{"id":"01e652e3-6720-40cd-bb12-1286a18322d3","arxiv_id":"2607.23062","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A new ratio of short- to long-term star formation rates splits Lyman-alpha emitting galaxies at cosmic noon into the same three archetypes that full star formation histories reveal.","lead":"This paper maps 270 distant star-forming galaxies from the HETDEX survey onto a new 'star formation stochasticity diagram' and finds they split into three types based on whether their current burst is their first, their largest, or an echo of an older one. The diagram is a candidate simple diagnostic for galaxy evolution surveys.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed GMM recovery of SFH archetypes is only demonstrated visually; no quantitative classification test links the three components to archetype labels, and the input ratio is constructed from the same recent-vs-past SFR contrast used to define the archetypes.","rationale":"The reader's verdict is CONDITIONAL and identifies the constant fret=0.67 as the weakest assumption. I do not think fret is the most load-bearing issue for the central claim: for the LAE sample itself, SFRavg is computed directly from Mformed in Eq. 2, so an age-dependent fret would not shift the GMM input; fret only enters when converting the comparison correlation from Mérida et al. The more serious gap is that the claimed recovery of archetypes by the GMM is supported only by the visual histogram overlay in Fig. 4, with no quantitative classification test. This matters because the ratio log(SFR10/SFRavg) is constructed from the same recent-vs-past star-formation contrast used to define the archetypes in §3.3, making some separation expected. The missing test is directly addressable and would settle whether the headline claim is real. Since this concern points to the same conditional status the reader already assigned, I recommend no change to the verdict.","tokens_in":17055,"tokens_out":5464,"duration_ms":57573,"concrete_test":"Perform 5-fold cross-validation on the 270 LAEs. In each fold, fit a 3-component GMM to log(SFR10/SFRavg) on the training set, then assign each held-out galaxy to its most probable component. Compare these assignments to the archetype labels inferred from the full SFHs (First/Dominant/Nondominant) and report the confusion matrix, accuracy, and adjusted Rand index. Compare against (a) the majority-class baseline (≈61%) and (b) a null distribution obtained by permuting archetype labels. If held-out accuracy is not significantly above baseline, the claim that the stochasticity ratio recovers the archetypes without full SFHs is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in §5.2 and Figure 4 is that a 3-component GMM on log(SFR10/SFRavg) 'recovers' the three archetypes defined from full SFHs. This is the only evidence offered for the abstract's statement that the ratio can summarize mass assembly without viewing full SFHs. Two issues make this insecure. First, the feature is not independent of the classification: the archetype definitions in §3.3 (First Burst vs. Dominant Burst vs. Nondominant Burst) are explicitly thresholds on the contrast between star formation in the last 200 Myr and earlier star formation, and log(SFR10/SFRavg) is a smoothed version of that same contrast. A separation in this ratio is therefore expected by construction. Second, the comparison in Figure 4 is purely visual: histograms of the three archetypes are overlaid on the fitted Gaussian components, but no classification accuracy, confusion matrix, mutual information, or null-model comparison is reported. Because the GMM components and the archetype labels are both derived from the same Dense Basis SFHs, even a strong visual match does not establish that the ratio alone can classify galaxies without the full SFH. The practical claim—summarizing stellar mass assembly with one number—requires a quantitative out-of-sample validation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses 270 HETDEX-detected LAEs in the COSMOS field with CANDELS photometry, reconstructs star formation histories with the Dense Basis method, and classifies them into the three SFH archetypes from Firestone et al. (2025): First Burst, Dominant Burst, and Nondominant Burst. It introduces a 'Star Formation Stochasticity Diagram' based on log(SFR10/SFRavg) and log(SFR100/SFRavg), compares LAEs with a generic-galaxy correlation from Mérida et al. (2026), and applies a Gaussian Mixture Model to the stochasticity metric. The central claim is that a 3-component GMM on log(SFR10/SFRavg) recovers the three archetypes, so that the ratio summarizes stellar mass assembly without viewing full SFHs.","tokens_in":17406,"tokens_out":6185,"duration_ms":61443,"significance":"If substantiated, the paper would provide a useful one-dimensional summary statistic for the diverse star-formation histories of LAEs and a diagnostic diagram that removes the redshift evolution of