{"id":"1923086e-efea-4048-abb3-f4ce251a0ae6","arxiv_id":"2507.23654","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"JWST imaging of 817 Lyman-alpha emitters at z=3-7 shows a bimodal population: low-mass isolated 'pristine' galaxies and massive 'merger-driven' galaxies, with a 39% late-stage merger fraction.","lead":"A JWST survey of 817 early-universe galaxies that emit Lyman-alpha light finds that about 39% are in the final stages of merging and 61% show some sign of interaction. The paper splits these galaxies into two classes, low-mass pristine systems and massive merger-driven systems, suggesting mergers help make galaxies visible in Lyman-alpha.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The unquantified mass and redshift dependence of visual merger classification threatens the central pristine/merger-driven dichotomy; a completeness simulation is required.","rationale":"The reader's weakest assumption correctly identifies the point on which the central claim depends: the visual morphological classification is asserted to be reliable, but its completeness is never demonstrated as a function of mass, surface brightness, or redshift. I find the same concern in the paper itself: Section 5.1 argues that late-stage mergers are identified by compact multiple nuclei and hence are robust to cosmic dimming, but this is an assertion rather than a measured completeness curve. Section 4.1's nonparametric diagrams provide useful supporting evidence that securely classified mergers have distinct G, M20, and AO values, but they do not test whether the visual classifier misses faint mergers or misclassifies bright isolated clumpy galaxies. The central evolutionary dichotomy maps directly onto the mass dependence of the merger fraction, so any mass-dependent selection effect in the classifier propagates into the headline result. My concrete check—a blind injection-recovery experiment with realistic Sersic profiles, noise, and classifier re-run—would settle the question cleanly. I do not see a more fundamental flaw: the SED fitting, photometry, and spectroscopic sample construction are standard, the nonparametric measurements give some independent support, and the Iani et al. (2024) bimodality in M*-sSFR is an independent observational anchor. The main gap is specific and testable, so a conditional verdict remains appropriate; if the completeness simulation shows strong mass/redshift-dependent recovery, the verdict should move toward rejection of the central dichotomy claim, but the current reader's conditional assessment is the right intermediate position.","tokens_in":13726,"tokens_out":3837,"duration_ms":46389,"concrete_test":"Run an injected-galaxy completeness simulation using real NIRCam F115W images: place pairs of compact Sersic profiles (simulating unresolved or partially resolved late-stage mergers at physical separations of 2–10 kpc) and single Sersic profiles (simulating isolated galaxies) into the observed noise, spanning log(M*/Msun)=7–9.5, z=3–7, and the observed S/N distribution of the sample. Have the same two classifiers re-run the visual classification blind, and compute merger recovery rate and isolated-to-merger false-positive rate as functions of mass, redshift, and S/N. If recovery strongly increases with mass or decreases with redshift in the same sense as Figure 4's trends, the mass and redshift dependence of the merger fraction—and hence the pristine/merger-driven split—is not secure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that LAEs split into low-mass Pristine isolated and massive Merger-driven populations—rests entirely on the morphological classification in Section 3.3. The paper asserts that late-stage mergers are identified by multiple compact nuclei and are therefore resilient to surface brightness dimming (Section 5.1), but it never quantifies how classification completeness and purity vary with stellar mass, UV luminosity, redshift, or S/N. A low-mass, high-redshift galaxy with two faint close components will be seen as a single compact source and classified as isolated; a bright massive galaxy will show substructure regardless of whether it is a true late-stage merger. The nonparametric validation in Section 4.1 applies only to sources with S/N>3 and shows that mergers occupy a certain region of G-M20/AO-M20 space; it does not calibrate the visual classifier or measure false-positive/negative rates as functions of the very quantities (M*, M_UV, z) that drive the paper's trends. Since the mass dependence of the merger fraction is the observational basis for the two evolutionary classes, this completeness uncertainty is load-bearing: if low-mass mergers are preferentially missed and bright massive systems preferentially overclassified, the claimed dichotomy could be partially an observational artifact.