction; the half-light radius $R_e$, S\\'ersic index $n$, and axis ratio $b/a$ from these fits are the parametric measurements. Non-parametric concentration and asymmetry are measured with Morfometryka, and the asymmetry criterion $A > 0.35$ from the CAS system is used to classify mergers. To rule out redshift-dependent observational biases, a subsample of 285 galaxies is artificially redshifted in steps of $\\Delta z = 1$ up to $z = 12$ using the AREIA code, which rebins, dims, adds shot noise, and re-convolves sources with the F444W PSF, and the remeasured parameters are compared as fractional changes. The flat fractional-change trends for size, axis ratio, and asymmetry constitute the key check that the observed evolution is not an artifact.","core_discovery":"The central claim is that galaxy structure evolution continues into the Epoch of Reionization: in rest-frame optical light (F444W), galaxies become progressively smaller with increasing redshift out to $z \\sim 12.5$, and the size-mass relation shifts so that galaxies of a given stellar mass are smaller at earlier times. The paper quantifies this with a single S\\'ersic half-light radius that declines as $R_e = (2.12 \\pm 0.28)\\,(1+z)^{-0.67 \\pm 0.06}$ kpc, and with a size-mass relation whose normalization drops from $\\log_{10}(R_{e0}/\\mathrm{kpc}) \\sim -0.6$ to $-0.77$ across redshift bins, corresponding to a roughly 53% size increase for a $\\log(M_*/M_\\odot) = 8$ galaxy in about 400 Myr. In addition, non-parametric asymmetry measurements give a merger fraction of $f_m = 0.12 \\pm 0.07$, consistent with the roughly constant $f_m \\sim 0.1$ seen at $z < 9$, and the sample shows an excess of round, high-axis-ratio objects among the most compact galaxies. The paper's redshift-simulation check finds flat fractional changes in size, axis ratio, and asymmetry when real galaxies are artificially moved to higher redshifts, which is taken as evidence that the observed trends reflect real evolution rather than surface-brightness or resolution effects.","pith_inferences":["If confirmed with spectroscopy, the size-mass evolution provides a tight timeline for feedback-driven size growth: dense, lower-mass galaxies must puff up by about 50% within roughly 400 Myr, which simulations will need to reproduce without overproducing massive compact remnants.","Because no mass cut was applied, the sample is dominated by low-mass galaxies; deeper, wider surveys should test whether the same power law holds for $\\log(M_*/M_\\odot) > 9$ at $z > 8$, where current numbers are small.","The high axis ratios of compact galaxies could partly reflect a bright unresolved core; stacking the compact objects to search for faint extended emission would distinguish intrinsically round galaxies from core-dominated ones, a test the paper leaves implicit.","If photo-z scattering at $z \\sim 10$ is as strong as the paper suggests, the true size at $z \\sim 10$-$12$ could be even smaller or larger than measured; targeted spectroscopic follow-up of the dip at $z \\sim 10$ would sharpen the evolutionary interpretation."],"forward_implications":["If the trend is real, the small sizes of very high-redshift galaxies are a genuine evolutionary phase, not a selection or resolution artifact.","Galaxies of a given mass being roughly 53% larger only 400 Myr later implies rapid size growth in the first half-billion years of galaxy assembly.","A merger fraction near 0.12 with little redshift evolution from $z \\sim 4$ to $z \\sim 12$ implies that mergers were already assembling galaxies in the Epoch of Reionization.","The trend also holds for Petrosian radii, so galaxies grow in total size, not just in their half-light radii.","The excess of round, compact objects suggests that many early galaxies were either intrinsically round or dominated by a bright core whose outer diffuse component is too faint to detect."],"supporting_citations":[{"why":"Supplies the EPOCHS v1 catalogue of high-redshift candidates from which the 520-galaxy sample is drawn.","marker":"Conselice et al. (2024)"},{"why":"Provides GALFIT, the two-dimensional light-profile fitting code used to measure the Sersic parameters.","marker":"Peng et al. (2002)"},{"why":"Provides Morfometryka, used to measure non-parametric concentration and asymmetry.","marker":"Ferrari et al. (2015)"},{"why":"Defines the asymmetry merger criterion $A > 0.35$ used to classify mergers.","marker":"Conselice (2003)"},{"why":"Provides the lower-redshift JWST size evolution and Sersic-index comparison that this work extends.","marker":"Ormerod et al. (2024)"},{"why":"Supplies the redshift-simulation approach used to test whether observed trends are redshift effects.","marker":"Whitney et al. (2021)"},{"why":"Presents the AREIA code used for the artificial redshift experiments.","marker":"Tohill et al. (2021)"},{"why":"Provides ASTRID mock observations whose size-mass evolution is compared with the observed trend.","marker":"LaChance et al. (2024)"},{"why":"Supplies the axis-ratio versus size 'banana' comparison and the lower-redshift $b/a$ distribution.","marker":"Pandya et al. (2024)"}],"fun_headline_variants":["JWST: 521 galaxies trace size drop to z=12.5","Rounder galaxies at high z: JWST sees excess axis-ratios","Merger fraction constant at 12% from z=4 to z=12","Size-mass relation holds at z~12, galaxies shrink with time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that the photometric redshifts and SED-derived stellar masses are accurate enough to bin galaxies by redshift and mass; the paper itself notes that photo-z scattering occurs near $z \\sim 10$ because the Lyman break lies between the F115W and F150W filters, and only a small fraction of the sample has spectroscopic confirmation.","fun_headline_variants_meta":{"raw":{"variants":["JWST: 521 galaxies trace size drop to z=12.5","Rounder galaxies at high z: JWST sees excess axis-ratios","Merger fraction constant at 12% from z=4 to z=12","Size-mass relation holds at z~12, galaxies shrink with time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000814,"raw_usage":{"total_tokens":3698,"prompt_tokens":1206,"completion_tokens":2492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":822,"completion_tokens_details":{"reasoning_tokens":2410}},"tokens_in":822,"tokens_out":2492,"duration_ms":15938,"temperature":1.0,"reasoning_tokens":2410,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:44:47.350562+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"If a substantial fraction of the $z \\sim 9$-$12$ candidates were reassigned to neighboring redshifts by JWST/NIRSpec spectroscopy, and the size-redshift or size-mass trends flattened after reassignment, the central claim would fail. A direct test would take the 520 galaxies, measure spectroscopic redshifts for as many as possible, and re-derive the $R_e(z)$ power law and the size-mass normalization with the secure redshifts only; if the smaller sizes at $z \\sim 10$-$12$ disappear, the evolutionary interpretation is wrong.","supporting_citations":[],"review_version":1}