{"id":"66fd457b-2328-40fd-81f8-997b3f29c437","arxiv_id":"2608.06611","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper reports 9 new candidate intermediate-scale extinction features and organizes all 17 such features into two families, alpha and beta, with distinct correlations to the 2175 Å carbon bump.","lead":"Using new HST/STIS spectra for 24 stars plus 50 literature sightlines, this paper expands the census of broad interstellar extinction features, proposing 9 new candidate features and grouping all features into two families, alpha and beta. The families show distinct correlations with the 2175 Å UV bump, hinting that they trace different carbonaceous dust populations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The candidate census rests on the assumption that all residual structure beyond a fourth-order polynomial plus Drude profiles is astrophysical; because candidate Drudes are added iteratively to the same residuals used for their significance, continuum/model mismatch can masquerade as new ISS…","rationale":"The reader's weakest assumption correctly identifies the load-bearing point: the detection scheme assumes that, after subtracting a fourth-order polynomial and known Drude features, the remaining residual is astrophysical. This is exactly where the argument is least secure. The iterative procedure in Section 4.3 selects features from the residual and then tests those same features against the same residual, so the statistical significances in Table 5 do not protect against a wrong continuum order, non-Drude shapes for the literature features, or common instrumental/model systematics. The downstream claims of two families and correlation with the 2175 Å bump are built on the candidate list, so an error at the census stage propagates to the classification. The location of several candidates on the shoulders of broad features or near the edges of the fitted range strengthens the concern. The paper's own Section 7 limitation statement is consistent with this reading. I do not see a reason to move away from the reader's CONDITIONAL verdict: the study is careful, the data and code are public, and independent confirmation or a targeted robustness test could settle the question, but the central empirical claim is not yet fully secured.","tokens_in":25327,"tokens_out":6928,"duration_ms":72141,"concrete_test":"Using the public code, refit all 74 extinction curves with a fifth-order polynomial continuum (Np=5) in Eq. 4 and repeat the Section 4.3 iterative candidate search. Then repeat once more using a non-parametric spline continuum with knots spaced roughly 1000 Å instead of a polynomial. Require each of the 11 candidate features to be recovered at >5σ with central wavelength and width consistent within their quoted uncertainties under both alternative continuum models, and additionally require that the recovered amplitudes agree within 2σ between the 24 new Prince sightlines and the 50 F19 sightlines when analyzed separately. Any candidate that fails these tests is likely absorbing continuum or model flexibility rather than a robust interstellar extinction feature.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of nine new candidate ISS features depends on the decomposition in Eq. 4: each optical extinction curve is represented as a fourth-order polynomial plus Drude profiles for known features. Section 4.3 then iteratively inspects the mean residual, adds Drudes at 4010, 5150, 5890, 6210, 7420, and 8130 Å, then adds more at 3240, 3660, 4460, 6960, and 9520 Å, continuing until the residual is flat. The significances in Table 5 are obtained by MCMC-fitting this same mean residual with the standard deviation of the mean as the uncertainty. This is partly circular: the features were selected because they appear in that residual, so a non-Drude residual shape or a mismatch between the true continuum and a fourth-order polynomial will be absorbed into an extra Drude and can appear highly significant if the mismatch is common to many sightlines. The paper itself concedes in Section 7 that if the six literature ISS features are not well-modelled by Drude profiles, some candidates may be artifacts. Several candidate placements make this worry concrete: ISS44 sits on the broad wing of ISS43, ISS52 lies between ISS48 and ISS54, ISS75 lies between ISS77 and ISS81, and ISS32 and ISS96 sit near the 3000 Å and 10000 Å ends of the fitted range, where a fourth-order polynomial is least constrained and where instrumental fringing and order-edge effects are strongest. The quoted 5σ significances do not test the null hypothesis that the continuum has a different order or that the literature features have non-Drude shapes.