{"id":"a55cf2bd-b9ff-4eae-b84a-42dc09a80851","arxiv_id":"2507.15944","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The authors show that Splash-like stellar populations in simulated Milky Way-mass galaxies track the mass of retrograde accreted stars rather than the single largest merger, and that the observed Splash chemistry is the high-eccentricity extension of the high-alpha disc.","lead":"The Splash, a retrograde stellar population in the Milky Way, is not necessarily the result of one major merger: the authors show that low-mass accreted galaxies on retrograde orbits can produce Splash-like populations in simulations, and that the observed Splash chemistry is a smooth extension of the high-alpha disc. The paper combines APOGEE abundances for 16 elements with ARTEMIS cosmological simulations to link the Splash to retrograde accretion history.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The observed e>0.6 and simulated Lz<0 Splash definitions are never calibrated; if they do not match, the r=0.92 correlation and MW comparisons in Sections 5.2-5.4 may not apply to the Milky Way Splash.","rationale":"The reader's weakest_assumption identifies exactly the load-bearing step: the equivalence between the observed Splash selection (e>0.6) and the simulated Splash selection (Lz<0). My reading of the full text confirms this is not merely a minor calibration issue but the hinge on which the paper's central conclusion depends. Section 5.1 introduces the substitution of Lz for eccentricity with only a footnote saying the correlation was 'verified' in the simulations; no plot, correlation coefficient, or overlap fraction is provided. Since the observed Splash is defined by eccentricity while the simulated Splash is defined by orbital direction, the two selections could diverge substantially in mixed populations. The rest of Section 5 (Fig. 13, Fig. 16, Fig. 17, Fig. 19) uses the Lz<0 definition for all simulated Splash fractions and then compares these fractions directly to observed Milky Way values, implicitly assuming the mapping is exact. This is a genuine external-validity risk, not an internal inconsistency in the simulations: the r=0.92 correlation may well be real for the simulated definition, but whether it constrains the Milky Way Splash is untested. My proposed check is direct and uses existing simulation outputs with the same potential/orbit calculations already used in the paper, so it is feasible without new simulations. The verdict CONDITIONAL remains appropriate because the concern is addressable: a calibration plot and a re-computed correlation would either validate or invalidate the mapping. I see no reason to reject the paper, and the reader's condition is already the right level of caution.","tokens_in":32770,"tokens_out":3361,"duration_ms":38322,"concrete_test":"For each of the seven ARTEMIS galaxies, select in-situ star particles in the same solar-neighbourhood volume (5<R<11 kpc, |Z|<3 kpc, [Fe/H]>-2.5) and apply the exact observed Splash criterion: the [Mg/Fe]-[Fe/H] high-alpha cut used for that galaxy (as in Fig. 12) and eccentricity e>0.6 computed in the simulation potential. Compute the resulting Splash fraction for each galaxy (and each azimuthal section) and compare it with the Lz<0-based fraction used in Fig. 16. Reconstruct the correlation plot and the G34/G44 comparison using the e>0.6-based fractions. If the Pearson r drops significantly from 0.92 or the ordering of G44 versus G34 changes, the assumed mapping fails and the paper's applicability to the Milky Way Splash is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim that Splash-like populations are ubiquitous and correlate with retrograde accreted fraction rests on treating the observed Splash (high-alpha, e>0.6; Section 2.1) and the simulated Splash (in-situ stars with Lz<0 at z=0; Section 5.2) as the same physical population. This equivalence is asserted in Section 5.1, where Lz is substituted for eccentricity because \"Lz correlates very well with eccentricity,\" with a footnote claiming this was verified in the simulations. No quantitative calibration is shown. The two criteria are not trivially equivalent: a star on a radial prograde orbit can have e>0.6 while Lz>0, and a star on a nearly circular retrograde orbit can have Lz<0 while e<0.6. Consequently, the simulated Splash fraction (in-situ retrograde fraction) may include low-e retrograde stars absent from the observed sample and exclude high-e prograde stars that are present. This mapping underpins every MW-to-simulation comparison: the confusion-map agreement (Fig. 7 vs Fig. 17), the [Mg/Fe]-Lz trends in Fig. 13, the age comparisons in Fig. 19, and the conclusion that low-mass retrograde mergers (e.g., G44) generate Splash-like features. If the mapping is poor, the strong r=0.92 correlation of Fig. 16 remains a valid internal simulation result, but it does not necessarily constrain the origin of the Milky Way's Splash. The manuscript provides no direct test of this equivalence despite it being the crucial link between the observational and simulation halves of the paper.