{"id":"cac1d146-06d0-43d6-a272-3e1643f36a02","arxiv_id":"2608.01647","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Little Red Dots at z=2-6 show less than 4% intrinsic H-alpha variability on 1-3 month rest-frame timescales, a flat white-noise pattern unlike normal AGNs, implying different broad-line production.","lead":"This paper measures how much the light from 17 Little Red Dots and 14 blue quasars at redshifts 2 to 6 changed over a few months using repeated JWST spectra. It finds Little Red Dots vary far less than normal quasars in both continuum and hydrogen line emission, suggesting their broad lines are produced differently.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The local-continuum normalization (Sec. 3.1) can cancel correlated Hα+continuum variability, so the ~4% LRD upper limit and white-noise conclusion depend on the assumption that LRD continua are constant; the 3-epoch difference-imaging check is not sufficient.","rationale":"The reader's weakest assumption identifies the same load-bearing point: the spectral normalization assumes the LRD continuum is non-variable, and if that fails the central Hα variability constraint and the white-noise conclusion are biased. I agree with that assessment. The paper is otherwise careful: it uses repeated MSA spectra, compares to an independent reduction, tests for wavelength-calibration systematics with ELGs, and repeatedly frames the variability estimates as upper limits. Those strengths do not remove the need for an independent flux calibration. The proposed test is directly feasible with existing data (F444W aperture photometry for all six epochs, difference imaging for the first three) and would settle whether the normalization is responsible for the low measured Hα rms. Because the concern is substantial but not fatal—the paper's caveats and upper-limit language partially mitigate it—the appropriate disposition remains CONDITIONAL, matching the reader's verdict. No change to the verdict is needed.","tokens_in":34546,"tokens_out":6354,"duration_ms":75116,"concrete_test":"Recompute §3.3 σ0,lc for the 7 typical LRDs using an independent absolute calibration: normalize each of the six MSA epoch spectra with the F444W aperture photometry (already available for all six epochs) instead of the local continuum, and also apply the difference-image photometry calibration for the first three epochs where it exists. If the photometrically calibrated σ0,lc rises to the SDSS-RM level (≳6%) or the per-pair |ΔF_Hα| values increase systematically, the low LRD variability is an artifact of the continuum normalization. As a cross-check, inject a simulated continuum+line light curve with a known 6% Hα response into the real MSA spectra and run the same pipeline; a recovered σ0,lc below ~3% would confirm the bias.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1 normalizes every LRD epoch spectrum by dividing the line profile by the fitted local continuum and rescaling to the epoch with the largest continuum. This removes any fractional flux change common to both line and continuum. In the normal-AGN picture the paper contrasts with, Hα is driven by the ionizing continuum, so a few-percent intrinsic continuum change over 1-3 months would be mirrored by Hα; the normalization would suppress exactly the variability the paper aims to measure. The only direct evidence that the continuum is constant is difference-image photometry from the first three Deep imaging epochs (§4.3), while the spectroscopic variability analysis uses up to six MSA epochs; additionally, F200W/F444W aperture photometry shows ~6% raw scatter before difference imaging, comparable to the quoted 4% Hα rms. If continuum variability is at the few-percent level, the reported σ0,lc=4.02% is not an upper limit on intrinsic line variability but potentially an underestimate biased by the normalization. The white-noise pattern across timescales would then be indistinguishable from an artifact of common-mode removal. The paper is transparent about this assumption and labels results as upper limits, and it does exclude NX7607 from the headline estimate, but the central population claim and the 3.8σ contrast with SDSS-RM rest on this normalization.