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a homogeneous analysis of intermediate-scale structure (ISS) in optical extinction curves using 74 lines of sight, combining 50 literature sightlines with 24 new HST/STIS observations. The authors fit each extinction curve with a fourth-order polynomial plus Drude profiles for known ISS features, then iteratively add Drude profiles to the mean residual until it is flat. They confirm five of six literature ISS features at >5σ, fail to confirm ISS54, and report ten new candidate features (later reduced to nine after excluding ISS96). Based on Pearson correlations between fitted amplitudes, they group most features into two families, α and β, and report strong correlations of ISS43, ISS48, ISS64, and ISS77 with the 2175 Å bump strength, which they interpret as evidence for carbonaceous carriers.","tokens_in":25661,"tokens_out":5516,"duration_ms":51161,"significance":"If the candidate features are real, this work substantially expands the census of broad extinction structure in the Milky Way and provides a new phenomenological organization (two families) that could constrain dust models. The paper's strengths include the new STIS data, the homogeneous re-analysis of the combined sample, the public release of code and data, and an explicit statement of the main assumption (Drude representation of literature features) in Section 7. However, the detection significance and the correlation-based family classification rest on several methodological choices that need additional validation before the central claims can be accepted.","major_comments":[{"comment":"The significance estimates for the candidate ISS features are computed by MCMC-fitting the same mean residual from which the features were selected, using the standard deviation of the mean as the uncertainty. This is a post-hoc selection: a feature that appears in the residual is then tested against that same residual, so the quoted 5σ values do not include the multiplicity of trials or the uncertainty in the continuum model. The paper's own Section 7 states that the decomposition assumes the six literature features are well-modelled by Drude profiles and that 'some of the proposed candidate ISS features may be artifacts of the method' if this is not the case. Because the discovery claim of nine new features depends directly on this procedure, the authors should provide robustness tests, e.g., fitting with a 5th-order polynomial or an alternative continuum model, a bootstrap or false-discovery-rate analysis, and split-sample validation, to demonstrate that the candidates are not absorbing continuum mis-specification.","section":"4.3, Table 5, Section 7"},{"comment":"The two-family classification is derived entirely from Pearson correlation coefficients between fitted ISS amplitudes, but the paper does not give uncertainties on r, does not state significance thresholds, and does not explain the clustering rule used to assign features to families. Several low-significance features (ISS32, ISS54, ISS62) are tentatively placed in families, while ISS96 is excluded on the basis of a lack of correlation. The authors should propagate the amplitude uncertainties into the correlation analysis (e.g., via bootstrap or Monte Carlo), test the significance of the family separation with a permutation test or a clustering algorithm, and state the criterion used to assign a feature to a family. Without this, the two-family classification is not quantitatively supported.","section":"5, Figure 10, Figure 11"},{"comment":"The claim of strong correlations between the 2175 Å bump strength B3 and ISS43, ISS48, ISS64, and ISS77 is based on Pearson r values without uncertainties. Because B3 and the ISS amplitudes are derived from fits to the same extinction curves and are normalized by A(55), possible correlated errors or the shared normalization could bias r upward. The paper should report p-values or bootstrap confidence intervals for these correlations and ideally use a regression method that accounts for errors in both variables before concluding a carbonaceous origin.","section":"6, Figure 11"}],"minor_comments":[{"comment":"The text says ISS54 cannot be confirmed, yet Table 5 lists it as a feature and Figure 11 places it in Family α; please clarify whether it is considered a detection or a tentative feature.","section":"4.2, Table 5, Figure 11"},{"comment":"The statement that a fourth-order polynomial is adequate is taken from M20; please state whether this was re-verified for the present 74-line-of-sight sample, given the extended wavelength range to 10000 Å.","section":"3.3, Eq. (4)"},{"comment":"The panel labels such as 'model = 4D Poly. + 3 Drudes' are inside the figure and not defined in the caption; define the abbreviations