{"id":"b9508952-08cd-4ef4-843d-84d90da8dc55","arxiv_id":"2507.05468","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"Pantheon+ supernova distance moduli and model residuals are found to be non-Gaussian, better fit by a skew-t distribution, and bootstrap confidence intervals are provided for an interacting dark-energy model.","lead":"This paper tests whether supernova brightness measurements follow a normal distribution and finds they do not. It then fits an interacting dark energy model using statistical methods that do not assume normality.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Paper's central statistical finding (non-Gaussian residuals, skew-t fit) is credible as an internal analysis, but the abstract overclaims a redshift t-distribution and the bootstrap ignores the Pantheon+ covariance matrix, so the practical impact on dark-energy parameter estimation is not…","rationale":"The reader's weakest_assumption—that the tests and bootstrap ignore the published Pantheon+ covariance matrix—is exactly the load-bearing concern. I agree with the reader's assessment. The strongest claim is the rejection of Gaussianity and the skew-t fit for the residuals, which is plausible and internally consistent by standard tests, but the practical impact on dark-energy parameter estimation is not shown because the analysis treats the Pantheon+ data as independent and does not compare against a covariance-aware standard analysis. The abstract's statement about 'redshift distribution' being a t-distribution overstates what the paper actually tested (residuals, not redshifts). These issues mean the paper should be accepted only conditionally, after the authors redo the analysis with the covariance matrix and clarify what is actually being claimed about the redshift distribution. No fraud or misconduct is implied; the concern is purely technical and testable.","tokens_in":7348,"tokens_out":1563,"duration_ms":15747,"concrete_test":"Redo the residual test and the bootstrap using the full Pantheon+ covariance matrix: first, compute the standardized residuals using C^{-1} and the best-fit model, then re-run the Lilliefors and Jarque-Bera tests on those residuals. Second, replace the independent bootstrap with a covariance-aware procedure (e.g., resample the fitted residuals and add a multivariate normal draw with covariance C, or use the published Pantheon+ covariance in a chi-square fit). If the skew-t fit is still not rejected and the parameter intervals broaden by more than, say, 20%, the practical claim is weakened; if the intervals are robust, the paper's practical conclusion survives.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim—that Gaussianity is untenable and that a skew-t distribution with six degrees of freedom fits the residuals—is internally supported by the Lilliefors (D=0.039, p=0.0007), Jarque-Bera (JB=26.19) and Kolmogorov-Smirnov (p=0.9099) tests. However, the load-bearing weakness is in the leap from residuals to cosmological parameter estimation. The residual distribution is not the redshift distribution; the abstract says 'the redshift distribution is more accurately described by a t-distribution', but the KS test in Section 3.1 is applied only to the model residuals, not to the redshift variable itself. Moreover, the bootstrap confidence intervals in Section 3.1 are built by resampling the 1,358 Pantheon+ distance moduli independently and re-fitting via ordinary least squares, without using the published Pantheon+ covariance matrix. Since Pantheon+ distance moduli are known to have significant correlated systematic uncertainties (e.g., bias corrections, photometric calibration, peculiar velocities), treating them as exchangeable draws can produce intervals that are too narrow and point estimates that are subtly biased. The 95% bootstrap intervals reported (e.g., Omega in [0.323,0.343]) are extremely tight compared with typical Pantheon+ analyses, which suggests that the systematic covariance has not been propagated. If that covariance dominates, the reported p-values and confidence intervals are not a reliable basis for inferring dark-energy parameters or for claiming that the skew-t residual model changes parameter constraints. The central statistical finding about residuals is plausible, but the paper's advertised consequence—that this changes parameter estimation for dark energy models—is not demonstrated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper