{"id":"15e06eac-4fd3-4de8-83fc-71c78de49b04","arxiv_id":"2507.00118","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Adding BOSS galaxy clustering with simulation-based priors modeled as Gaussian mixtures shifts the DESI plus CMB plus supernova constraints on dark energy toward a cosmological constant and improves the w0-wa figure of merit by about 20 percent.","lead":"The authors combine BOSS galaxy clustering data with a new Gaussian-mixture version of simulation-based priors, and find that the dynamical dark energy preference favored by DESI weakens once this information is added. The new prior method also makes the dark energy constraints about 20 percent tighter, which could influence how future DESI results are interpreted.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The SBP-induced shift of wa toward Lambda-CDM is the least secure step of the central claim: the GMM prior is trained on Ref [34] HOD mocks at a fiducial Lambda-CDM cosmology, and no non-Lambda-CDM mock validation is reported.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing point: the SBP is derived from Lambda-CDM HOD mocks, and its transferability to varying w0 and wa is not demonstrated. That is the most important concern because the headline 'preference disappears' has two layers. The first layer, the effect of adding BOSS full-shape data, is robust to the SBP: even with the conservative prior, Table I shifts wa to -0.342 +/- 0.220, consistent with earlier work [32]. The second layer, the claim that the SBP 'weakens the case for dynamical dark energy' by shrinking the posterior around Lambda-CDM and improving the FoM by about 20%, is uniquely attributable to the simulation-based prior. If that prior is Lambda-CDM-dependent, the extra shrinkage is not independent evidence.\n\nThe paper's internal checks do not resolve this. The convergence from GMM6 to GMM10 and the Appendix A comparison against normalizing flows validate the GMM as an accurate approximation to the HOD sample distribution at the fixed cosmology of the mocks. They do not validate that the distribution of EFT parameters is unchanged when w0 and wa vary across the DESI-preferred region. Thus the concern is empirical rather than formal, and it can be addressed by mock-based validation.\n\nI am not recommending a change to the reader's verdict, because the CONDITIONAL verdict already captures this uncertainty. The analysis is careful, the analytic marginalization derivation is standard, and the BOSS full-shape result has independent support. But the paper should either demonstrate cosmology-independence of the SBP or restrict its conclusion to the BOSS-only effect when discussing the dynamical dark energy preference. A formal model-comparison statistic would also strengthen the wording 'preference disappears,' but I do not rest the objection on that point since credible-interval statements are common in this literature.","tokens_in":20026,"tokens_out":6784,"duration_ms":83402,"concrete_test":"Generate HOD mocks with the same HOD prescription, abundances, and redshifts as Ref. [34] but in at least two non-Lambda-CDM cosmologies bracketing the DESI-preferred region (for example, w0 = -0.9, wa = 0.3 and w0 = -0.95, wa = -0.3). Measure the 11 EFT parameters at field level, fit a GMM with the same 10-component protocol, and re-run the Planck + BOSS + Pantheon+ + DESI analysis. If the alternative GMM shifts wa by more than about 0.1 or changes the FoM by more than about 10% relative to Table I, then the SBP is cosmology-dependent and the SBP-specific weakening of dynamical dark energy is not established; if the shifts are negligible, the concern is resolved and the claimed shrinkage is legitimate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central physical conclusion has two parts: (i) adding BOSS EFT full-shape to DESI+CMB+SNe removes the dynamical-dark-energy preference, and (ii) the simulation-based prior further shrinks the w0-wa posterior around Lambda-CDM and improves the FoM by about 20%. Part (i) is supported by the conservative-prior row in Table I and by earlier work, so it does not depend on the SBP. Part (ii) is the paper's distinctive claim, and it rests entirely on the SBP.\n\nThe SBP is imported from Ref. [34]: Section 2.1 states that EFT parameters \"have been measured from large catalogs of the HOD galaxies in [34]\" and that these measurements \"constitute the simulations-based priors.