the SFR–M* relation. The analysis benefits from a spectroscopically selected sample, a non-parametric SFH reconstruction method, and a comparison catalog analyzed with consistent assumptions. The archetype fractions are consistent with the previous ODIN results, adding confidence to the empirical archetype framework. However, the headline claim rests on a purely visual comparison in Figure 4, and the archetype definitions already encode the same recent-versus-past SFR contrast that the stochasticity metric measures. The paper also leaves a normalization correction unspecified and adopts a single mass-retention fraction that it admits is age-dependent. These issues need to be resolved before the central claim can be accepted.","major_comments":[{"comment":"The claim that a 3-component GMM on log(SFR10/SFRavg) 'recovers' the three SFH archetypes is supported only by a visual overlay of histograms in Figure 4. No classification accuracy, confusion matrix, mutual information, or null-model comparison is reported. Because the archetype definitions (§3.3) are thresholds on the contrast between star formation in the last 200 Myr and earlier star formation, and log(SFR10/SFRavg) is a smoothed version of that same contrast, a separation in this ratio is expected by construction. Please provide a quantitative assignment of GMM components to archetype labels (e.g., using posterior probabilities and a contingency table) and a test of whether the agreement exceeds a randomized labeling or a mass/redshift-matched null.","section":"§5.2 / Fig. 4"},{"comment":"The comparison to the Mérida et al. (2026) correlation uses an unspecified 'simple correction to the correlation's normalization based on the behavior of a subset of our LAEs' (§4.1). The correction's functional form, sample size, and selection are not given. In addition, the conversion via Eq. (3) assumes a single mass-retention fraction fret=0.67, yet §5.1 states that mass losses exceed 30% and are 'heavily dependent on the age of a galaxy's stars.' If fret varies systematically across the archetypes, the metric is shifted differentially and the GMM components could be partly artifacts. Please specify the normalization correction, test sensitivity to fret, and consider an age-dependent or marginalised treatment.","section":"§4.1, §5.1–5.2"},{"comment":"Equation (3) gives log(SFR100/SFRavg) = (β−1) log(M*) + α + log(fret t_univ). Thus the converted reference correlation remains a function of stellar mass unless β=1. The white horizontal reference line in Figure 3 cannot hold for all LAEs unless a representative M* is specified or the correlation is marginalised over the sample mass distribution. The claim that the stochasticity diagram removes redshift evolution and enables direct comparison requires explicit treatment of this mass dependence; otherwise the placement of the reference line is ambiguous.","section":"Eq. (3) / §5.2"}],"minor_comments":[{"comment":"The abstract contains 'CANDELS\\null', likely a LaTeX error. Also, 'better 98% accuracy' should read 'better than 98% accuracy'.","section":"Abstract / §2.1"},{"comment":"BIC values and the fitted GMM parameters (means, variances, weights) should be tabulated for both the burst-sensitive and generic metrics. The current presentation does not allow readers to assess the strength of the 3-component preference.","section":"Figure 4"},{"comment":"For the generic metric, the BIC does not strongly prefer 3 components and the component count is fixed at 3 for comparison. This should be stated more prominently in the main text and abstract so that 'statistically motivated sub-populations' is not overclaimed.","section":"§5.2"},{"comment":"The archetype fractions (61%, 31%, 7%) are given without uncertainties. Bootstrap or posterior-sampling uncertainties would help assess the consistency with the ODIN fractions.","section":"§3.4"},{"comment":"SFR10 and SFR100 are used in §4.2 before being explicitly defined in the main text; a sentence in §4.1 or a footnote defining both timescales would improve readability.","section":"§4.1–4.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a good fit for the journal and I do not see a novelty or attribution problem. The main challenge is the lack of a quantitative link between the GMM components and the archetype labels; this is fixable with a classification test and by specifying the normalization and mass-retention assumptions. I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper does two things. First, it confirms the Firestone et al. (2025) SFH archetype fractions in an independent, spectroscopically selected HETDEX sample (61/31/7 vs 67/28/5). That is a real result. Second, it introduces a “star formation stochasticity diagram” — log(SFR/SFRavg) versus redshift, where SFRavg is formed mass divided by the age of the universe — and claims that a Gaussian mixture model on this ratio recovers the three archetypes without needing full SFH fitting.