tests whether the Gaussianity assumption is valid for the Pantheon+ supernova data and for residuals of an interacting dark-energy model. Using the Lilliefors and Jarque-Bera tests on residuals from a restricted-gradient fit, the authors report strong rejection of normality (D = 0.039, p = 0.0007 and JB = 26.19, p = 10^-16). They then fit a skewed generalized t (SGT) distribution with q = 6 degrees of freedom, obtaining a Kolmogorov-Smirnov p-value of 0.9099, and construct bootstrap 95% confidence intervals for the cosmological parameters Omega_m, H0, gamma_m, and gamma_x. The paper concludes that non-Gaussianity is untenable for the standard assumption and that a skew-t distribution describes the residuals better.","tokens_in":7740,"tokens_out":4538,"duration_ms":55654,"significance":"If the residual non-Gaussianity finding is robust, it is a useful caution for the many cosmological analyses that assume Gaussian distance-modulus errors. The paper reports explicit test statistics and p-values, and the bootstrap procedure is clearly described. The main limitation is that the analysis ignores the Pantheon+ covariance matrix and applies the normality tests to residuals from a single fit, so the reported p-values and confidence intervals are conditional on an independence assumption that is likely violated for this dataset. In addition, the abstract's claim about the redshift distribution is not supported by the tests performed, and the paper does not demonstrate that the non-Gaussianity actually changes parameter estimates. These issues are load-bearing for the paper's main message and require substantial revision.","major_comments":[{"comment":"The abstract states that 'the redshift distribution is more accurately described by a t-distribution,' but no normality test is applied to the redshift variable. The Lilliefors, Jarque-Bera, and Kolmogorov-Smirnov tests in Section 3.1 are applied to the residuals of the restricted-gradient fit, not to the redshift distribution. The manuscript should either test the redshift variable directly or revise the abstract to refer to model residuals rather than the redshift distribution.","section":"Abstract and Section 3.1"},{"comment":"The bootstrap resamples the 1,358 distance moduli independently and does not use the Pantheon+ covariance matrix. Because Pantheon+ distance moduli have significant correlated systematic uncertainties (for example, calibration, bias corrections, and peculiar velocities), treating them as exchangeable draws can produce confidence intervals that are too narrow and p-values that are not exact. The 95% intervals in Table 5 should be recomputed using the published covariance matrix, for instance by sampling from the multivariate normal with the full covariance or by a residual bootstrap that accounts for the correlation structure, and the effect on the intervals should be reported.","section":"Section 3.1, bootstrap paragraph"},{"comment":"The skew-t distribution is fitted to the same residuals that are subsequently used for the Kolmogorov-Smirnov test. The reported p = 0.9099 is therefore a goodness-of-fit diagnostic rather than an independent confirmation of the model. Additionally, Equation (7) is the skewed generalized t distribution, not the ordinary skew-t distribution; the terminology should be corrected, and the fit should be compared with a Gaussian model using information criteria or cross-validation to justify the choice.","section":"Section 3.1, skew-t fit"},{"comment":"The paper promises an analysis of the impact of non-Gaussianity on parameter estimation, but no comparison is made between parameter estimates obtained under Gaussian and non-Gaussian assumptions. The bootstrap confidence intervals in Table 5 are based on least-squares estimation, which is equivalent to Gaussian maximum likelihood; the fitted skew-t distribution is not used in the estimation. To support the claim that non-Gaussianity affects parameter estimation, the authors should present a likelihood or weighted fit using the SGT error model and compare the resulting parameters and confidence intervals with the Gaussian-based results.","section":"Title, Abstract, and Section 3.1"}],"minor_comments":[{"comment":"There is a typo in the abstract: 'comprehensive statistical, analysis' should read 'comprehensive statistical analysis.'","section":"Abstract"},{"comment":"The prime notation in Equation (5) is used without defining the derivative