{"id":"14118813-8a00-4c59-9a52-e981dc54ed78","arxiv_id":"2607.18048","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Four stellar-population models overpredict the 2.4–5 μm stellar continuum of quiescent compact galaxies by 0.1–0.3 mag, with the biggest excess at intermediate ages (1–5 Gyr).","lead":"This paper tests four widely used stellar population synthesis models against new SPHEREx infrared observations of nearly 3,900 compact galaxies. All four models predict 10–30% too much infrared light at 2.4–5 μm, a bias that would change galaxy mass and age estimates if confirmed.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The z-band scalar cross-calibration is the key load-bearing step: if SPHEREx-to-SDSS normalization has a color term, the claimed 2.4–5 μm overpredictions and the E-MILES CO advantage could be instrumental rather than stellar-physics.","rationale":"The reader's weakest_assumption identifies the same load-bearing concern: the wavelength-independent scalar cross-calibration in Sect. 3.2 is applied to all SPHEREx channels, and the entire wavelength-dependent offset measurement rests on this scalar being correct at all wavelengths. I agree that this is the most load-bearing step. The paper does not test for a color term in the SPHEREx-to-SDSS normalization; the WISE W1 comparison at 3.4 μm is a single-point check that cannot constrain a slope toward 4–5 μm. Since the claimed E-MILES advantage at 3.8–5 μm and the overall model overpredictions are the core results, an undiagnosed color term would directly undermine the central claim. The reader's verdict is CONDITIONAL, which is appropriate: the concern is concrete and testable, but not yet demonstrated. No additional load-bearing concerns outweigh this one. The 10% model-flux floor is acknowledged as a caveat but does not invalidate the 0.2–0.3 mag offsets seen in the non-E-MILES models. The Appendix C regression (R² < 0.4) does address the age/metallicity degeneracy, so the attack is not on that front. Therefore I keep the verdict UNCHANGED, as the reader's condition already captures the needed test.","tokens_in":27967,"tokens_out":2718,"duration_ms":26595,"concrete_test":"Compute the SPHEREx-to-SDSS scaling factor separately in at least three independent wavelength windows (e.g., SDSS z and y, and WISE W1) and re-run the offset analysis with each. If the 2.4–5 μm median offsets (Figs. 6, 9) shift by more than the claimed 0.05–0.1 mag when using a factor derived from a different band, the scalar calibration is insufficient. Independently, directly compare SPHEREx Band 6 (4.42–5.0 μm) fluxes to WISE W2 (4.6 μm) for the same compact-galaxy sample; a non-flat SPHEREx/WISE ratio across 3.4–4.6 μm would indicate a calibration slope that could produce the E-MILES 'CO advantage' artifactually.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that four SPS models overpredict the 2.4–5 μm stellar continuum by 0.1–0.3 mag with a strong intermediate-age dependence (Sects. 4.2, 5.1). This claim rests on the wavelength-dependent shape of the observed SPHEREx SEDs relative to optical-fitted models. In Sect. 3.2, the SPHEREx-to-SDSS cross-calibration is computed entirely from the SDSS z-band ratio and applied as one wavelength-independent scalar to every SPHEREx channel. The paper states that this correction 'does not affect the overall shape of observed SPHEREx SEDs,' but that is only true if the correction is genuinely wavelength-independent. If the true PSF/aperture difference between SDSS and SPHEREx has a color term—e.g., a few percent per octave in wavelength—the wavelength-dependent offset trends in Figures 6 and 9 would be partly instrumental, not intrinsic to the models. The WISE W1 check at 3.4 μm samples only one wavelength, near the peak of the claimed 2.4–3.2 μm offset; it cannot rule out a slope toward 4–5 μm. Notably, the E-MILES advantage at 3.8–5 μm, attributed to CO absorption, is exactly where no external cross-check is provided. The median 4% offset and the >20% cut do not test for a wavelength-dependent residual. Thus the load-bearing assumption is not merely an uncertainty on the zero-point; it directly affects the shape of the measured offset curves.