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents multi-epoch JWST/NIRSpec MSA spectroscopy from the first six NEXUS Deep epochs for 17 Little Red Dots and 14 blue broad-line AGNs at 2 ≲ z ≲ 6. After fitting the Hα complex and normalizing each epoch spectrum (local-continuum normalization for LRDs, aperture-photometric normalization for BLAGNs), the authors measure Balmer decrements, line ratios, and Hα and continuum variability. They report that LRDs have enhanced Balmer decrements, low continuum variability (σ0,lc ≈ 2–3%), and low Hα variability (σ0,lc = 4.02%) on rest-frame ~1–3 month timescales, compared with 6.29% for an Hα-luminosity-matched SDSS-RM sample, a difference they quote as 3.8σ. Combining this with literature results at longer timescales, they argue that LRD broad-line variability follows a white-noise pattern, in contrast to the red-noise behavior of normal AGNs, suggesting different broad-line production mechanisms in LRDs.","tokens_in":34934,"tokens_out":6466,"duration_ms":78102,"significance":"If the central result holds, it is valuable: it provides one of the first spectroscopic, ensemble constraints on high-redshift LRD broad-line variability and connects to a growing body of photometric non-detections. The use of an external SDSS-RM comparison sample, the explicit treatment of flux uncertainties via the σ0 maximum-likelihood estimator, and the authors' repeated caution that their estimates are upper limits are strengths. The sample is small (12 LRDs, 26 Hα flux pairs), however, and the headline 4% constraint is conditional on a local-continuum normalization that assumes a non-varying LRD continuum. That assumption, plus the exclusion of the most variable LRD (NX7607), means the population-level claim and the white-noise interpretation are not yet fully secured. With additional robustness tests and more careful reporting of the NX7607 exclusion, the paper would provide a solid upper-limit benchmark for LRD variability models.","major_comments":[{"comment":"The LRD normalization divides each epoch's Hα profile by the fitted local continuum and rescales to the epoch with the largest continuum. This removes any fractional flux change that is common to Hα and the adjacent continuum. In the standard AGN reverberation picture the paper contrasts with, a few-percent ionizing-continuum change over 1–3 months would induce a proportional Hα response, so the normalization would suppress exactly the signal the paper aims to measure. The only direct evidence that the LRD continuum is constant comes from difference-image photometry of the first three Deep imaging epochs (§4.3), whereas the spectroscopic variability uses up to six MSA epochs; raw F200W aperture photometry shows ~6% scatter, comparable to the quoted 4% Hα rms. Thus σ0,lc = 4.02% (Table 3) is an upper limit only under the assumption of a non-varying LRD rest-optical continuum. Please quant","section":"§3.1, normalization scheme"},{"comment":"The text says 'We exclude NX7607 when obtaining σ0,lc, as its ∼30% variability significantly inflates the estimated intrinsic variability,' yet Table 3 lists 'All LRDs' with N_obj = 12, N_pair = 26, and σ0,lc = 4.02%, which appears to include NX7607 (which has 2 reliable epochs in Table 1). No row is given for the sample excluding NX7607, and §4.2 quotes 4.02% as the main LRD result. Please clarify which sample produced the headline value and report both with and without NX7607. Because NX7607 is the single most variable LRD, its inclusion/exclusion is pivotal to the claimed 3.8σ suppression; this must be transparent and not confined to a single sentence.","section":"§3.3 and Table 3"},{"comment":"The σ0 estimate is obtained by duplicating all |ΔF| values with negative signs, thereby doubling the number of data points without adding independent information. The quoted uncertainty (e.g., 4.02+0.59, Table 3) is therefore likely optimistic. With only 26 independent flux pairs from 12 LRDs, the 3.8σ excess over SDSS-RM is fragile. Please validate with bootstrap resampling of the 26 pairs and with one-object-out/jackknife tests, and state the number of independent (non-duplicated) pairs when quoting significances. The SDSS-RM uncertainty is tiny because of its large pair count; the comparison error is dominated by the LRD side, so a proper small-sample treatment is essential.","section":"§3.3, σ0 estimator"},{"comment":"The conclusion of a 'white-noise pattern across all timescales' is derived by combining the NEXUS 4% upper limit with literature points on yearly-to-decade timescales, several of which are themselves upper limits or