\n\nThe diagram is a sensible extension of the Scalo b parameter, and the paper is upfront about the lineage. The move to SFR10 instead of SFR100 for the burst-sensitive version is well motivated: the SFR100-M* diagram genuinely underrepresents young first-burst galaxies. The use of Dense Basis with consistent priors across the LAE and comparison samples (Mérida et al. 2026) is methodologically clean.\n\nThe soft spots are real, though not fatal.\n\nThe central claim — that the GMM components “recover” the archetypes — is supported only by overlaying histograms of the archetype ratio values on the fitted Gaussians. There is no confusion matrix, classification accuracy, or any quantitative link between GMM components and archetype labels. Given that the archetypes are themselves defined by the contrast between recent and past star formation, and the ratio log(SFR10/SFRavg) is a smoothed version of that same contrast, a separation is largely expected by construction. A reviewer should ask for a proper classification test: fit the GMM, assign each galaxy to its most probable component, and compare component membership to archetype labels.\n\nThe constant mass-retention fraction fret=0.67 is a single number applied to galaxies of very different ages. The paper itself notes that mass loss is heavily age-dependent and can exceed 30%. If real retention fractions vary systematically across the archetypes — young First Burst galaxies should retain more — the metric gets shifted by different amounts, and the GMM separation could partly be an artifact. Varying fret or marginalizing over it would address this.\n\nThe conversion of the Mérida et al. correlation to the stochasticity diagram includes an unspecified normalization correction “based on the behavior of a subset of our LAEs.” Which subset, and what criterion? That needs to be pinned down.\n\nThe generic metric version fixes the component count at 3 despite BIC not preferring it; that is disclosed, but it weakens the “statistically motivated” language.\n\nBottom line: the archetype confirmation and the new diagram are worth having. The one-number summary claim overreaches until the GMM-to-archetype mapping is validated quantitatively and the fret assumption is stress-tested. This deserves a serious referee, with the expectation of a moderate revision.\n\nRecommendation: send to peer review.","headline":"Useful new diagnostic and a clean confirmation of the LAE archetype fractions, but the headline GMM recovery claim rests on a visual overlay and a constant mass-retention assumption.","tokens_in":17970,"tokens_out":3827,"would_cite":true,"duration_ms":37994,"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":"This paper claims that one ratio — log(SFR10/SFRavg) — carries enough information to recover the three star-formation-history archetypes of Lyman-alpha emitting galaxies, so that stellar mass assembly can be summarized without reconstructin","keywords":["Lyman-alpha galaxies","star formation stochasticity","star formation history","SFR-M* diagram","Gaussian mixture model","cosmic noon","HETDEX","stellar mass assembly"],"falsifier":"Recompute log(SFR10/SFRavg) for the same 270 galaxies using per-galaxy mass-retention fractions derived from stellar-population models with age-dependent mass loss, then rerun the GMM on the new values; if the BIC no longer prefers three components, or the components no longer align with the archetypes assigned from full SFHs, the claim that the ratio alone recovers the archetypes fails.","tokens_in":16955,"feed_emoji":"🌌","tokens_out":7348,"duration_ms":68050,"temperature":0.7,"pith_summary":"This paper claims that one quantity — the ratio of a Lyman-alpha emitting galaxy's star formation rate over the last 10 million years to its average rate over cosmic time — is enough to summarize how the galaxy assembled its stars. The authors reconstruct flexible, non-parametric star formation histories for 270 HETDEX LAEs in the CANDELS/COSMOS field and confirm the same three archetypes previously found in ODIN LAEs: First Burst, Dominant Burst, and Nondominant Burst. They then show that an unsupervised Gaussian mixture model applied to the single ratio recovers those same three populations without ever seeing the full histories. The new Star Formation Stochasticity Diagram removes the redshift evolution that complicates the standard star formation rate–stellar mass diagram, and it reveals that rapidly rising First Burst galaxies are hidden by a 100-million-year averaging timescale. If correct, this gives observers a quick way to classify how distant galaxies are forming stars.","feed_headline":"A single ratio recovers three galaxy archetypes","feed_subtitle":"The 10-million-year vs. lifetime star-formation ratio recovers the three archetypes without full history fitting","key_machinery":"The central object is the Star Formation Stochasticity Diagram, plotting log(SFR/SFRavg) against redshift, where SFRavg = Mformed/tuniv = M*/(fret·tuniv) is the galaxy's average star formation rate since the Big Bang. The paper's 'burst-sensitive' version uses SFR10, the SFR averaged over the last 10 Myr, to catch young bursts that a 100 Myr average washes out. The accompanying Gaussian Mixture Model (a probabilistic model that decomposes a distribution into Gaussian subpopulations, with the number of components chosen by the Bayesian Information Criterion) is the mechanism that recovers the three archetypes from the ratio alone.","core_discovery":"On the paper's own terms, the discovery is that the empirical LAE SFH archetypes — First Burst, Dominant Burst, Nondominant Burst — are not just artifacts of detailed SFH fitting but appear as statistically preferred Gaussian components in the distribution of log(SFR10/SFRavg). The traditional SFR100–M* diagram averages the recent burst over 100 Myr and merges First Burst LAEs with ordinary galaxies; switching to a 10 Myr average separates them. Because the ratio normalizes out the redshift-dependent scaling of SFR with mass, the full HETDEX sample can be analyzed without binning, and the Gaussian mixture (selected by BIC) chooses three components for the burst-sensitive metric. The componen","pith_inferences":["If mass retention varies with stellar age, the single-fret assumption could shift the stochasticity metric differently for young and old archetypes, making the three Gaussian components partly a systematic effect rather than a purely physical separation.","The ratio could in principle be estimated from empirical tracers — e.g., an emission-line SFR for SFR10 and a mass-to-light or continuum measure for SFRavg — offering a faster way to classify large samples without full SED fitting.","The same diagram should apply to other emission-line-selected populations, such as [OII] or H-alpha emitters; if the three-component structure is universal, it would argue for a common stochasticity in gas accretion, not something peculiar to Lyα selection.","If the archetypes reflect genuinely different mass-assembly phases, the stochasticity diagram could be used to test galaxy formation simulations by comparing predicted distributions of log(SFR10/SFRavg) at fixed redshift."],"forward_implications":["First Burst LAEs, which appear ordinary in SFR100–M* space, are revealed as genuine starbursts when the 10 Myr average is used.","The stochasticity diagram allows the LAE sample to be analyzed as a whole over Δz~1.6 without redshift binning.","A Gaussian mixture on log(SFR10/SFRavg) recovers the same three archetypes as full SFH reconstruction, so classification can be done from the ratio alone.","The archetype frequencies are consistent between the HETDEX sample (61/31/7%) and the earlier ODIN sample (67/28/5%), supporting a common population description.","Existing SFR–M* correlations can be converted into the new diagram via the relation log(SFR/SFRavg) = (β−1)log M* + α + log(fret·tuniv), placing generic galaxies and LAEs on a common scale."],"fun_headline_variants":["SFR10/SFRavg ratio splits LAEs into three archetypes","New diagram uses 10-Myr SFR to reveal three LAE types","A single star-formation ratio recovers three galaxy archetypes","Stochasticity diagram: one ratio, three archetypes for LAEs","HETDEX: 10-Myr vs. lifetime SFR yields three LAE populations"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing assumption is that a single mass-retention fraction (fret = 0.67) converts formed stellar mass to present stellar mass for every galaxy; because the paper itself notes that mass loss is age-dependent, any systematic variation in retention across the three archetypes would shift the stochasticity metric differentially and could make the three Gaussian components an artifact of that assumption.","fun_headline_variants_meta":{"raw":{"variants":["SFR10/SFRavg ratio splits LAEs into three archetypes","New diagram uses 10-Myr SFR to reveal three LAE types","A single star-formation ratio recovers three galaxy archetypes","Stochasticity diagram: one ratio, three archetypes for LAEs","HETDEX: 10-Myr vs. lifetime SFR yields three LAE populations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00028,"raw_usage":{"total_tokens":1561,"prompt_tokens":869,"completion_tokens":692,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":592}},"tokens_in":613,"tokens_out":692,"duration_ms":7281,"temperature":1.0,"reasoning_tokens":592,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:41:15.444886+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute log(SFR10/SFRavg) for the same 270 galaxies using per-galaxy mass-retention fractions derived from stellar-population models with age-dependent mass loss, then rerun the GMM on the new values; if the BIC no longer prefers three components, or the components no longer align with the archetypes assigned from full SFHs, the claim that the ratio alone recovers the archetypes fails.","supporting_citations":[],"review_version":1}