{"id":"4351da21-45d0-4e47-9eeb-947addc92ee6","arxiv_id":"2505.07880","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A new two-template empirical SN Ia color model trained on Nearby Supernova Factory data finds that about 13% of phase-independent flux variation is not explained by dust.","lead":"This paper builds a data-driven model of Type Ia supernova brightness with two separate color variation templates, one of which behaves like dust and one of which does not. The authors estimate that about 13 percent of the color variation is not dust-like, suggesting that intrinsic supernova diversity is mixed into standard light curve models.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 13% 'non-dust-like' fraction is measured perpendicular to the CCM89 a(λ)/b(λ) plane; any non-CCM89 extinction dependence in the sample is therefore counted as intrinsic, so the quantitative 'intrinsic variation' claim is weaker than stated.","rationale":"The paper's central claim is not just that a second phase-independent chromatic direction exists (plausible and supported by the coherence of the MVR and CCM89-derived bases, the 80° plane separation, and the lack of correlation between c2' and SALT2 c), but that the additional direction is intrinsic rather than dust-like, quantified as 13% of phase-independent flux variance. That quantitative label is defined relative to the CCM89 a(λ)/b(λ) plane. All plausible real extinction curves—Milky Way with non-CCM89 grain-size distributions, host-galaxy curves with different power-law slopes, circumstellar dust—do not lie exactly in that plane, and Section 4.4's projection places any such component in the 'not dust-like' bucket. The paper acknowledges the complementary contamination (intrinsic variation entering L1', §3.4.2) and the phase-averaging ambiguity (§4.1.3), but does not test how robust the 13% is to the assumed dust family. Because the claim is stated as a variance fraction in the abstract, and the metric is actually the median of normalized per-SN projection magnitudes rather than a variance decomposition, the quantitative headline has additional fragility: reframing as a proper variance ratio could move the number substantially even before dust-model questions are considered. The sampling issues the reader notes (§3.5 non-convergence, SALT2-based sigma clipping) are real but secondary: the paper checks that per-SN refits agree across chain groups, and the clipping direction (removing reddened SNe) mainly affects the size of the dust variance. The recommended remedy is the cheap geometric re-computation described in the concrete test; because that test could move the headline claim, CONDITIONAL is the right stance, and this read does not change the reader's verdict.","tokens_in":31203,"tokens_out":15794,"duration_ms":165975,"concrete_test":"Recompute the Section 4.4 perpendicular fraction and the Figure 8 decomposition with the CCM89 plane replaced by an alternative extrinsic-dust basis, e.g., the span of two power-law curves E(λ)∝λ^{-β} with different β, or the Fitzpatrick & Massa (2007) parametrization, using the already-sampled bivector L, L1', L2', and per-SN c distribution (no refit required). If the median perpendicular fraction falls from 0.13 to below ≈0.03 for any plausible dust family, the 13% 'non-dust-like' variance is an artifact of assuming CCM89 and the intrinsic claim is unsupported; if the fraction remains ≈0.1 or larger for all tested dust bases, the intrinsic interpretation is robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's headline number—'approximately 13% of modeled phase-independent flux variance is not dust-like'—is an output of Section 4.4's geometric computation, in which 'dust-like' is defined as lying in the plane spanned by the CCM89 curves a(λ) and b(λ) (§3.4.2). The claim that the detected second phase-independent component is intrinsic therefore carries a hidden premise: that any real extrinsic extinction in the SNfactory sample lies within the CCM89 two-parameter family. That premise is insecure. Host-galaxy or circumstellar extinction with different functional forms (power-law E(λ)∝λ^{-β} slopes, Fitzpatrick & Massa-style parametrizations, or dust with varying grain properties) would contribute a nonzero component perpendicular to the CCM89 plane, and that component is exactly what Section 4.4 reports as 'not dust-like' and interprets as intrinsic. The paper explicitly flags the reverse leakage (intrinsic variation contaminating L1', §3.4.2) and the phase-averaging ambiguity (§4.1.3), but not this forward leakage (non-CCM89 dust contaminating the perpendicular fraction). The large 80° plane-separation angle (§4.1) and the sign flip in L2' are consistent with non-CCM89 dust as well as with intrinsic variation, so they do not resolve the ambiguity. The weaker, structural claim—that SALT2-class models have only one phase-independent degree of freedom and can leak—survives, but the quantitative 13% and the label 'not dust-like' rest on a single assumed dust family.