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper combines APOGEE DR17/Gaia data with ARTEMIS simulations to characterize the Splash. The observational part selects high-[Mg/Fe] giants in the solar neighbourhood and divides them into high-alpha disc (e<0.6) and Splash (e>0.6) samples. Comparing 16 abundance ratios in four metallicity bins, the authors find statistically significant differences, with Splash stars more alpha-enhanced and lower in [Mn/Fe], but also show that these differences vary smoothly with eccentricity. The simulation part defines Splash-like populations in seven ARTEMIS galaxies as in-situ star particles on retrograde orbits (Lz<0), and finds that such populations exist in both GE/S-like and minor-accretion hosts, with a strong correlation (Pearson r=0.92) between the Splash fraction and the fraction of accreted retrograde stars. The authors conclude that retrograde minor mergers can generate Splash-like populations and that orbital orientation matters more than total merger mass.","tokens_in":33146,"tokens_out":6582,"duration_ms":71471,"significance":"The paper's observational analysis is a careful, high-precision chemo-dynamical comparison that extends previous Splash studies to 16 elements. The simulation result, if validated, would be an important step beyond the single-merger narrative: it would imply that Splash-like features are generic to accretion histories and that orbital orientation is a key parameter. Strengths include strict quality cuts, explicit correction for APOGEE log-g abundance systematics, bootstrap-based null chi-squared distributions, and use of a cosmological zoom-in suite with in-situ/accreted labels. However, the central observational-to-simulation link rests on an uncalibrated equivalence between eccentricity and angular-momentum selections, and the simulation sample is small and selected to have high accreted fractions. These caveats are load-bearing for the Splash ubiquity and correlation claims as applied to the Milky Way.","major_comments":[{"comment":"The manuscript equates the observed Splash definition (e>0.6, Section 2.1) with the simulated definition (in-situ stars with Lz<0, Section 5.2) on the assertion that \"Lz correlates very well with eccentricity\" (footnote 3). Quantitatively, the two criteria are not equivalent: a star on a highly eccentric prograde orbit has e>0.6 and Lz>0, while a nearly circular retrograde star has Lz<0 and e<0.6. Because every Milky Way comparison in Sections 5.2-5.4 (Fig. 16 correlation, Fig. 13 trend, Fig. 19 ages, and the Fig. 7 versus Fig. 17 confusion maps) uses the simulated Lz<0 definition, the applicability of the main conclusions to the observed Splash depends on this mapping. Please provide a quantitative calibration in ARTEMIS: for the same solar-neighbourhood volume, report the contamination and completeness of the Lz<0 selection relative to an e>0.6 selection (or apply the e>0.6 criterion directly to simulated star particles), and re-derive Fig. 16 and the confusion maps with the matched definition. Without this, the r=0.92 correlation remains an internal simulation result whose connection to the Milky Way Splash is unestablished.","section":"Section 5.1, footnote 3; Sections 5.2-5.4"},{"comment":"The headline correlation (Pearson r=0.92) is computed from 42 measurements that are not independent: 6 azimuthal sectors per galaxy, with galaxies strongly clustered in the plane. The significance and confidence interval of r need to account for this clustering, for example by block-bootstrap resampling whole galaxies or using a mixed-effects model. In addition, the two plotted quantities share a common Lz<0 orientation definition at z=0, so a correlation is partly expected if a retrograde merger both deposits retrograde accreted stars and heats the in-situ disc in the same rotational sense. The paper does not control for this shared-orientation effect, for example by comparing with a null that randomizes the sign of Lz, or by partial correlation with the total accreted fraction. Please add such a control before claiming that retrograde orientation specifically drives Splash formation.","section":"Section 5.2, Fig. 16"},{"comment":"The conclusion that Splash-like populations are \"ubiquitous\" (abstract and Section 5.4) is drawn from seven ARTEMIS galaxies, and the paper states that \"all systems selected here have high total accreted fractions ... >40%\" (Section 5). This selection enriches the sample in accretion-dominated systems and does not justify statements about Milky Way-mass galaxies in general. The ubiquity claim should either be restricted to the selected class of accretion-rich galaxies or supported by a test on a larger, unselected ARTEMIS sample (for example, the full 45-galaxy suite). As written, the claim overreaches the data.","section":"Section 5.2 and Section 5.4"}],"minor_comments":[{"comment":"The text refers to \"six simulated galaxies\" in two places, but the paper defines and uses seven (G29, G30, G34, G42, G17, G19, G44). Please correct the inconsistency.","section":"Sections 5.1 and 