in the caption.","section":"Figure 7"},{"comment":"Entries like 'Yesnew', 'Nonew', 'Unsurenew' in the 'Dust feature?' column are cryptic; use separate columns for 'Literature?' and 'Dust feature?' with a clear key.","section":"Table 5"},{"comment":"The informal naming of the sample and the footnote about Prince, while harmless, is out of keeping with the tone of an ApJ paper; consider a more standard designation.","section":"2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for ApJ and the data/code release is a positive. My main concern is the statistical validation of the detection and classification; the authors' own caveat in Section 7 is appropriately honest but indicates that the central claim is not yet fully supported. I would not reject, but the revision should include the robustness tests described."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a solid, useful paper. The authors re-reduce and homogeneously fit 74 Milky Way sightlines, 24 of them with new HST/STIS data, and apply the established Drude-fitting method of Massa et al. (2020) to look for intermediate-scale structure in optical extinction curves. They confirm the six literature features (though ISS54 only at 3.2σ), add nine new candidates, and arrange the features into two correlation families, α and β, with α features correlating strongly with the 2175 Å bump. The data and code are public, and the paper is refreshingly honest about its own limitations: it explicitly says in Section 7 that if the literature features are not well-modelled by Drudes, some candidates may be artifacts.\n\nThe soft spots are real but not fatal. The main one, which the stress-test note correctly identifies, is that the candidate Drudes are added iteratively to the mean residual, and the quoted significances come from fitting that same mean residual. So a common continuum mismatch or a non-Drude shape in a known feature can masquerade as a new candidate. The fact that the features are averaged over 74 sightlines and are required to correlate with A(55) helps, but it does not fully remove the circularity. The paper acknowledges this in Section 7, which is good, but it means the nine new candidates should be labelled provisional until confirmed on independent data.\n\nThe second soft spot is the statistics of the correlations. Pearson r coefficients are given without uncertainties, and because both ISS amplitudes and B3 are normalized by A(55), a common normalization can artificially inflate correlations. The family grouping is also somewhat qualitative, based on a circle graph. These are minor issues for a discovery paper, but they should be addressed in revision (e.g., bootstrap uncertainties, partial correlations).\n\nWho is this for? Anyone building dust models or interpreting extinction curves. The census is the most complete to date, and the carbonaceous connection to the 2175 Å bump is a genuinely useful constraint. I would send it to peer review without hesitation, and I'd tell the editor the main thing to ask the authors is to make the significance test robust to the continuum choice (e.g., fit with a 5th-order polynomial as a null check) and to report bootstrap errors on the correlations.","headline":"A well-documented, honest census of intermediate-scale extinction features; the nine new candidates are plausible but need independent confirmation.","tokens_in":26236,"tokens_out":2801,"would_cite":true,"duration_ms":25877,"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":"A systematic census of 74 Milky Way sightlines reports nine new intermediate-scale dust extinction features and shows that the full set sorts into two families, α and β, with the α family strongly correlated with the 2175 Å carbon bump.","keywords":["interstellar dust extinction","intermediate-scale structure","2175 Å bump","diffuse interstellar bands","Drude profile","carbonaceous dust","Hubble Space Telescope STIS","Milky Way sightlines"],"falsifier":"Re-fit the same 74 extinction curves with a differently shaped continuum (for instance a spline or a fifth- or sixth-order polynomial) and with an independent set of stellar atmosphere models, then check whether the nine new candidate features still appear at >5σ in the mean residual; if they disappear or drop below 5σ under a reasonable alternative continuum, the claim that they are real interstellar features would be falsified. A complementary test is to look for the same features in higher signal-to-noise spectra of individual sightlines, or to require that the candidate features correlate with mid-infrared carbonaceous emission or absorption features as the carbon-origin claim predicts.","tokens_in":25142,"feed_emoji":"🔭","tokens_out":8471,"duration_ms":69058,"temperature":0.7,"pith_summary":"The