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper compiles 817 spectroscopically confirmed Lyman-alpha emitters at 3<z<7 in GOODS-S from MUSE, VANDELS, and CANDELS-z7, fits their SEDs with CIGALE, measures non-parametric morphological indicators (Gini, M20, outer asymmetry) from JWST/NIRCam imaging, and visually classifies them into isolated, pair, compact-merger, diffuse-merger, and tidal-feature systems. It reports a late-stage merger fraction of 39.4%±2.5% and an interaction fraction of 60.6%±6.3%, with strong trends toward higher merger fractions at lower redshift, higher stellar mass, and brighter UV luminosity. It further reports a bimodal M*-sSFR distribution in which isolated LAEs cluster at low mass and high sSFR while late-stage mergers cluster at higher mass and slightly lower sSFR, and interprets this as evidence for two evolutionary classes: 'Pristine LAEs' and 'Merger-driven LAEs'. The central claim is that mergers are a dominant driver of LAE evolution across the first two billion years.","tokens_in":13956,"tokens_out":5628,"duration_ms":55932,"significance":"If the central claims hold, this would be a valuable large-sample JWST study: 817 spectroscopically confirmed LAEs with NIRCam imaging is a substantial resource, and the explicit link between merger morphology and the previously reported M*-sSFR bimodality (Iani et al. 2024) is interesting and timely. The paper's strengths include the multi-survey spectroscopic sample, the use of quantitative non-parametric indicators alongside visual classification, and the attempt to connect observed merger fractions to simulations of LAE demographics. However, the central dichotomy—Pristine versus Merger-driven populations—rests on visual morphological classifications whose completeness and purity are never quantified as functions of mass, luminosity, redshift, or S/N. Because the merger fraction's mass and luminosity dependence is the observational basis for the two-class interpretation, this missing calibration is load-bearing. The paper also defers key sample-selection details to in-preparation works, and contains several internal numerical inconsistencies that need clarification.","major_comments":[{"comment":"The central merger and interaction fractions, and the resulting Pristine/Merger-driven dichotomy, rest on visual classifications whose completeness and purity are never quantified as functions of stellar mass, M_UV, redshift, or S/N. Section 5.1 argues that late-stage mergers are identified via multiple compact nuclei and are therefore resilient to surface-brightness dimming, but this is an assertion, not a measurement: no injection/recovery simulation or comparison with an independent automated classifier is presented. Since Figure 4 shows the merger fraction rising steeply with M* and luminosity, the possibility remains that low-mass, high-redshift compact mergers are classified as isolated because they are unresolved, while bright massive systems are preferentially resolved into multiple components and classified as mergers. A completeness simulation, or at a minimum a quantitative S/N- and size-dependent purity analysis, is required to support the mass dependence of the merger fraction that underpins the bimodal evolutionary interpretation.","section":"Section 3.3, Section 5.1, Figure 4"},{"comment":"The non-parametric validation in Section 4.1 does not calibrate the visual classification. Figure 3 plots only securely classified sources with S/N>3 and shows that late-stage mergers occupy the G-M20/AO-M20 selection regions, but this is a consistency check, not a completeness or contamination estimate: it does not report the fraction of visually classified mergers recovered by the automated criteria, the false-positive rate among isolated galaxies, or how these rates vary with M*, M_UV, z, and S/N. The statement that 'late-stage mergers separate cleanly from isolated systems' is qualitative; without a confusion matrix or efficiency curves, the quantitative agreement claimed in the text is not established. The authors should either provide these measurements or restrict the text to the observed segregation.","section":"Section 4.1, Figure 3"},{"comment":"The abstract and Section 4.3 claim negligible UV-slope differences between late-stage mergers and isolated LAEs 'at fixed M* and M_UV', but Figure 5 