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses SPHEREx QR2 spectrophotometry as an out-of-sample test of four NIR-covering stellar population synthesis models (E-MILES, BC03, CB19, FSPS). From 3,889 SDSS compact galaxies (2,726 non-ELGs, 1,163 ELGs), the authors perform pPXF full-spectrum fits to SDSS optical spectra, predict rest-frame NIR SEDs without using any SPHEREx data, and compare them to SPHEREx photometry. The central findings are that all models overpredict the 2.4–5 μm continuum of non-ELGs by ~0.1–0.3 mag, with the overprediction strongest for intermediate-age (~1–5 Gyr) populations; E-MILES shows the smallest offsets at 3.8–5 μm, attributed to its CO absorption treatment; and ELGs show the opposite (underprediction) trend from non-stellar emission.","tokens_in":28281,"tokens_out":4771,"duration_ms":55316,"significance":"If correct, these results constitute one of the first large-sample, externally validated tests of NIR SPS predictions, with direct implications for TP-AGB and cool-star modeling, mass-to-light estimates, and full-SED fitting. The design has real strengths: SPHEREx data are held out of the fits, the sample is large, stellar kinematics are validated against SDSS, and the Appendix C regression (R²<0.4) supports the claim that offset differences are not driven simply by the small age/metallicity differences among model fits. The WISE W1 check gives some reassurance on SPHEREx zero-points, although it is a single wavelength. The main caveat is the cross-calibration's assumed wavelength independence; if that fails, the spectral shape of the offsets could be partly instrumental.","major_comments":[{"comment":"The SPHEREx-to-SDSS scaling factor is derived from the z-band ratio and applied as a scalar to all channels; the paper's assertion that this does not affect SED shape is untested. The WISE W1 check at 3.4 μm cannot rule out a color term that produces a slope toward 4–5 μm, exactly where the E-MILES CO advantage is claimed. Please provide a multi-wavelength cross-calibration test: e.g., compare SPHEREx synthetic photometry with WISE W2, or derive the scaling in each band using stellar calibrators, or at least show the SPHEREx/SDSS ratio as a function of wavelength for galaxies with high S/N. Absent this, the central wavelength-dependent offset curves are not yet robustly established as intrinsic to the models.","section":"Section 3.2, Eq. (1), Figs. 6/9/10"},{"comment":"The SDSS fiber-to-total correction is a single r-band scalar applied to both the observed spectrum and the model SED. For galaxies with radial color gradients, this can also distort the NIR spectral shape, since the correction should vary with wavelength. The compact selection mitigates but does not eliminate this. Please quantify the possible color-gradient effect, e.g., by comparing SDSS model magnitudes in g,r,i,z and checking whether the fiber-to-total ratio is color-independent.","section":"Section 3.1, aperture correction"}],"minor_comments":[{"comment":"The text says 'we adopted the magnitude unit for the weighted offset parameter,' but Eqs. (1)-(2) define a flux-weighted average. Please provide the explicit conversion from the flux ratio Δf to the magnitude offset μΔ, or revise the equations to compute the offset directly in magnitudes.","section":"Section 3.2, Eq. (2)"},{"comment":"The caption states that WISE photometry is multiplied by the SPHEREx-to-SDSS z-band scaling factor, contradicting Section 3.2, where no scaling is applied to WISE data because only the W1−W3 color is used. Correct this inconsistency.","section":"Figure 3 caption"},{"comment":"Minor typo: 'Andreas L. F aisst' should presumably be 'Andreas L. Faisst'.","section":"Author list"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong candidate if the calibration concern is addressed. The missing multi-wavelength cross-check is a data-analysis addition rather than a conceptual change, so I would not reject. The central result is plausible and the out-of-sample design is a definite strength."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nIf you work on SPS models, read this. Using SPHEREx photometry of ~3900 compact SDSS galaxies, the authors fit four SPS models (E-MILES, BC03, CB19, FSPS) to optical spectra only, then compare their predicted NIR SEDs to SPHEREx data that never entered the fit. The result is a clear, out-of-sample benchmark: all four models overpredict the stellar continuum at 2.4–5 μm by 0.1–0.3 mag, the excess grows for intermediate-age populations, and E-MILES visibly outperforms the others at 3.8–5 μm, plausibly because of CO absorption in its library.