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents a Stan-based empirical SN Ia spectral flux model with one phase-dependent and two phase-independent chromatic flux variation templates, trained on 73 sigma-clipped SNfactory SNe Ia using nλ=10 spectrally binned photometry. Because the two phase-independent templates span a plane, the authors use bivector methods to define two interpretable bases: a maximum-variance-ratio (MVR) basis and a CCM89-derived basis whose first vector lies on the intersection of the model plane with the CCM89 dust plane. In both bases, the second template shows rapidly varying, spectral-feature-like structure blueward of about 4500 Å that is not consistent with standard dust, and the authors report that approximately 13% of phase-independent chromatic flux variability is not dust-like. They argue that SALT2-class models with a single phase-independent color template can leak intrinsic variation into that template.","tokens_in":31588,"tokens_out":7360,"duration_ms":74234,"significance":"If the qualitative detection holds, the paper is a useful contribution to SN Ia standardization: it provides evidence for a second phase-independent chromatic degree of freedom and warns that single-color-template models may conflate intrinsic and extrinsic variation. The model architecture is genuinely agnostic in its training, the use of bivectors to handle basis ambiguity in the two-dimensional template plane is elegant, and the comparison with SALT2, SNEMO, and Twins Embedding is informative. The authors also deserve credit for explicitly flagging reverse leakage (intrinsic variation contaminating L1') and the phase-averaging ambiguity. However, the quantitative headline is currently not as robust as the abstract suggests: the 13% statistic as computed is not a variance fraction, and the label 'not dust-like' is conditional on the completeness of the CCM89 two-parameter dust family.","major_comments":[{"comment":"The abstract states that 'approximately 13% of modeled phase-independent flux variance is not dust-like,' but the calculation described in §4.4 does not compute a variance fraction. The procedure normalizes each SN's vector c = c1' L1' + c2' L2', computes the perpendicular component of each normalized vector with respect to the CCM89 plane, and reports the median of the resulting per-object values, with 68% interval [0.05, 0.4]. A variance fraction would instead be a variance-weighted quantity such as Σ_i |c_{i,⊥}|^2 / Σ_i |c_i|^2. The median of normalized directions can differ substantially from the variance fraction, especially when the variance is dominated by a subset of objects. The wording of the abstract and conclusion should be changed to describe the statistic actually computed, or the variance-weighted quantity should be reported.","section":"§4.4"},{"comment":"The 'not dust-like' classification is made with respect to the plane spanned by the CCM89 a(λ) and b(λ) curves. The paper explicitly warns that intrinsic variability can contaminate L1' (reverse leakage), but it does not address the forward leakage: any extrinsic extinction whose wavelength dependence lies outside the CCM89 span—for example, power-law extinction with varying β, Fitzpatrick-Massa style parametrizations, or dust with different grain properties—contributes to the perpendicular component and is therefore counted as 'intrinsic.' Because the 13% figure is computed as the perpendicular component relative to that same plane, the headline quantitative claim is conditional on CCM89 being a complete description of the dust in the sample. The 80° plane-separation angle and the sign flip in L2' are consistent with non-CCM89 dust as well as with intrinsic variation. The authors should either test alternative dust families or reframe the 13% as 'non-CCM89-like' rather than 'not dust-like.'","section":"§3.4.2; §4.4"},{"comment":"The training set is curated by a 2σ clip of SALT2 c and x1 along their longer tails, which explicitly removes heavily reddened SNe Ia with peak B−V > 0.18. Since the reported 13% is a relative measure of non-dust-like to dust-like variance in this selected sample, the removal of reddened objects can inflate the non-dust fraction by reducing the dust-like variance in the denominator. The manuscript should report how the 13% value changes with the clipping threshold, or analyze an unclipped sample, before presenting the number