paper reports a systematic census of intermediate-scale structure (ISS) in the optical extinction curves of 74 Milky Way sightlines, combining 24 new HST/STIS spectra with 50 literature targets. It finds that the Milky Way's extinction curve contains far more broad structure than previously known: nine new candidate features beyond the six already reported, with fifteen of the seventeen total features detected above 5σ. The paper's central interpretive claim is that these features sort into two families, α and β, according to how their strengths correlate across sightlines. Family α sits at shorter wavelengths and tracks the strength of the 2175 Å bump and the dust size parameter 1/R(55); family β sits at longer wavelengths and correlates more weakly with the bump. If correct, this means the carriers of these broad extinction features form two distinct, possibly competing carbonaceous dust populations, and the strength of the 2175 Å bump can be used to predict part of the optical extinction structure.","feed_headline":"Two families of broad dust features emerge from 74 sightlines","feed_subtitle":"The short-wavelength family tracks the 2175 Å carbon bump and grain size; the long-wavelength family does not.","key_machinery":"The analysis rests on the Drude profile as the model for each ISS feature, $D(x,x_0,\\gamma)=x^2\\gamma^2/((x^2-x_0^2)^2+(x\\gamma)^2)$ in inverse-micron wavenumber $x=1/\\lambda$, added to a fourth-order polynomial that represents the optical continuum. The measurement pipeline fits each sightline's extinction curve with this sum, then examines the residuals after removing known interstellar lines and diffuse interstellar bands and convolving with a Gaussian that suppresses structure narrower than 150 Å. The mean residual across 74 sightlines is used to iteratively add Drude components until the residual is flat; the statistical significance of each feature is computed from the area under its Drude profile relative to the standard deviation of the mean. This combination of a fixed functional form for the broad features and an ensemble average over many independent lines of sight is what lets the paper claim detections of features whose individual amplitudes are only a few percent of A(55).","core_discovery":"The authors establish, on the basis of a homogeneous fit of 74 extinction curves, that the optical spectrum of interstellar extinction contains at least 17 broad (FWHM ≳ 150 Å) features with widths larger than any known diffuse interstellar band. Six of these had been reported before; this work confirms five of them, leaves the 5400 Å feature unconfirmed at 3.2σ, and proposes nine new candidates that survive both a 5σ detection threshold and a check that their strengths rise with dust column A(55). The central new result is a two-family classification: correlation analysis of the feature amplitudes splits the sample into family α (ISS32, ISS36, ISS40, ISS43, ISS44, ISS48, ISS52, ISS54) and family β (ISS59, ISS62, ISS64, ISS69, ISS75, ISS77, ISS81, ISS84), with ISS96 in neither family. Members of family α strongly track each other and the 2175 Å bump strength; family β members track each other and show moderate bump correlations, while ISS36 is anti-correlated with several β features, indicating two carrier populations that can be enhanced or suppressed relative to one another. The paper argues that the correlations with the 2175 Å bump point to a carbonaceous origin for the carriers, with family α's additional correlation with 1/R(55) suggesting a size-sensitive carbonaceous population and family β a size-insensitive one.","pith_inferences":["If the two-family pattern survives in the Magellanic Clouds, the wavelength split between α and β may map onto a difference in electronic transition energies of two carbonaceous grain populations, which would make the α/β ratio a metallicity-sensitive diagnostic.","The authors' comparison of ISS with DIBs suggests a testable extension: a dedicated cross-correlation of ISS family strengths with DIB equivalent widths along the same 74 sightlines would show whether the two are different manifestations of the same carbonaceous carrier population at different size scales.","Because family α tracks 1/R(55) while family β does not, one could use the α/β amplitude ratio as a new observational handle on grain size distributions, independent of the standard R(V) parametrization.","A direct laboratory test: if the carriers are carbonaceous, the features' central wavelengths should match electronic transitions of specific PAH cations or hydrogenated amorphous carbon clusters, which could be checked against published laboratory spectra."],"forward_implications":["The optical extinction curve of the