shows only separate binned relations versus M* and M_1500, not a matched sample controlling both quantities simultaneously. Given that the merger fraction depends strongly on both M* and M_UV (Figure 4), the small Δβ in one-dimensional bins could be a residual covariate effect. The authors should either construct matched control samples in the joint (M*, M_UV) space or rephrase the claim to say that the binned relations are consistent within the stated uncertainties.","section":"Section 4.3, Figure 5"},{"comment":"There are numerical inconsistencies in the reported fractions that need clarification. The abstract states overall late-stage merger and interaction fractions of 39.4%±2.5% and 60.6%±6.3%, while Section 4.2 describes the merger fraction as rising from ~15% at z>6 to 45-50% at z=3 and the interaction fraction from ~45% to 70%. Section 5.3 then refers to 'the exceptionally high merger fraction observed in massive LAEs (f_merger ~80%)', which appears to contradict Figure 4 where the late-stage merger fraction at log(M*/M_sun)>9 is ~50% and only the interaction fraction reaches ~80%. Please specify which quantity is meant in each place and ensure the figure and text use consistent definitions.","section":"Section 4.2, Section 5.3, Abstract"},{"comment":"The sample selection function is not presented in the manuscript: the detailed criteria are deferred to Song et al. (2025, in preparation), and the imaging reduction to Liu et al. (2025, in preparation). Because the MUSE-Wide, MUSE-Deep, VANDELS, and CANDELS-z7 surveys have different depths, areas, and selection functions, the redshift and mass dependence of the merger and interaction fractions in Figure 4 could be biased by sample construction. At minimum, the paper should provide the number of sources from each survey, the redshift and mass distributions per survey, and a statement of how survey depth limits low-mass completeness as a function of redshift.","section":"Section 2.3"}],"minor_comments":[{"comment":"The Introduction states a sample of 819 LAEs, while the Abstract and Section 2.3 state 817; please correct the inconsistency.","section":"Section 1 vs. Abstract and Section 2.3"},{"comment":"The statement that discrepancies occurred in 'less than 20% of cases' should be replaced with the exact number of discrepant objects, the number resolved by consensus, and the number excluded as tentative, so the reader can assess the robustness of the final 'secure' sample.","section":"Section 3.3"},{"comment":"The table header contains the typo 'Late-stege Merger' and the p-values are reported as '0' or '0.0'; continuous test statistics cannot yield exactly zero, so the entries should be given as p<0.001 or with actual rounded values.","section":"Table 1"},{"comment":"The headings/text contain typos: 'indentifaication' in the Section 5.2 heading and 'classifed' in Section 5.3; these should be corrected.","section":"Section 5.2 and Section 5.3"},{"comment":"The axis labels 'log(M* /M )' and 'M*' are missing the solar-mass subscript/symbol and should read log(M*/M_sun) and M* in the proper notation.","section":"Figure 4 and Figure 5"},{"comment":"The bound 'f_true < 0.10' is introduced without derivation or uncertainty; please explain how this upper limit is obtained from the observed ~10% tidal-feature fraction and the assumed incompleteness.","section":"Section 5.1"},{"comment":"The JWST program list ends with '654' in the Acknowledgements but Section 2.1 lists '6541'; this appears to be a truncation and should be corrected.","section":"Acknowledgements"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about unquantified completeness of the visual morphological classification is valid and is the main reason for major revision. The paper's central claim is potentially important, but the mass and luminosity dependence of the merger fraction must be demonstrated to be robust against resolution and S/N biases before the Pristine/Merger-driven interpretation can be accepted. The numerical inconsistencies in the reported fractions and the deferred sample-selection details should also be fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real new thing here is the sample: 817 spectroscopically confirmed LAEs at 3<z<7 with JWST/NIRCam imaging, visually classified and cross-checked with nonparametric indicators. That is the largest systematic morphological census of LAEs at these redshifts, and the numbers that come out of it—39% late-stage mergers, 61% interacting, with strong mass and luminosity dependence—are a genuinely useful reference point. The paper also does something honest: it validates the visual classes against G-M20 and AO-M20 diagnostics, and the separation in those planes is reasonably clean. The bimodality in M*-sSFR is real and already established by Iani et al. 2024, so they are not manufacturing it. Credit where due: this is a solid, workmanlike observational study with a clear presentation.