marginal detections (TWINKLE; Burke et al. 2026; Furtak et al. 2025). Upper limits cannot demonstrate a flat structure function; they only place an envelope. Moreover, the combined data are heterogeneous (different objects, different normalizations, some photometric, some spectroscopic). Please fit a power-law structure function to the upper limits or provide a statistical test that the ensemble is inconsistent with a red-noise model; otherwise soften the conclusion to 'consistent with low-level variability bounded by current upper limits.'","section":"§4.2 and §5.1, white-noise claim"}],"minor_comments":[{"comment":"The extinction coefficient κ(λ) is used in Eq. (1) before being defined; define R_V and κ in the text immediately before the equation.","section":"§3.2, Eq. (1)"},{"comment":"The F200W 'Typical LRDs' row contains only 1 object and 15 pairs. This is too little to support an ensemble claim; the text should explicitly caution that the F200W typical-LRD constraint is dominated by a single source.","section":"§4.3/Table 3"},{"comment":"Typo: 'inclde' should be 'include' in the final paragraph.","section":"§6"},{"comment":"A few references are arXiv-only or lack full bibliographic details (e.g., Naidu et al. 2025, Chen et al. 2026, Sneppen et al. 2026). Please ensure all cited works have complete journal/volume/page information where available.","section":"References"},{"comment":"The cool LRDs are normalized with aperture photometry while typical LRDs use local continuum, yet both are combined in Table 3's 'All LRDs' row. The text should remind the reader of this methodological difference when interpreting the combined σ0,lc.","section":"§3.1, cool LRDs"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me start with the bottom line. This is a useful, transparent paper that delivers the first systematic multi-epoch NIRSpec MSA constraints on H-alpha variability for 12 LRDs over rest-frame 1-3 months. The headline result - LRDs have intrinsic H-alpha rms <=4%, well below matched SDSS-RM quasars at 6.3%, and a flat white-noise structure function across timescales - is plausible and consistent with prior photometric work, but it rests on a normalization choice that may suppress exactly the variability the paper aims to measure.\n\nWhat is genuinely good: the sample is well-defined, the SDSS-RM comparison is luminosity-matched, the sigma0 estimator is a proper maximum-likelihood approach, they run a control sample of emission-line galaxies to check for systematics, and they repeatedly label their variability numbers as upper limits. The Balmer decrement analysis is thorough. The paper is honest about its caveats.\n\nThe soft spot is Section 3.1. Each LRD epoch spectrum is normalized by dividing the line profile by the fitted local continuum and rescaling to the epoch with the largest continuum. If the LRD continuum varies at the few-percent level and H-alpha responds to the ionizing continuum, this normalization removes exactly the common-mode signal they want to detect. Their only direct check that the continuum is constant is difference imaging from the first three epochs; aperture photometry on all six shows ~6% scatter, comparable to the quoted 4% H-alpha rms. So the 4% upper limit could be an underestimate. They also exclude NX7607, the one LRD with ~30% variability, from the headline estimate, which matters with 12 objects, and the white-noise pattern is a qualitative interpretation of results from different samples and methods. The stress-test concern on the normalization is fair; it is not a manufactured flaw.\n\nNone of this kills the paper. The measurement is genuinely new and the field needs these constraints. But the central physical conclusion - a different broad-line production mechanism in LRDs - depends on a normalization that is not yet fully validated. A serious referee should ask for difference imaging on all epochs, or a joint line-plus-continuum variability analysis that does not assume the continuum is constant. I'd send it to peer review; with that revision it could be a solid contribution. Anyone working on LRD demographics or high-z AGN variability will want to read it.","headline":"First systematic NIRSpec multi-epoch H-alpha variability sample for LRDs gives a plausible 4% upper limit, but the local-continuum normalization can suppress exactly the variability it measures, so the white-noise conclusion is not yet