6"},{"comment":"The y-axis label reads \"Splash fractoin\"; this should be \"Splash fraction\".","section":"Fig. 7"},{"comment":"All p-values are printed as 0.0; please report them as <0.001 or with a precision consistent with the bootstrap method, since a literal zero probability is not a meaningful statement.","section":"Fig. 9"},{"comment":"The Splash fraction in simulations is defined as the ratio of in-situ retrograde to prograde stars, while the abstract calls it the \"mass fraction of Splash stars\". These are different quantities; please clarify which quantity is plotted in Fig. 16 and used in the text.","section":"Section 5.2"},{"comment":"The Milky Way panel uses -0.6<[Fe/H]<-0.4 while the simulated panels use -1.0<[Fe/H]<-0.8; the text compares the slopes without testing sensitivity to this metallicity choice. A brief check or caveat would be useful.","section":"Fig. 13"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a serious candidate for this journal. My main concern is the uncalibrated e<->Lz mapping between the observed and simulated Splash definitions; this is fixable with additional simulation analysis and should be a condition for acceptance. I also note that the simulation sample is small and deliberately selected for high accreted fractions, so the ubiquity claim needs tempering. The observational chemical analysis is careful and, on its own, close to publishable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on Kisku et al. It's really two papers in one. The observational half—a 16-element abundance comparison between the Splash and the high-alpha disc—is careful and convincing. They clean APOGEE sensibly, correct for log g systematics, use a solar-neighbourhood sample to dodge the selection function, and test both with binned chi-square and a bootstrap null. The finding that the Splash sits at the high-alpha, high-[Al,K/Fe], low-[Mn/Fe] end of the high-alpha disc, and that the chemical differences are a smooth function of eccentricity, is a real contribution. The abundance work stands on its own.\n\nThe simulation half is more interesting and more fragile. The new diagnostic is a strong correlation (Pearson r=0.92) between the simulated Splash fraction (in-situ retrograde stars) and the accreted retrograde fraction, across four MW-GES and three MW-MA ARTEMIS galaxies. That G44, a minor-merger galaxy, outdoes G34, a major prograde merger, does suggest orbital orientation matters more than total accreted mass. That reframes the Splash as a generic retrograde-accretion outcome, not a unique GE/S fingerprint.\n\nThe soft spot is the junction between the halves. The observed Splash is high-alpha stars with e>0.6; the simulated Splash is in-situ stars with Lz<0. The paper asserts Lz correlates with eccentricity, with a footnote saying it was verified, but gives no quantitative calibration. The observed sample includes prograde high-e stars; the simulated sample includes low-e retrograde stars. Those are not the same population, and this mapping underpins the confusion-map, age, and [Mg/Fe]-Lz comparisons. Until that equivalence is shown, the Milky-Way-specific conclusions stay provisional. Also, the r=0.92 partly reflects that both axes are retrograde fractions in the same spatial region; the 42 points are six azimuthal sectors per galaxy, so the effective independence is closer to 7 than 42. Minor issues: p-values quoted as 0.0 should be inequalities, and the e=0.6 and metallicity windows are post-hoc, though the authors are transparent about that.\n\nThis is not a rejection. The abundance work is publishable, and the simulation-side diagnostic is worth pursuing. But I would require a quantitative Lz-eccentricity calibration, ideally re-running Fig 16 with an e>0.6 cut on the simulation stars, and a check on within-galaxy clustering before publication. That is major revision, not a desk rejection.\n\nI'd send this to a serious referee without hesitation. Astronomers working on the Splash, disc heating, and merger diagnostics are the audience; I'd bring it to reading group to argue about definitions.","headline":"A solid abundance study with a provocative simulation-side claim that is currently undercut by an uncalibrated definitional bridge between observed and simulated Splash.","tokens_in":33731,"tokens_out":4873,"would_cite":true,"duration_ms":51984,"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":"The paper argues that Splash-like populations in Milky Way-mass galaxies are ubiquitous and correlate with retrograde accreted mass, so a single massive merger is not required to produce the Milky Way's Splash.","keywords":["Splash population","galactic archaeology","high-alpha disc","retrograde mergers","Gaia-Enceladus/Sausage","APOGEE abundances","ARTEMIS simulations","Milky Way formation"],"falsifier":"In the ARTEMIS galaxies, compute both eccentricity and angular momentum for every in-situ star particle at $z=0$; if stars with $e>0.6$ turn out to be frequently prograde, or if stars with $L_z<0$ frequently have $e<0.6$, then the observed selection and the simulated definition do not