\n\nThe soft spots are real and I agree with the stress-test note. The central dichotomy rests entirely on visual classification by two human classifiers, with no public catalog and no completeness simulation. The paper argues that late-stage mergers are identified by multiple compact nuclei and are therefore resilient to surface brightness dimming (Section 5.1), which handles the redshift trend reasonably well. But it does not address the mass-dependent detectability problem: a low-mass, high-z pair with two faint close components will be classified as a single isolated source, while a bright massive galaxy will show substructure regardless of whether it is a true late-stage merger. Since the mass dependence of the merger fraction is the observational basis for the Pristine vs Merger-driven split, this is load-bearing, not a minor footnote. The paper needs either a completeness simulation using simulated JWST images or a calibration of the visual classifier as a function of M*, M_UV, and S/N. Without that, the specific fractions and the two-class interpretation are plausible but not fully demonstrated.\n\nI also think the causal language goes a bit beyond the evidence. The paper shows a morphological association; the claim that mergers 'drive' LAE evolution, enhance LyC escape, and transform ordinary SFGs into detectable LAEs leans heavily on external results (Witten, Zhu, etc.). That is a reasonable interpretation, but it is not established by this dataset, and the abstract oversells it.\n\nMinor points: the interaction fraction includes pairs and tidal features, and tidal features are acknowledged to suffer from cosmic dimming; the hand-wave that they are only ~10% so it doesn't matter is a bit quick. Sample selection from three surveys with different depths (MUSE-Deep, MUSE-Wide, VANDELS, CANDELS-z7) could bias the redshift evolution, and that is not quantified either.\n\nWho is this for? Anyone working on LAE demographics, merger fractions at high redshift, or the reionization-probe interpretation of LAEs. It is a useful dataset and a plausible qualitative result, but the headline fractions need a completeness control before I would cite them as numbers. A serious referee should be engaged; this deserves peer review, not desk rejection. The referee should ask for the completeness simulation and the public catalog, and for toning down the causal language.","headline":"Large, useful morphological census of 817 LAEs, but the pristine-versus-merger dichotomy needs a completeness simulation before the headline fractions are taken at face value.","tokens_in":14577,"tokens_out":1954,"would_cite":true,"duration_ms":22656,"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":"Using JWST/NIRCam imaging of 817 Lyman-alpha emitters at 3<z<7, this paper shows that mergers split the LAE population into low-mass pristine systems and massive merger-driven systems.","keywords":["Lyman-alpha emitters","galaxy mergers","galaxy evolution","JWST","high-redshift galaxies","galaxy morphology","pristine LAEs","merger-driven LAEs"],"falsifier":"A direct completeness test is to take the JWST/NIRCam images of classified massive mergers, degrade them to the resolution and signal-to-noise of the faint low-mass LAEs, and ask whether the merger signatures survive independent reclassification. If the merger fraction at fixed stellar mass drops when images are degraded to a common S/N, the reported mass dependence—and with it the pristine/merger-driven dichotomy—would be an observational artifact. Alternatively, comparing the visual merger fraction at matched S/N with a machine-learning classifier trained on simulated mergers would settle whether the classification is biased.","tokens_in":13510,"feed_emoji":"🔭","tokens_out":7027,"duration_ms":62216,"temperature":0.7,"pith_summary":"This paper uses JWST high-resolution imaging to ask whether galaxy mergers are central to the evolution of Lyman-alpha emitters, the star-forming galaxies used as tracers of the early universe. By visually classifying 817 spectroscopically confirmed LAEs at redshifts 3 to 7, it reports that nearly 40% are in the late stages of a merger and about 61% are in some form of interacting system, with merger fractions