as a population statement about SNe Ia.","section":"§2"},{"comment":"The error model is an important modeling choice for the central claim: the paper adds 2% of maximum observed flux in quadrature and adopts a Cauchy likelihood, explicitly to achieve stable convergence. Because L2' and L2_mvr are residual-driven templates, the evidence for a second phase-independent component depends on this error model not creating or suppressing the signal. The pull-distribution check is useful, but the robustness of the recovered L2 template and of the 13% statistic should be demonstrated against alternative choices, such as a Student-t or Normal likelihood with the same added variance, or different values of the 2% term.","section":"§3.2, Eq. (6)"}],"minor_comments":[{"comment":"The sentence 'These two phase-dependent components provide the flexibility...' should refer to the two phase-independent components; as written it contradicts the model description.","section":"§1.2"},{"comment":"The caption says both yellow and blue points compare c′2 and c′1; one of these should be s1 versus c′2 (or c′1 versus c′2) to match the corner plot.","section":"Figure 14 caption"},{"comment":"The normalization expression contains a bare 'q' before the sums; this appears to be a typesetting error for a square root and should be corrected.","section":"Eq. (7)"},{"comment":"The table formatting introduces spaces in decimal numbers, e.g., '0 .12'; this is a LaTeX rendering issue that should be fixed for readability.","section":"Table 1"},{"comment":"The 'space station crew' analogy and the term 'tail wagging' are informal for a journal article; consider tightening this paragraph.","section":"§3.5"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and the qualitative claim is interesting and worth publishing if the quantitative analysis is made rigorous. The main concerns are fixable: the 13% statistic needs to be either recomputed as a variance fraction or described accurately, and the CCM89-completeness caveat should be made prominent in the abstract and conclusions. I do not see concerns about novelty or disclosure; the authors are transparent about the model's limitations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take on 2505.07880. The genuinely new thing is the model architecture: three chromatic templates—one phase-dependent, two phase-independent—trained with no assumed dust law on SNfactory spectrophotometry. That alone distinguishes it from SALT2, BayeSN, SNEMO, and SUGAR. The paper then examines the 2D phase-independent plane with two independently motivated bases (MVR and CCM89-derived), and both lead to the same picture: the first template is dust-like, the second carries spectral-feature-like variation blueward of ~4500 Å that aligns with Ca II H&K and other lines and is uncorrelated with SALT2 c (SCC 0.01) while the first correlates at 0.78. The structural conclusion—that single-color-template models can leak intrinsic variation into their color law—is well supported and matters for LSST-era standardization.\n\nThe paper is unusually candid about its own limits: HMC chains don't fully converge (three similar group solutions; two samplers rejected), the training set is sigma-clipped in SALT2 c/x1, and the model can't separate phase-averaged from truly phase-independent variation. Those are real caveats, but they weaken the quantitative result, not the qualitative one.\n\nThe soft spot is the headline 13% number. “Not dust-like” is defined as “not in the plane spanned by CCM89 a(λ) and b(λ).” L1' is the intersection of the model plane with that CCM89 plane, and the perpendicular fraction is measured to that same plane. If actual extinction (or circumstellar dust, or wavelength-dependent scattering) has functional forms outside that two-parameter family, those contributions are automatically scored as non-dust. The paper flags the reverse leakage—intrinsic variation contaminating L1'—but not this forward one. The 13% should be read as “not CCM89-like,” not “intrinsic.” This doesn't kill the paper; the second template's small-scale spectral structure is hard to explain with any smooth extinction, so the qualitative point survives.\n\nMath, data, citations all look solid. The Stan implementation is standard, the geometric algebra is overkill but correct, and the comparison to SALT2 and SNEMO is fair.