Milky Way is not smooth: it contains at least fifteen statistically significant broad features between roughly 3200 and 9600 Å, nine of them newly reported here.","The two families give a new organizing axis for extinction-curve studies: a sightline's family-α to family-β strength ratio encodes which dust population dominates, allowing extinction curves to be classified by carrier family rather than only by R(55).","The strong correlation of ISS43, ISS48, ISS64 and ISS77 with the 2175 Å bump means the bump strength can be used to predict part of the optical extinction structure, and it ties these four features to carbonaceous carriers.","ISS54 (5400 Å) is not confirmed by this sample, and ISS62 is only marginally detected, so those two features need higher signal-to-noise data before they can be treated as established.","The anti-correlation of ISS36 with family-β features implies the two carrier populations are not simply independent but can be competitively enhanced or suppressed in different environments."],"supporting_citations":[{"why":"Established the three original ISS features at 4370, 4870, 6300 Å and the Drude-plus-polynomial fitting approach the paper adopts.","marker":"D. Massa et al. (2020)"},{"why":"Reported the 7700 Å ISS feature, which this paper confirms as ISS77.","marker":"J. Maíz Apellániz et al. (2021)"},{"why":"Reported the 5400 Å feature (ISS54), which this paper tests and does not confirm.","marker":"R. Zhang et al. (2024)"},{"why":"Reported the 8500 Å feature (ISS84), confirmed here.","marker":"G. M. Green et al. (2025)"},{"why":"Provided 50 of the 74 sightlines and the extinction-curve data this analysis builds on.","marker":"E. L. Fitzpatrick et al. (2019)"},{"why":"Introduced the stellar-atmosphere-based method for deriving extinction curves used here.","marker":"E. L. Fitzpatrick & D. Massa (2005)"},{"why":"Supplied the R(V)-dependent extinction curve model used in the fitting pipeline.","marker":"K. D. Gordon et al. (2023)"},{"why":"Provided the TLUSTY non-LTE model atmospheres used to fit the stellar SEDs.","marker":"I. Hubeny et al. (2025)"},{"why":"Supplied the f(H2) relation and the C4-molecular-phase interpretation used in the correlation analysis.","marker":"D. Van De Putte et al. (2023)"}],"fun_headline_variants":["Broad dust features sort into two families across 74 sightlines","17 broad dust features fall into two distinct families","Dust feature families link to carbon bump and grain size","Two dust families: one tracks carbon bump, other doesn't","Optical extinction reveals two contrasting dust families"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole detection scheme assumes that any residual larger than 5σ left after subtracting a fourth-order polynomial continuum plus known Drude features is a real astrophysical feature, rather than an artifact of the continuum model, the stellar atmosphere models, or the fringe correction; the paper itself warns that if the literature features are not exactly Drude-shaped, some of the new candidates could be artifacts.","fun_headline_variants_meta":{"raw":{"variants":["Broad dust features sort into two families across 74 sightlines","17 broad dust features fall into two distinct families","Dust feature families link to carbon bump and grain size","Two dust families: one tracks carbon bump, other doesn't","Optical extinction reveals two contrasting dust families"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000253,"raw_usage":{"total_tokens":1630,"prompt_tokens":1078,"completion_tokens":552,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":694,"completion_tokens_details":{"reasoning_tokens":474}},"tokens_in":694,"tokens_out":552,"duration_ms":5791,"temperature":1.0,"reasoning_tokens":474,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T04:15:25.024787+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-fit the same 74 extinction curves with a differently shaped continuum (for instance a spline or a fifth- or sixth-order polynomial) and with an independent set of stellar atmosphere models, then check whether the nine new candidate features still appear at >5σ in the mean residual; if they disappear or drop below 5σ under a reasonable alternative continuum, the claim that they are real interstellar features would be falsified. A complementary test is to look for the same features in higher signal-to-noise spectra of individual sightlines, or to require that the candidate features correlate with mid-infrared carbonaceous emission or absorption features as the carbon-origin claim predicts.","supporting_citations":[{"cited_title":"M., Zhang, X., & Zhang, R","cited_arxiv_id":null,"evidence_quote":"Reported the 8500 Å feature (ISS84), confirmed here."}],"review_version":1}