\" Those HOD catalogs are generated at a fiducial Lambda-CDM cosmology, and the paper gives no evidence to the contrary. The GMM in Section 2.3 is an approximation to that fixed-cosmology distribution. Bias and stochastic parameters in the EFT are not guaranteed to be cosmology-independent over the w0-wa range explored; the paper reports no test on non-Lambda-CDM mocks. Table I shows the GMM10 prior moves wa from -0.342 +/- 0.220 (conservative) to -0.094 +/- 0.188, and Table II attributes the roughly 20% FoM gain to the SBP. If the prior encodes Lambda-CDM, the shrinkage is partly prior-driven rather than data-driven. This is a calibration and transferability gap, not an internal inconsistency, and it can be settled empirically.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a Gaussian-mixture-model (GMM) approximation to simulation-based priors (SBPs) on effective-field-theory (EFT) nuisance parameters in galaxy-clustering full-shape analyses, preserving the ability to marginalize analytically over the linearly entering parameters. It applies this method to a joint analysis of BOSS DR12 power spectrum, bispectrum, and BAO data combined with Planck 2018 CMB, Pantheon+ supernovae, and DESI DR2 BAO, and reports two main results: (i) adding BOSS full-shape data removes the dynamical-dark-energy preference seen in DESI+CMB+SNe, and (ii) the simulation-based prior further shrinks the w0-wa posterior around the cosmological constant and improves the dark-energy figure of merit by roughly 20%.","tokens_in":20397,"tokens_out":6045,"duration_ms":71742,"significance":"If both claims hold, the paper is a valuable methodological and phenomenological contribution: it makes SBP analyses analytically tractable via GMMs and shows that high-precision BOSS full-shape data disfavor the dynamical-dark-energy interpretation of DESI DR2. The use of public BOSS likelihoods, the convergence checks across GMM3/GMM6/GMM10, and the normalizing-flow comparison in Appendix A are concrete strengths. The central quantitative claim of a ~20% figure-of-merit gain is, however, conditional on the SBP being transferable from Lambda-CDM HOD mocks to the general w0-wa cosmologies explored in the chains, and the paper does not demonstrate that transferability.","major_comments":[{"comment":"The simulation-based prior is taken from HOD mock catalogs of Ref. [34] generated at a fiducial Lambda-CDM cosmology, and the GMM trained on those samples is then applied to chains that vary w0 and wa. No test of the cosmology dependence of the EFT parameter distribution is provided. Because Table I and Table II attribute the shift of wa toward 0 and the ~20% FoM improvement to the SBP, the distinct claim that the SBP 'further weakens the case for dynamical dark energy' (abstract and Section 5) could be partly prior-driven rather than data-driven. This is a transferability/calibration gap, not an internal inconsistency; it should be closed by recalibrating the prior from mocks at several non-Lambda-CDM cosmologies or by demonstrating that the EFT prior is approximately cosmology-independent over the explored range. Without such a test, the SBP-based conclusions should be softened.","section":"2.1, 4, 5"},{"comment":"The abstract states that the dynamical-dark-energy preference 'disappears' once BOSS full-shape is added, but the support is only the 95% credible region of the w0-wa posterior; no Bayes factor, evidence ratio, or Delta-chi-squared is reported. The baseline DESI+CMB+SNe hint is itself only at the ~2.7-sigma level in this paper, so a quantitative model-comparison statistic is needed to justify the 'disappears' language and to compare the conservative-prior and SBP rows of Table I. Similarly, the 20% FoM gain in Table II is given without an uncertainty or a sensitivity estimate.","section":"4, Fig. 3, Table I"}],"minor_comments":[{"comment":"There is a typo in Section 3: 'In each two regions, the we have mixed samples' should be corrected.","section":"3"},{"comment":"The text says the Gelman-Rubin statistics satisfies R < 0.01; since R is always at least 1, this should read R-1 < 0.01.","section":"4"},{"comment":"GMM3 has a higher FoM (268.21) than GMM10 (248.81), yet the text claims convergence between GMM6 and GMM10 and dismisses GMM3 as a numerical artifact; a