\n\nBottom line: worth a serious referee. I'd bring it to the reading group and would cite it in my own work. A revision that renames the 13% “not CCM89-like” and adds a posterior predictive check would put it on firmer footing. Engage with it.","headline":"A genuinely new two-template phase-independent SN Ia model that supports the qualitative leak claim; the headline 13% is CCM89-relative, not a direct measure of intrinsic variance.","tokens_in":32333,"tokens_out":3562,"would_cite":true,"duration_ms":34244,"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":"This paper presents evidence that Type Ia supernovae have a second, non-dust mode of phase-independent chromatic variation, about 13% of the modeled variance, which standard single-color light-curve models can mistake for dust.","keywords":["Type Ia supernovae","empirical light-curve models","chromatic flux variation","dust extinction","intrinsic variability","supernova standardization","spectrophotometry"],"falsifier":"A concrete check would be to train the same model on synthetic light curves generated with only CCM89 dust extinction and no intrinsic phase-independent variation, then measure the perpendicular variance fraction; if the recovered fraction is comparable to 13%, the metric's non-dust signal is an artifact of the dust-plane completeness assumption rather than intrinsic variability.","tokens_in":30974,"feed_emoji":"🔭","tokens_out":9651,"duration_ms":85187,"temperature":0.7,"pith_summary":"The paper sets out to test whether the phase-independent color variation of Type Ia supernovae is really one-dimensional. It builds an empirical light-curve model with two independent chromatic templates and no assumed dust law, fits it to spectrophotometric time series of 73 supernovae, and examines the two-dimensional plane those templates span. The plane intersects the standard CCM89 dust-extinction plane along the dust-like direction, but the orthogonal direction carries variation centered blueward of 4500 Å that dust cannot produce; the paper estimates that about 13% of the phase-independent flux variance is non-dust-like. If the claim holds, common one-color-template models such as the SALT family are likely absorbing intrinsic supernova variation into their dust-like color parameter, which would bias both the inferred dust law and standardized distances.","feed_headline":"Supernovae show 13% of color variance is intrinsic, not dust","feed_subtitle":"A two-template color model finds standard single-color light-curve fits may absorb that intrinsic variation as dust.","key_machinery":"The load-bearing object is a two-dimensional plane in wavelength space, denoted by the unit bivector $L = L_1 \\wedge L_2$, which encodes all phase-independent chromatic variation while remaining independent of the arbitrary choice of basis vectors. To interpret the plane, the paper constructs a CCM89-derived basis: the first template $L_1'$ is the intersection of the model plane with the plane spanned by the CCM89 dust curves $a(\\lambda)$ and $b(\\lambda)$, and $L_2'$ is a $90^\\circ$ rotation within the model plane, maximizing its component perpendicular to the dust plane. The per-supernova coefficients in this basis are then projected onto the complement of the CCM89 plane, and the median fraction of the resulting perpendicular component yields the 13% non-dust variance figure.","core_discovery":"On the paper's own terms, the central discovery is that a physics-agnostic three-template model for Type Ia supernova light curves—one phase-dependent template and two phase-independent chromatic templates—converges to a stable two-dimensional phase-independent variation plane. In the CCM89-derived basis, the first template is the line shared with the CCM89 dust plane and behaves like a dust law, while the second is orthogonal in the model plane and is dominated by rapidly varying chromatic flux, with its strongest features in the Ca II H&K region and other lines blueward of 4500 Å. Projecting each supernova's $c_1, c_2$ coefficients onto the complement of the CCM89 plane gives a median non-dust variance fraction of 0.13 with a 68% range [0.05, 0.4]. The paper concludes that this non-dust component is effectively intrinsic variation that previous single-phase-independent-template models can 'leak' into their color parameter, so the dust-like template and the color-luminosity relation are not purely extrinsic.","pith_inferences":["A direct test of the 13% claim is to simulate supernovae with a known two-component intrinsic color distribution, fit them with a single-color template model, and check whether the recovered non-dust residual has the same strength and wavelength signature; the paper does not run this injection test.","The geometric-algebra plane machinery could be carried into higher spectral resolution: with more than ten wavelength bins, the second template should either sharpen into identifiable spectral features, supporting intrinsic variation, or dissolve into noise, undermining the 13% figure.","If the two-dimensional phase-independent plane is a general