rising steeply for massive and bright galaxies. The LAE population splits into two distinct classes in the plane of stellar mass versus specific star-formation rate: low-mass isolated systems that appear young and gas-rich, and massive systems whose structures betray merger activity. The paper argues these two classes represent different evolutionary pathways, making mergers a central driver of Lyman-alpha visibility and mass assembly in the first two billion years. If the classification is unbiased, LAEs cannot be treated as a single homogeneous population.","feed_headline":"Mergers split early Lyman-alpha emitters into two classes","feed_subtitle":"Low-mass isolated emitters look pristine, while mergers dominate massive systems and boost Lyman-alpha escape.","key_machinery":"The argument rests on a five-category visual morphological classification of LAEs performed on JWST/NIRCam pseudo-color images: isolated, pair, compact-merger, diffuse-merger, and tidal-feature systems. Late-stage mergers combine compact and diffuse mergers, and interacting systems combine pairs, mergers, and tidal features. The non-parametric diagnostics Gini, $M_{20}$, and outer asymmetry $A_O$—quantities that measure flux concentration, spatial clumping, and outer asymmetry—are used to show that visual mergers satisfy the same selection boundaries calibrated for lower-redshift massive galaxies, indicating the signatures (double nuclei, tidal tails, diffuse multiple components) are robust to surface-brightness dimming. The bimodality in the $M_*$–sSFR plane is established from SED-fitting-derived stellar masses and star-formation rates, and the paper ties the two peaks to the morphological classes through K-S tests comparing stellar mass and sSFR distributions of young/old LAEs versus isolated/merger systems.","core_discovery":"The paper reports a systematic morphological census of Lyman-$\\alpha$ emitters (LAEs) in the GOODS-S field based on JWST/NIRCam imaging. Among 817 spectroscopically confirmed LAEs at $3<z<7$, it identifies late-stage mergers with a fraction of $39.4\\%\\pm2.5\\%$ and interacting systems (pairs, mergers, and tidal features) with a fraction of $60.6\\%\\pm6.3\\%$. The merger fraction rises sharply with stellar mass and UV luminosity, from roughly 10% among the faintest, lowest-mass systems to more than 70% for massive ($\\log(M_*/M_\\odot)>9$) and bright ($M_{\\rm UV}<-20$) ones. In the stellar-mass–specific-star-formation-rate ($M_*$–sSFR) plane, the full sample is bimodal: isolated LAEs peak at $\\log(M_*/M_\\odot)\\approx7.8$ with $\\log({\\rm sSFR/yr^{-1}})\\approx-7.4$, while late-stage mergers peak at $\\log(M_*/M_\\odot)\\approx8.6$ with $\\log({\\rm sSFR/yr^{-1}})\\approx-7.6$. The paper interprets this as two evolutionary classes—Pristine LAEs (low-mass, isolated, early-stage galaxies with minimal merger interactions) and Merger-driven LAEs (massive systems where mergers enhance star formation and enable Lyman-$\\alpha$ escape).","pith_inferences":["If the two-class picture is correct, the common practice of stacking all LAEs to measure average properties will blend a young, low-mass, merger-free population with an old, massive, merger-dominated one; interpretations of averaged Lyman-alpha escape fractions from such stacks would need revision.","A testable extension is to compare these visual merger classifications with dynamical signatures from JWST/NIRSpec or ALMA kinematics; genuine mergers should show velocity offsets or disturbed rotation, and a mismatch would indicate classification bias.","The transition near $10^{8.5}\\,M_\\odot$ could correspond to a halo mass at which merger rates become dominant; cosmological simulations that predict merger rates as a function of halo mass for LAEs could verify or refute this threshold."],"forward_implications":["Nearly all massive LAEs ($\\log(M_*/M_\\odot)>8.5$) show merger or interaction signatures, so any model of Lyman-alpha escape in bright, evolved galaxies must include merger-triggered gas dispersal and feedback.","Low-mass, isolated LAEs constitute a distinct pristine population, meaning faint-end LAE surveys trace early galaxy assembly without significant merger contamination.","The merger fraction increases from about 15% at $z>6$ to about 45% at $z\\sim3$, matching hierarchical-assembly predictions that massive, older LAEs emerge as halos grow.","The dependence of merger rate on stellar mass and UV luminosity implies that mass- and luminosity-selected LAE samples probe different physical populations with different escape physics.","The morphological