brief quantitative explanation of why GMM3 is not the preferred model would strengthen the convergence argument.","section":"Table II, Section 4"},{"comment":"Some panels in the triangle plot lack readable axis labels, and the legend uses 'Normalizing Flow' while Fig. 4 uses 'SBP Normalizing Flow'; the figures should be made consistent and self-explanatory.","section":"Fig. 1"},{"comment":"The selection of DESI DR2 BAO samples with z > 0.75 and the exact overlap-removal procedure with BOSS are described only briefly; listing the specific samples (LRG, ELG, QSO) and their redshift cuts would improve reproducibility.","section":"3"}],"recommendation":"major_revision","confidential_remarks":"The SBP transferability issue is the key risk to the paper's headline claim. If the authors can add a non-Lambda-CDM mock test, the paper would be substantially stronger; otherwise the abstract and Section 5 should be reworded to distinguish the BOSS-full-shape effect (which appears robust) from the SBP-specific shrinking effect (which is not yet demonstrated to be cosmology-independent). The GMM methodology itself is a useful technical contribution and the paper fits the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the GMM representation of EFT-SBP priors with analytic marginalization is a genuine technical advance—it makes simulation-based priors practical for expensive extended-model runs, and the convergence checks across GMM3/6/10 plus the normalizing-flow comparison in Appendix A give me reasonable confidence in the approximation. Second, the paper's headline that BOSS full-shape erases the DESI w0-wa preference survives contact with the data, but the extra claim that the simulation-based prior further shrinks the posterior toward LCDM rests on priors trained on LCDM mocks, and that part is not on the same footing.\n\nWhat the paper does well: it uses public, established likelihoods, includes a conservative-prior baseline, and the reduction in preference from adding BOSS FS is visible in the conservative row of Table I, independent of the SBP. That part is consistent with earlier work [32] and is probably robust. The GMM machinery is well explained and the analytic marginalization is clean. The FoM gain of ~20% from GMM10 over the conservative prior is plausible, though GMM3 scoring higher than GMM10 makes me want a bit more discussion of stability.\n\nWhere it's soft. The 'disappears' language outruns the statistics: there is no Bayes factor or delta chi-squared, just posterior contours that shrink and shift. I would not call that a fatal flaw—the contours do move—but it is an overstatement. More substantively, the SBP transferability gap is real. The priors come from HOD mocks at a fiducial LCDM cosmology, and the paper does not test them on non-LCDM mocks. Bias and stochastic parameters can shift with the growth history, so the SBP-driven movement of wa toward zero is partly a statement about the prior, not purely the data. That said, the claim that the DESI preference weakens with BOSS FS does not depend on the SBP, so the main conclusion stands. The missing code and GMM components are a minor reproducibility friction.\n\nWho should read it: anyone working on full-shape galaxy clustering or the DESI dynamical-dark-energy question. It deserves a serious referee; the SBP transferability critique is addressable with additional mock tests, and the paper would be stronger if the authors added a model-comparison statistic and released the GMM components.","headline":"A real technical advance in EFT-SBP, with a solid but over-claimed physical conclusion; the SBP-induced shrinkage toward LCDM is the weak link.","tokens_in":20930,"tokens_out":2532,"would_cite":true,"duration_ms":30816,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["98.80.-k","95.36.