property of Type Ia supernovae, the same agnostic architecture could be applied to other transient classes to test whether a second chromatic mode appears there as well.","The paper does not correlate the $c_2'$ coefficient with host-galaxy properties; testing whether the second mode tracks stellar mass or local environment would help decide whether it is intrinsic to the explosion or an environmental extinction effect."],"forward_implications":["Standard one-color-template models, including the SALT family, cannot represent the second phase-independent mode, so their color parameter mixes dust extinction with intrinsic variation; the paper's 13% figure quantifies the fraction of variance at risk of being misattributed.","If a second chromatic template is included, residual scatter in standardized distances should drop or, at minimum, the systematic floor from modeling should be lower than the roughly 0.08 mag level of spectral standardization.","The non-dust template's concentration blueward of 4500 Å implies that spectral regions containing Ca II H&K, Si II, and Fe II lines carry information about intrinsic diversity that photometric color alone could mistake for reddening.","The recovered dust-like first template gives $R_V \\approx 2.4$ in the CCM89 basis and $R_V \\approx 2.18$ in the maximum-variance basis, but the paper stresses that these values are not physical extinction-law measurements because intrinsic variation can leak into them."],"supporting_citations":[{"why":"Supplies the published rest-frame Nearby Supernova Factory spectra used as the training data.","marker":"Aldering et al. 2020"},{"why":"Defines the CCM89 dust-extinction plane that anchors the non-dust variance metric.","marker":"Cardelli et al. 1989"},{"why":"Provides SALT2.4 fits of the sample and the comparison baseline for per-supernova parameter correlations.","marker":"Betoule et al. 2014"},{"why":"The SNEMO empirical model trained on the same data, used for comparison of phase-independent eigenvector behavior.","marker":"Saunders et al. 2018"},{"why":"The Twins Embedding nonlinear model whose phase-independent spectral variation near 4500 Angstroms aligns with the second template.","marker":"Boone et al. 2021b"},{"why":"The SUGAR model and source of the linear transformation used to decorrelate the achromatic offset from the chromatic coefficients.","marker":"Léget et al. 2020"},{"why":"Provides SALT3's recovered color template, used to verify that standard SALT color curves align with CCM89 where the second template begins to vary.","marker":"Kenworthy et al. 2021"}],"fun_headline_variants":["Supernovae: 13% of color shift is intrinsic, not dust","Agnostic model reveals intrinsic color variation in supernovae","Type Ia supernovae: 13% intrinsic variability, not dust","Dust-like assumptions hide 13% intrinsic supernova color variation","Empirical model: supernova color variation is 13% intrinsic"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the CCM89 dust-extinction model spans all possible extrinsic chromatic variation; if real extinction or other environmental effects can produce wavelength dependence outside the CCM89 $a(\\lambda), b(\\lambda)$ subspace, the perpendicular component attributed to intrinsic variability will be overestimated.","fun_headline_variants_meta":{"raw":{"variants":["Supernovae: 13% of color shift is intrinsic, not dust","Agnostic model reveals intrinsic color variation in supernovae","Type Ia supernovae: 13% intrinsic variability, not dust","Dust-like assumptions hide 13% intrinsic supernova color variation","Empirical model: supernova color variation is 13% intrinsic"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001095,"raw_usage":{"total_tokens":4566,"prompt_tokens":937,"completion_tokens":3629,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":3536}},"tokens_in":553,"tokens_out":3629,"duration_ms":22510,"temperature":1.0,"reasoning_tokens":3536,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:36:50.453614+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check would be to train the same model on synthetic light curves generated with only CCM89 dust extinction and no intrinsic phase-independent variation, then measure the perpendicular variance fraction; if the recovered fraction is comparable to 13%, the metric's non-dust signal is an artifact of the dust-plane completeness assumption rather than intrinsic variability.","supporting_citations":[{"cited_title":"2020, Research Notes of the American Astronomical Society, 4, 63, doi: 10.3847/2515-5172/ab8fa5","cited_arxiv_id":null,"evidence_quote":"Supplies the published rest-frame Nearby Supernova Factory spectra used as the training data."}],"review_version":1}