\n\nThe central claim mostly holds. The design is strong—large sample, kinematics validated against SDSS, WISE W1 agreement at ~3.5%, and the offsets correlate with stellar age and WISE color in ways that point to stellar-physics origin. The Appendix C regression (R² < 0.4) does what it should: differences in recovered ages/metallicities don't drive the offsets.\n\nThe stress-test note points at the real soft spot: the SPHEREx-to-SDSS cross-calibration is a single scalar from the z-band ratio applied to all channels. If that scalar has a color term, the wavelength trends in Figures 6 and 9 could be partly instrumental. I read the paper looking for a test; there isn't one. The WISE W1 check sits at 3.4 μm, near the peak of the offset, so it can't rule out a slope into Bands 5–6. This is the main technical request for revision: test for wavelength-dependent scaling, e.g., by using stellar templates or cross-calibrating against a second photometric band at 4–5 μm. It doesn't sink the paper because the intermediate-age bump and the SSP-level comparisons in Appendix B don't depend on the absolute zero-point, but the 4–5 μm E-MILES advantage does deserve a check that isn't just one model's SSP ratios.\n\nTwo smaller caveats. The 10% uncertainty floor on model fluxes makes the 0.1-mag end of the range borderline; the 0.2–0.3 mag offsets are solid, but the \"slight overprediction\" for E-MILES at 2.6–3 μm is near the floor. And the paper honestly states that the CO features in E-MILES come from the theoretical Phoenix extension, not observed spectra, so the CO interpretation is reasonable rather than proven. No code or sample catalog is released, which would help reproduction.\n\nSend it to a serious referee. It's a genuinely new empirical constraint on SPS models, the methodology is largely sound, and the open issues are the kind a revision can address.","headline":"Genuinely new out-of-sample test: all four SPS models overpredict 2.4–5 μm flux by 0.1–0.3 mag; caveats are fixable, not fatal.","tokens_in":29032,"tokens_out":3621,"would_cite":true,"duration_ms":39538,"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":"Four stellar population models overpredict galaxy near-infrared light by 0.1–0.3 mag.","keywords":["stellar population synthesis","near-infrared spectral energy distributions","SPHEREx","SDSS galaxies","TP-AGB stars","CO absorption","spectral fitting","galaxy evolution"],"falsifier":"Recompute the offsets using a wavelength-dependent SPHEREx calibration built from multiple standard sources across 2.4–5 μm instead of one z-band ratio; if the 2.6–3 μm overprediction peak disappears or shifts, the claimed stellar-physics bias would be undermined.","tokens_in":27744,"feed_emoji":"🔭","tokens_out":4618,"duration_ms":51502,"temperature":0.7,"pith_summary":"This paper tests four stellar population synthesis models by fitting the optical spectra of about 3,900 compact galaxies and comparing each model's predicted near-infrared SED against SPHEREx spectrophotometry from 0.75 to 5 μm. It finds that all four models overpredict the stellar continuum at 2.4–5 μm by 0.1–0.3 magnitudes, with the largest excess for intermediate-age (1–5 Gyr) stellar populations. The paper argues this is a systematic flaw in how the models treat NIR-bright evolved stars—TP-AGB stars and cool giants—not a calibration artifact. Among the models, E-MILES matches the observed fluxes best at 3.8–5 μm because its empirical cool-star library includes CO absorption features the other models lack.","feed_headline":"Models overpredict galaxy near-IR light by 0.1–0.3 mag","feed_subtitle":"Four population codes fit to optical spectra shine too bright at 2.4–5 μm; one model gets the longest wavelengths right.","key_machinery":"The key machinery is a two-step comparison: full-spectrum fitting of SDSS optical spectra (3500–7000 Å) to determine stellar population parameters, then projecting the best-fit model SEDs into the NIR and measuring a weighted offset against SPHEREx spectrophotometry on a common grid of 102 wavelength