variable; the authors should specify whether the prime denotes differentiation with respect to ln a or another variable.","section":"Section 2, Equation (5)"},{"comment":"The statement that a standard deviation of 0.14 'provides evidence' of a platykurtic distribution is not justified; kurtosis should be estimated directly rather than inferred from the standard deviation.","section":"Section 3.1, Table 3"},{"comment":"The goodness-of-fit column mixes minimum sum of squares for the Grid and Gradient methods with a maximum for the EM method, making the values non-comparable across methods; this should be clarified or presented in separate columns.","section":"Table 2"},{"comment":"The text refers to the 'original Pantheon dataset, which contains 1,358 observations' while the paper otherwise uses the Pantheon+ dataset; the exact sample and the number of distance moduli used should be clarified.","section":"Section 3.1, bootstrap paragraph"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a statistically important question, and the reported non-Gaussianity tests are internally consistent. However, the omission of the Pantheon+ covariance matrix is a serious technical gap that affects the confidence intervals and p-values, and the abstract overstates the redshift-distribution result. These issues are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the residual non-Gaussianity result is real and cleanly demonstrated, but the paper overreaches when it claims this changes dark-energy parameter estimation. The Lilliefors and Jarque-Bera tests are decisive (D=0.039, p=0.0007; JB=26.19), and the skew-t fit with q=6 and KS p=0.91 is a reasonable description of those residuals. That part is solid and worth citing as a confirmation of earlier work by Dainotti et al. and others.\n\nThe soft spots are where the paper tries to extend the result. First, the abstract says the redshift distribution is better described by a t-distribution, but the tests are on model residuals, not on the redshift variable itself. That is a plain misstatement and should be fixed. Second, the bootstrap confidence intervals resample the 1,358 distance moduli independently and ignore the published Pantheon+ covariance matrix. Given how tight the reported intervals are (Omega in [0.323,0.343]), I suspect the correlated systematics are not being propagated. Without the covariance matrix, the intervals are not trustworthy as a statement about dark-energy parameters. Third, the skew-t fit is validated on the same residuals it was fitted to, so the KS test is a fit diagnostic rather than an independent confirmation—minor, but worth stating. Finally, the paper uses only one fit to generate residuals, so we don't know if the non-Gaussianity is robust to the choice of model or fitting method.\n\nThe model fitting itself is fine as an exercise, and the three methods (grid, EM, gradient) give consistent estimates. The interaction model is a legitimate extension of previous work. But the paper's advertised consequence—that the Gaussian assumption materially changes parameter constraints—is not demonstrated here. The reader's report and the stress-test note both land this correctly.\n\nRecommendation: send to a competent referee with instructions that the covariance issue is load-bearing. The paper deserves review time because the residual test is clean and the topic matters. But I wouldn't accept the current version as is, and I wouldn't cite it for the parameter intervals.","headline":"The paper convincingly rejects Gaussianity for Pantheon+ residuals but fails to show the practical impact on dark-energy parameters because it ignores the covariance matrix and overclaims a redshift t-distribution.","tokens_in":8251,"tokens_out":2267,"would_cite":false,"duration_ms":24555,"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":"Normality of supernova residuals is rejected; a six-degree skew-t distribution fits the Pantheon+ residuals while bootstrap confidence intervals are built for an interacting dark-energy model.","keywords":["Gaussianity assumption","Pantheon+","Type Ia supernovae","interacting dark energy","skew-t distribution","normality tests","bootstrap confidence intervals","residual analysis"],"falsifier":"Repeat the Lilliefors and Jarque-Bera tests on residuals weighted by the full Pantheon+ covariance matrix, and bootstrap from the corresponding multivariate distribution; if normality is not rejected or the skew-t fit