homology between $z<1$ mergers and $z>3$ LAEs suggests the same merger indicators can be applied across cosmic time up to the epoch of reionization."],"supporting_citations":[{"why":"Supplies the deepest MUSE-Deep spectroscopic sample of LAEs, the backbone of the 817-galaxy sample.","marker":"Bacon et al. (2023)"},{"why":"Provides MUSE-Wide LAE detections with equivalent-width measurements that feed the sample selection.","marker":"Kerutt et al. (2022)"},{"why":"VANDELS survey identifications extend the redshift and luminosity coverage of the LAE compilation.","marker":"Talia et al. (2023)"},{"why":"CANDELS-z7 supplies the highest-redshift LAEs used in the sample.","marker":"Pentericci et al. (2018)"},{"why":"Defines the Gini and M20 diagnostics that independently validate the visual merger classification.","marker":"Lotz et al. (2004)"},{"why":"Establishes the Gini–M20 selection criterion used to identify late-stage mergers.","marker":"Lotz et al. (2008)"},{"why":"Introduces the outer-asymmetry AO–M20 plane that captures both compact and diffuse mergers.","marker":"Ren et al. (2023)"},{"why":"Provides redshift-dependent corrections to the AO threshold used in the non-parametric diagnostics.","marker":"Ren et al. (2024)"},{"why":"Theoretical model predicting young, low-mass and old, massive LAE populations that the paper identifies with pristine and merger-driven classes.","marker":"Shimizu & Umemura (2010)"},{"why":"Reports the observed bimodality in the sSFR–M* plane that this paper connects to morphology.","marker":"Iani et al. (2024)"}],"fun_headline_variants":["JWST survey: two LAE classes, pristine vs merger-driven","Mergers drive massive Lyman-alpha emitters, JWST shows","817 LAEs: low-mass isolated, high-mass merger-driven","Bimodal LAE population: pristine and merger-driven at z=3-7"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The bimodal split and the mass dependence of the merger fraction rest on the assumption that visual morphological classification is complete and unbiased across the sample—specifically, that low-mass isolated LAEs are not unresolved late-stage mergers and that massive LAEs are not preferentially classified as mergers simply because their higher surface brightness makes multiple components easier to see.","fun_headline_variants_meta":{"raw":{"variants":["JWST survey: two LAE classes, pristine vs merger-driven","Mergers drive massive Lyman-alpha emitters, JWST shows","817 LAEs: low-mass isolated, high-mass merger-driven","Bimodal LAE population: pristine and merger-driven at z=3-7"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000426,"raw_usage":{"total_tokens":2336,"prompt_tokens":1251,"completion_tokens":1085,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":867,"completion_tokens_details":{"reasoning_tokens":1006}},"tokens_in":867,"tokens_out":1085,"duration_ms":11490,"temperature":1.0,"reasoning_tokens":1006,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:28:34.138309+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct completeness test is to take the JWST/NIRCam images of classified massive mergers, degrade them to the resolution and signal-to-noise of the faint low-mass LAEs, and ask whether the merger signatures survive independent reclassification. If the merger fraction at fixed stellar mass drops when images are degraded to a common S/N, the reported mass dependence—and with it the pristine/merger-driven dichotomy—would be an observational artifact. Alternatively, comparing the visual merger fraction at matched S/N with a machine-learning classifier trained on simulated mergers would settle whether the classification is biased.","supporting_citations":[{"cited_title":"2022, A&A, 659, A183","cited_arxiv_id":null,"evidence_quote":"Provides MUSE-Wide LAE detections with equivalent-width measurements that feed the sample selection."},{"cited_title":"2023, A&A, 678, A25","cited_arxiv_id":null,"evidence_quote":"VANDELS survey identifications extend the redshift and luminosity coverage of the LAE compilation."},{"cited_title":"J., Garilli, B., et al","cited_arxiv_id":null,"evidence_quote":"CANDELS-z7 supplies the highest-redshift LAEs used in the sample."},{"cited_title":"S., et al","cited_arxiv_id":null,"evidence_quote":"Introduces the outer-asymmetry AO–M20 plane that captures both compact and diffuse mergers."},{"cited_title":"2010, MNRAS, 406, 913","cited_arxiv_id":null,"evidence_quote":"Theoretical model predicting young, low-mass and old, massive LAE populations that the paper identifies with pristine and merger-driven classes."}],"review_version":1}