line up, and the $r=0.92$ correlation cannot be assumed to describe the Milky Way's Splash. A second, data-side check would be to measure the Splash fraction in APOGEE outside the solar neighbourhood ($R<5$ kpc) and see whether the simulated dependence of the Splash fraction on radius and height still holds there.","tokens_in":32585,"feed_emoji":"🌌","tokens_out":10069,"duration_ms":97636,"temperature":0.7,"pith_summary":"The paper sets out to test whether the Milky Way's Splash—a population of chemically old, kinematically hot disc stars on highly eccentric orbits—must be the unique signature of one massive merger, the Gaia-Enceladus/Sausage event. Using APOGEE and Gaia data for 14,258 high-$\\alpha$ stars (626 with eccentricity $e>0.6$) and 16 abundance ratios, it finds the Splash is chemically distinct from the rest of the high-$\\alpha$ disc but sits at the smooth, high-eccentricity end of trends in $[\\alpha/\\mathrm{Fe}]$, $[\\mathrm{Al}/\\mathrm{Fe}]$, $[\\mathrm{K}/\\mathrm{Fe}]$, and $[\\mathrm{Mn}/\\mathrm{Fe}]$, suggesting one continuous disc population. Turning to ARTEMIS cosmological simulations of Milky Way-mass galaxies, the paper then shows that Splash-like populations appear in galaxies with and without a Gaia-Enceladus/Sausage-scale merger. The Splash fraction correlates strongly (Pearson $r=0.92$) with the fraction of retrograde accreted stars, and a galaxy with only minor retrograde mergers can host a stronger Splash than a galaxy with a major prograde one. If correct, the result reframes the Splash as a general consequence of retrograde accretion history rather than a unique single-event fossil.","feed_headline":"No single giant merger needed to make the Splash","feed_subtitle":"A 0.92 correlation in simulations ties Splash stars to retrograde accreted mass, not merger size.","key_machinery":"The central object is the Splash itself: a population of old disc stars whose orbits were heated so strongly that they reach high eccentricity and often retrograde motion, defined in the observations as high-$\\alpha$ stars with eccentricity $e>0.6$ and in the simulations as in-situ star particles with angular momentum $L_z<0$ at redshift zero. The load-bearing comparison is the correlation of the simulated Splash fraction with the fraction of accreted stars on retrograde orbits, computed within matched solar-neighbourhood annuli ($5<R<11$ kpc, $|Z|<3$ kpc) for four galaxies with a major early merger and three with only minor early accretion. That correlation, rather than any single visual feature, is what lets the paper argue that retrograde orbital orientation is the controlling factor. The observed chemical analysis is carried by two complementary statistics: a per-metallicity-bin comparison of medians with $\\chi^2$ values and 1,000 bootstrap resamples, and a $\\chi^2$ distribution method that compares Splash abundance trends to 1,000 random disc samples of the same size.","core_discovery":"On the paper's own terms, the discovery is that the Splash is both real and not a unique formation event. In the Milky Way, Splash stars selected by eccentricity $e>0.6$ within the high-$\\alpha$ disc differ significantly from the rest of the high-$\\alpha$ disc in 12 of 16 abundance ratios: they are richer in $\\alpha$ elements, aluminium, and potassium and poorer in manganese, which the paper reads as an older, less chemically enriched population. Yet these differences are not a separate chemical track; they are the extreme end of smooth abundance gradients across eccentricity, meaning the Splash is the heated tail of the old disc. In the ARTEMIS simulations, the same kind of population—defined there as in-situ stars on retrograde orbits at $z=0$—is ubiquitous across Milky Way-mass hosts, whether or not they suffered a major early merger. The decisive variable is retrograde accreted mass: Splash fraction and accreted retrograde fraction correlate with Pearson $r=0.92$, and the minor-merger galaxy G44 out-performs two major-merger galaxies in Splash fraction. The paper concludes that lower-mass retrograde mergers can generate Splash-like populations, so the Milky Way's Splash does not by itself prove a single massive Gaia-Enceladus/Sausage collision.","pith_inferences":["Because the paper uses $L_z<0$ in simulations and $e>0.6$ in observations without direct calibration, a natural next step is to tag ARTEMIS star particles with both quantities simultaneously; if the mapping between the two definitions is loose, the $r=0.92$ correlation would need to be re-derived for the Milky Way's adopted eccentricity cut.","The paper's simulated [$\\mathrm{Mg}/\\mathrm{Fe}$]–$L_z$ plane shows a discontinuity at $L_z\\sim0$ that is absent in the Milky Way sample; one testable reading is that the Milky Way's Splash is more phase-mixed, or that the observed sample washes out the feature through selection, which could be checked by extending the APOGEE comparison to $R<5$ kpc.","If retrograde orientation is the controlling variable, then direct measurements of the orbital poles of surviving Milky Way satellites and streams, combined with