then fails the Kolmogorov-Smirnov test, the paper's central conclusion collapses.","tokens_in":7172,"feed_emoji":"🔭","tokens_out":6705,"duration_ms":74497,"temperature":0.7,"pith_summary":"Type Ia supernova cosmology has long taken distance-modulus residuals to be Gaussian, and this paper argues that assumption breaks down on the Pantheon+ catalogue. Fitting a sign-changeable interacting dark-energy model, the authors test the residuals with the Lilliefors and Jarque-Bera tests and obtain $p = 0.0007$ and $p = 10^{-16}$, rejecting normality. They then show that a skew-t distribution with six degrees of freedom fits the residuals well, with a Kolmogorov-Smirnov $p$-value of $0.9099$, and they build bootstrap confidence intervals for the model parameters. A sympathetic reader should care because every uncertainty quoted from Gaussian likelihood fits in this supernova analysis is suspect if the error distribution really is heavy-tailed.","feed_headline":"Six-degree skew-t beats Gaussian for Pantheon+ residuals","feed_subtitle":"Normality tests reject Gaussian residuals (p=0.0007; p<10^-16); bootstrap intervals follow for an interacting dark-energy model.","key_machinery":"The load-bearing statistical object is the residual distribution, tested by the Lilliefors and Jarque-Bera normality tests and compared with the skewed generalized $t$ density, a four-parameter family with location, scale, skewness, and tail-weight parameters. On the cosmological side, the model is a sign-changeable interaction $Q = q(\\alpha \\rho_x' + \\beta \\rho_x)$, where the deceleration parameter $q$ controls the sign change and the energy densities follow from a second-order differential equation. Bootstrap percentiles provide confidence intervals without invoking a Gaussian likelihood.","core_discovery":"The central discovery is that the Pantheon+ data and the residuals of the paper's interacting dark-sector model are not Gaussian, and the residuals are better described by a slightly skewed, heavy-tailed $t$-distribution. Descriptive statistics show asymmetry and a residual standard deviation of $0.1425$; the Lilliefors test gives $D = 0.039194$ with $p = 0.0007$, and the Jarque-Bera test gives $JB = 26.19$ with $p = 10^{-16}$, both rejecting normality. Fitting the skewed generalized $t$ density produces location $\\mu = -0.03$, scale $\\sigma = 0.12$, skewness $\\lambda = 0.04$, and $q = 6$ degrees of freedom, and the Kolmogorov-Smirnov test does not reject this fit ($p = 0.9099$). The paper then uses bootstrap resampling of the 1,358 distance moduli to construct 95% percentile confidence intervals for the matter density parameter, the Hubble parameter, and the dark-sector equation-of-state parameters.","pith_inferences":["If the residual distribution is truly heavy-tailed, published Pantheon+ parameter uncertainties computed under Gaussian likelihoods are likely too narrow; re-deriving them from a skew-t likelihood is a direct test of the practical impact.","The present tests fit residuals without the Pantheon+ covariance matrix, so a natural robustness check is to repeat the Lilliefors, Jarque-Bera, and skew-t fits using covariance-weighted residuals; the result could go either way.","The paper hints that the redshift distribution may be a mixture of two normals or two $t$ distributions; fitting such a mixture to the full Pantheon+ sample would be a testable extension.","The same battery of distribution tests could be applied to other cosmological probes, such as baryon acoustic oscillations or cosmic microwave background likelihoods, where systematic correlations may preserve or destroy Gaussianity."],"forward_implications":["Minimum chi-squared estimation, which assumes Gaussian errors, is not an adequate inferential basis for the interacting dark-energy model fit here.","A skew-t likelihood with about six degrees of freedom should replace or at least compete with the Gaussian likelihood for these supernova residuals.","Bootstrap percentile intervals, being distribution-free, give a defensible way to quote cosmological parameter uncertainties even when residuals are non-Gaussian.","The interacting dark-sector model remains viable under the non-Gaussian treatment: late-time acceleration occurs with $q_0$ near $-0.7$, and the interaction changes sign depending on the sign of $\\beta$."],"supporting_citations":[{"why":"Supplies the Pantheon+ catalogue of 1,358 supernova distance moduli used for all fits and