channels. This isolates each model's NIR predictive power because the fitting never sees the NIR data. The crucial model ingredient driving the wavelength-dependent differences is the stellar spectral library: E-MILES relies on an empirical library of cool-star spectra with resolved CO bands, while the other three use a smoother theoretical library that omits these features.","core_discovery":"The central claim is that the near-infrared continua predicted by E-MILES, BC03, CB19, and FSPS are systematically too bright at 2.4–5 μm relative to observed SPHEREx fluxes by 0.1–0.3 mag, even when the models are constrained only by optical SDSS spectra. The excess is strongest for galaxies with intermediate-age stellar populations (1–5 Gyr), linking the bias to TP-AGB and cool-star physics. E-MILES is the exception at 3.8–5 μm, where its empirical spectral library reproduces the 4.2–4.5 μm CO absorption that the other models do not include, bringing its predictions into much closer agreement with the data.","pith_inferences":["A natural extension, which the paper does not pursue, is to apply the same optical-fit-then-NIR-check procedure to star clusters of known age and metallicity; this would separate TP-AGB effects from the degeneracies of composite stellar populations.","If the age trend is real, re-fitting with TP-AGB luminous fractions artificially reduced should erase most of the 2.6–3 μm excess—a testable prediction that could guide library updates.","The lack of correlation between the offsets and WISE color in quiescent galaxies, noted in the paper, implies the bias is stellar, but the paper does not discuss that the same data could constrain the temperature and metallicity coverage needed in future NIR spectral libraries."],"forward_implications":["Stellar masses and star-formation histories derived from NIR photometry with these models carry a systematic 10–30% flux bias, which would propagate into mass-to-light ratios and age estimates.","The measured offsets give concrete, wavelength-resolved targets for recalibrating TP-AGB and cool-star prescriptions in the next generation of population synthesis models.","For emission-line galaxies, the models underestimate the NIR SED because they exclude non-stellar dust and PAH emission; this sets a floor for how much of any observed NIR excess must be attributed to dust rather than stars.","The persistence of the overprediction across four independently constructed models signals that the missing physics is common—likely molecular absorption or TP-AGB lifetimes—not a single model's quirk."],"fun_headline_variants":["Near-IR galaxy light overpredicted by 0.1–0.3 mag in all four codes","Stellar population models shine too bright at 2.4–5 μm","SPHEREx checks SPS models: systematic NIR overprediction found","Intermediate-age stars drive near-IR excess in population models","E-MILES matches data best at longest NIR wavelengths"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire measurement rests on a single scalar calibration between SPHEREx and SDSS derived only in the z-band; if that calibration has a wavelength-dependent error, the trends at 2.4–5 μm could be partly instrumental.","fun_headline_variants_meta":{"raw":{"variants":["Near-IR galaxy light overpredicted by 0.1–0.3 mag in all four codes","Stellar population models shine too bright at 2.4–5 μm","SPHEREx checks SPS models: systematic NIR overprediction found","Intermediate-age stars drive near-IR excess in population models","E-MILES matches data best at longest NIR wavelengths"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000256,"raw_usage":{"total_tokens":1497,"prompt_tokens":914,"completion_tokens":583,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":658,"completion_tokens_details":{"reasoning_tokens":495}},"tokens_in":658,"tokens_out":583,"duration_ms":6492,"temperature":1.0,"reasoning_tokens":495,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T16:15:49.103841+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the offsets using a wavelength-dependent SPHEREx calibration built from multiple standard sources across 2.4–5 μm instead of one z-band ratio; if the 2.6–3 μm overprediction peak disappears or shifts, the claimed stellar-physics bias would be undermined.","supporting_citations":[],"review_version":1}