+x"],"model":"deepseek-v4-flash","headline":"Adding BOSS galaxy clustering data to DESI, Planck, and Pantheon+ removes the reported preference for dynamical dark energy, and simulation-based priors tighten the dark energy constraint by about 20 percent.","keywords":["dynamical dark energy","w0-wa parametrization","galaxy clustering full-shape analysis","effective field theory of large-scale structure","simulation-based priors","Gaussian mixture model","BOSS survey","DESI BAO"],"falsifier":"Recalibrate the simulation-based prior from HOD mocks generated in a cosmology with $w_0$ and $w_a$ fixed away from $-1$ and $0$ (for example at the DESI best-fit values), then rerun the same joint likelihood; if the $w_0$-$w_a$ posterior moves toward those injected values, the prior is not cosmology-independent and the reported shift to $\\Lambda$CDM is not purely data-driven.","tokens_in":19783,"feed_emoji":"🌌","tokens_out":7758,"duration_ms":77529,"temperature":0.7,"pith_summary":"DESI's combination of baryon acoustic oscillations, Planck CMB data, and Pantheon+ supernovae has been reported to prefer dark energy that changes with time. This paper tests that preference by adding BOSS galaxy clustering data modeled with effective-field-theory perturbation theory, and it claims that the preference disappears once BOSS full-shape power spectrum and bispectrum information is included. The paper goes further: nuisance parameters describing galaxy bias and small-scale noise are constrained by priors measured from mock galaxies, and this additional information shrinks the $w_0$-$w_a$ posterior around the cosmological-constant point, making $w_0=-1$ and $w_a=0$ consistent at 95% confidence. The technical enabler is a Gaussian mixture model approximation of the simulation-based prior, which keeps the likelihood analytically marginalizable and makes the combined analysis computationally practical. If right, the paper undercuts the dynamical dark energy interpretation of DESI data and gives future full-shape analyses a fast way to use simulation-based priors.","feed_headline":"Adding BOSS clustering data erases DESI dark energy hint","feed_subtitle":"Simulation-based priors pull w0 and wa back to the cosmological constant within 95 percent.","key_machinery":"The load-bearing object is a Gaussian mixture model fit to the simulation-based prior distribution of EFT nuisance parameters. In the one-loop EFT model, eight of the eleven nuisance parameters—the cubic bias $b_{\\Gamma_3}$, the redshift-space counterterms $c_{s,0}$, $c_{s,2}$, $c_{s,4}$, the higher-derivative term $b_4$, and the stochastic parameters $P_{\\mathrm{shot}}$, $a_0$, $a_2$—enter the prediction linearly, so the likelihood is quadratic in them. For a single Gaussian prior, those parameters can be marginalized analytically; the paper extends the same analytic marginalization to a weighted sum of $K$ Gaussian components ($K = 3, 6, 10$), so the simulation-based prior, which is non-Gaussian and skewed, is captured without explicitly sampling all nuisance parameters. This is what makes the otherwise expensive joint run with Planck, Pantheon+, and DESI BAO feasible, and it is what converts the prior samples from HOD mocks into a posterior that is tighter around $\\Lambda$CDM.","core_discovery":"On the paper's own terms, the central discovery is that the reported hint for dynamical dark energy is not robust: when the DESI DR2 BAO, Planck, and Pantheon+ data are analyzed together with the BOSS EFT full-shape galaxy power spectrum and bispectrum likelihood, the $w_0$-$w_a$ posterior returns to the cosmological constant region. With the simulation-based priors modeled as a Gaussian mixture, the posterior contracts further: $w_0$ moves toward $-1$, $w_a$ moves toward $0$, the $\\Lambda$CDM point lies within the 95% credible region, and the figure of merit, defined as $\\mathrm{FoM} = 1/\\sqrt{\\det\\mathrm{Cov}(w_0,w_a)}$, improves by approximately 20% relative to the conservative EFT prior. The authors interpret this as evidence that the field-level simulation-based priors carry useful small-scale information that reduces degeneracies between cosmological parameters and galaxy nuisance parameters.","pith_inferences":["A testable implication the paper leaves implicit: the shift toward $\\Lambda$CDM could partly inherit the $\\Lambda$CDM cosmology used to generate the HOD mocks, so a cross-check is to recalibrate the simulation-based priors from simulations run in a $w_0w_a$ cosmology.","If the same field-level priors are applied to DESI's own full-shape data, which the paper explicitly leaves for future work, the figure-of-merit gain may be larger than 20% because DESI has greater volume than BOSS; this is a prediction of the method rather than a result reported here.","The GMM convergence check, in which the figure of merit changes by less than 4% between 6 and 10 components, tests stability within the