tests.","marker":"[4]"},{"why":"Argues that the Gaussianity assumption for supernova distance moduli can affect cosmological parameter determination, motivating this study.","marker":"[6]"},{"why":"Introduces the sign-changeable interaction that the paper extends with a distribution-free analysis.","marker":"[11]"},{"why":"Provides the interaction form and differential equation used to build the analytical energy-density solution.","marker":"[12]"},{"why":"Supplies the solution method for the second-order energy-density equation used to derive the model.","marker":"[13]"},{"why":"Supplies the restricted gradient optimization method used for parameter estimation.","marker":"[14]"},{"why":"Provides the Lilliefors normality test used to reject Gaussian residuals.","marker":"[19]"},{"why":"Provides the Jarque-Bera normality test used to reject Gaussian residuals.","marker":"[20]"},{"why":"Supplies the skewed generalized t density used to model the residuals.","marker":"[21]"}],"fun_headline_variants":["Gaussian fails for Pantheon+; skew-t fits residuals","Normality rejected for Pantheon+ dark-energy data","Skew-t beats Gaussian for supernova residuals","Pantheon+ residuals non-Gaussian; skew-t wins","Bootstrap intervals for dark energy after normality rejection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on treating the Pantheon+ distance moduli and the residuals of a single nonlinear least-squares fit as effectively independent draws, without using the survey's published covariance matrix; if the correlated systematic uncertainties are substantial, the reported $p$-values and bootstrap confidence intervals may not be valid.","fun_headline_variants_meta":{"raw":{"variants":["Gaussian fails for Pantheon+; skew-t fits residuals","Normality rejected for Pantheon+ dark-energy data","Skew-t beats Gaussian for supernova residuals","Pantheon+ residuals non-Gaussian; skew-t wins","Bootstrap intervals for dark energy after normality rejection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000217,"raw_usage":{"total_tokens":1437,"prompt_tokens":945,"completion_tokens":492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":561,"completion_tokens_details":{"reasoning_tokens":415}},"tokens_in":561,"tokens_out":492,"duration_ms":5171,"temperature":1.0,"reasoning_tokens":415,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:27:16.838247+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the Lilliefors and Jarque-Bera tests on residuals weighted by the full Pantheon+ covariance matrix, and bootstrap from the corresponding multivariate distribution; if normality is not rejected or the skew-t fit then fails the Kolmogorov-Smirnov test, the paper's central conclusion collapses.","supporting_citations":[{"cited_title":"On the statistical assumption on the distance moduli of Supernovae Ia and its impact on the determination of cosmological parameters","cited_arxiv_id":null,"evidence_quote":"Argues that the Gaussianity assumption for supernova distance moduli can affect cosmological parameter determination, motivating this study."},{"cited_title":"Dynamics and statefinder analysis of a class of sign-changeable interacting dark energy scenarios","cited_arxiv_id":"2202.05130","evidence_quote":"Introduces the sign-changeable interaction that the paper extends with a distribution-free analysis."},{"cited_title":"Linear and nonlinear interactions in the dark sector","cited_arxiv_id":"0911.5687","evidence_quote":"Supplies the solution method for the second-order energy-density equation used to derive the model."},{"cited_title":"Nesterov, Introductory Lectures on Convex Optimization: A Basic Course , Springer, 2004","cited_arxiv_id":null,"evidence_quote":"Supplies the restricted gradient optimization method used for parameter estimation."},{"cited_title":"Lilliefors, On the Kolmogorov-Smirnov test for normality with mean and variance unknown, J","cited_arxiv_id":null,"evidence_quote":"Provides the Lilliefors normality test used to reject Gaussian residuals."},{"cited_title":"Jarque and A","cited_arxiv_id":null,"evidence_quote":"Provides the Jarque-Bera normality test used to reject Gaussian residuals."},{"cited_title":"Davis, The skewed generalized t distribution: Tree package vignette,R Vignette (2015), https://cran.r-project.org/web/packages/sgt/vignettes/sgt.pdf","cited_arxiv_id":null,"evidence_quote":"Supplies the skewed generalized t density used to model the residuals."}],"review_version":1}