their masses, could predict the Milky Way's Splash fraction before more stars are observed; the paper's correlation gives a quantitative target for that prediction.","The chemical pattern—high $[\\alpha/\\mathrm{Fe}]$, high $[\\mathrm{Al},\\mathrm{K}/\\mathrm{Fe}]$, low $[\\mathrm{Mn}/\\mathrm{Fe}]$—may serve as a diagnostic of retrograde-heated old disc in other galaxies for which only integrated light is available, though that would require a stellar-population synthesis step the paper does not carry out."],"forward_implications":["A Splash-like population in a galaxy is no longer reliable evidence by itself for a Gaia-Enceladus/Sausage-scale single merger.","Retrograde accreted mass, not total accreted mass, becomes the predictor to measure when estimating how strongly a galaxy's disc has been heated.","The Milky Way's Splash could contain contributions from several relatively low-mass retrograde mergers, not just one event, and still show the observed high-alpha, high-Al/K, low-Mn chemistry.","The calibrated Splash fraction as a function of radius and height can be used to map retrograde accretion histories across the disc and to predict where undiscovered Splash stars might be found.","Simulations with a major prograde merger but low retrograde accretion predict a weak Splash, implying some Milky Way analogues with a massive early merger would show almost no Splash."],"supporting_citations":[{"why":"Introduces the ARTEMIS simulation suite and its in-situ/accreted star particle labelling, providing the simulated galaxies and the definition of Splash used in Section 5.","marker":"Font et al. (2020)"},{"why":"Earlier ARTEMIS study that defines Splash-like populations as in-situ retrograde stars, identifies MW-GES and MW-MA categories and MMAP accretion times/mass ratios, and reports the accreted-fraction correlation this paper sharpens by restricting to retrograde orbits.","marker":"Dillamore et al. (2022)"},{"why":"Auriga simulations establishing the Splash fraction versus MMAP mass relation, spatial Splash maps, and the starburst prediction, serving as the comparison benchmark for the ARTEMIS-based results.","marker":"Grand et al. (2020)"},{"why":"Observational study that associates the Splash with the metal-rich thick disc and defines its chemo-kinematic locus; the paper's selection is validated against its $v_\\phi$–[Fe/H] plane.","marker":"Belokurov et al. (2020)"},{"why":"Identifies the Splash as a heated retrograde disc population connected to Gaia-Enceladus/Sausage, motivating the question of whether a single merger is required.","marker":"Belokurov et al. (2018)"},{"why":"Supplies the $\\chi^2$ distribution comparison method used in Section 3.2 to quantify differences between Splash and high-alpha disc abundance trends.","marker":"Taylor et al. (2022)"},{"why":"Provides the abundance comparison procedure and the $\\log g$ systematics discussion that the observational analysis follows, including chemical evolution context for Fe-peak elements.","marker":"Horta et al. (2023)"}],"fun_headline_variants":["Splash arises from retrograde mergers, not just giant ones","Splash is the disc's heated tail, not a merger fossil","Retrograde mass fraction predicts Splash strength (r=0.92)","Minor retrograde mergers can build the Splash","Splash: a heated old disc, not a unique collision scar"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"That the stars the paper identifies as Splash in the Milky Way—high-$\\alpha$ stars with eccentricity above 0.6—are the same physical population as the simulated stars it counts as Splash, namely stars born in the host disc that now move on retrograde orbits.","fun_headline_variants_meta":{"raw":{"variants":["Splash arises from retrograde mergers, not just giant ones","Splash is the disc's heated tail, not a merger fossil","Retrograde mass fraction predicts Splash strength (r=0.92)","Minor retrograde mergers can build the Splash","Splash: a heated old disc, not a unique collision scar"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000541,"raw_usage":{"total_tokens":2685,"prompt_tokens":1126,"completion_tokens":1559,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":742,"completion_tokens_details":{"reasoning_tokens":1484}},"tokens_in":742,"tokens_out":1559,"duration_ms":11644,"temperature":1.0,"reasoning_tokens":1484,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:21:16.625027+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In the ARTEMIS galaxies, compute both eccentricity and angular momentum for every in-situ star particle at $z=0$; if stars with $e>0.6$ turn out to be frequently prograde, or if stars with $L_z<0$ frequently have $e<0.6$, then the observed selection and the simulated definition do not line up, and the $r=0.92$ correlation cannot be assumed to describe the Milky Way's Splash. A second, data-side check would be to measure the Splash fraction in APOGEE outside the solar neighbourhood ($R<5$ kpc) and see whether the simulated dependence of the Splash fraction on radius and height still holds there.","supporting_citations":[],"review_version":1}