Gaussian mixture family but not fidelity to the true simulation prior, so independent normalizing-flow-based sampling remains a useful validation as data errors shrink.","A broader consequence: because simulation-based priors shrink nuisance parameter posteriors without strongly reweighting cosmological parameters, they could also sharpen constraints on other cosmological parameters, such as neutrino mass or curvature, in the same combined-data setup."],"forward_implications":["If the claim holds, the DESI BAO+CMB+SNe preference for $w_0$ and $w_a$ away from $\\Lambda$CDM reflects a dataset combination issue rather than a robust detection of dynamical dark energy.","The BOSS full-shape likelihood with field-level simulation-based priors becomes a standard tool for non-minimal cosmological models, since analytic marginalization removes the main computational bottleneck.","The reported approximately 20% improvement in the $w_0$-$w_a$ figure of merit should be interpreted as a conservative gain; the comparison with a single Gaussian prior shows that the usual shortcut would miss part of this information.","The $w_0$-$w_a$ constraints are consistent with the cosmological constant within 95% confidence, so an extended dark energy model is currently not required by the combined data sets."],"supporting_citations":[{"why":"Provides the field-level EFT parameter samples from HOD mocks that define the simulation-based prior approximated by the Gaussian mixture model.","marker":"[34]"},{"why":"Introduced the EFT-SBP approach with normalizing flows and the field-level calibration method that the GMM replaces.","marker":"[33]"},{"why":"Supplies the BOSS EFT full-shape power spectrum and bispectrum likelihood and the conservative Gaussian prior used as the baseline.","marker":"[30]"},{"why":"Performed a similar DESI DR1 BAO plus BOSS full-shape dynamical dark energy analysis that this work extends to DESI DR2 with simulation-based priors.","marker":"[32]"},{"why":"Provides the DESI DR2 BAO measurements that, together with Planck and Pantheon+, create the dynamical dark energy preference being tested.","marker":"[3]"},{"why":"Supplies the Planck 2018 CMB likelihood used in the joint analysis.","marker":"[63]"},{"why":"Provides the Pantheon+ supernova dataset used alongside DESI BAO and Planck in the baseline dark energy constraints.","marker":"[64]"}],"fun_headline_variants":["Simulation priors pull dark energy hint back to cosmological constant","Gaussian-mixture priors make DESI dark energy hint vanish","BOSS and simulation-based priors erase DESI dark energy preference","Simulation priors shrink dark energy parameter space, favor static constant"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The simulation-based priors, derived from haloes and galaxies simulated in a $\\Lambda$CDM universe and tuned to look like BOSS, correctly describe galaxy bias and small-scale noise for the real Universe regardless of the true dark energy model.","fun_headline_variants_meta":{"raw":{"variants":["Simulation priors pull dark energy hint back to cosmological constant","Gaussian-mixture priors make DESI dark energy hint vanish","BOSS and simulation-based priors erase DESI dark energy preference","Simulation priors shrink dark energy parameter space, favor static constant"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001384,"raw_usage":{"total_tokens":5645,"prompt_tokens":1031,"completion_tokens":4614,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":4541}},"tokens_in":647,"tokens_out":4614,"duration_ms":37222,"temperature":1.0,"reasoning_tokens":4541,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:24:07.570129+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recalibrate the simulation-based prior from HOD mocks generated in a cosmology with $w_0$ and $w_a$ fixed away from $-1$ and $0$ (for example at the DESI best-fit values), then rerun the same joint likelihood; if the $w_0$-$w_a$ posterior moves toward those injected values, the prior is not cosmology-independent and the reported shift to $\\Lambda$CDM is not purely data-driven.","supporting_citations":[{"cited_title":"Assembly bias in quadratic bias parameters of dark matter halos from forward modeling","cited_arxiv_id":"2106.14713","evidence_quote":"Supplies the Planck 2018 CMB likelihood used in the joint analysis."}],"review_version":1}