solid.","tokens_in":35356,"tokens_out":5713,"would_cite":true,"duration_ms":61066,"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":"Little Red Dots show H-alpha variability of at most about 4 percent on monthly timescales—far below the ~6 percent red-noise flicker of normal quasars—suggesting their broad-line light is produced differently.","keywords":["Little Red Dots","AGN variability","H-alpha emission","JWST NIRSpec spectroscopy","High-redshift galaxies","Supermassive black holes","Broad-line region","Balmer decrement"],"falsifier":"Take the same or a larger LRD sample and normalize every NIRSpec epoch using independent difference-imaging photometry for all epochs rather than the spectral continuum. If the resulting H-alpha rms variability exceeds about 6 percent on rest-frame timescales below 100 days, or if the structure function rises with timescale instead of staying flat, the white-noise, weak-variability conclusion fails.","tokens_in":34505,"feed_emoji":"🔭","tokens_out":9930,"duration_ms":106502,"temperature":0.7,"pith_summary":"This paper tries to establish that Little Red Dots—compact, red, high-redshift sources thought to harbor accreting supermassive black holes—show almost no H-alpha or rest-optical continuum variability on rest-frame timescales of about one to three months, and that the little variability they do show is consistent with white noise rather than the red-noise flicker of normal quasars. Using multi-epoch JWST spectroscopy of 17 LRDs and 14 blue broad-line AGNs, the authors measure an intrinsic H-alpha rms variability of at most about 4 percent for the LRD population, compared with about 6 percent for a luminosity-matched low-redshift AGN sample over the same timescales. They also find enhanced Balmer decrements (median H-alpha/H-beta around 7.6) in LRDs. If the claim holds, the broad-line emission in LRDs is probably not produced by the same reverberation-driven mechanism as in ordinary AGNs, which would reshape theoretical models of these objects and of early black-hole growth.","feed_headline":"Little Red Dots barely flicker: H-alpha rms about 4%","feed_subtitle":"Multi-epoch JWST spectra show flat, noise-like variability over months to decades, unlike normal quasars.","key_machinery":"The load-bearing tool is the maximum-likelihood intrinsic rms variability estimator sigma_0, applied to pairwise fractional H-alpha flux differences between epochs: the paper symmetrizes the difference distribution with negative duplicates and divides the resulting scatter by the square root of two to obtain a sample-wide light-curve rms, sigma_0,lc. This estimator separates true source variability from per-epoch measurement noise, allowing 12 LRDs to be compared statistically with 56 low-redshift SDSS-RM AGNs. The secondary machinery is the ensemble structure function built from SDSS-RM H-alpha light curves, which supplies the red-noise baseline, together with a local-continuum normalizatio","core_discovery":"The central claim is that, as a population, LRDs have intrinsic H-alpha flux variability of about 4 percent rms or less on rest-frame timescales under roughly 100 days, with rest-optical continuum variability below about 3 percent, while luminosity-matched low-redshift AGNs show about 6 percent monthly H-alpha variability with a red-noise structure function that grows with timescale. Combining the present multi-epoch NIRSpec measurements with published LRD variability measurements on roughly six-month to decade timescales, the paper argues that LRD variability is flat—white-noise-like—across all sampled timescales, in both H-alpha and continuum. The authors present this as evidence that the","pith_inferences":["If the flat variability pattern is real, the reported 4 percent H-alpha upper limit may itself be optimistic: normalizing each epoch to a constant local continuum would erase any line variability that tracks the continuum, so the true line variability could be even lower than measured.","The comparison sample is luminosity-matched but not matched in Eddington ratio or black-hole mass; if LRD variability follows the same anti-correlation with accretion rate seen in local AGNs, the observed suppression might reflect extreme super-Eddington accretion rather than a fundamentally different emission mechanism—an alternative the current data may not fully exclude.","A direct test would be to obtain difference-imaging photometry for every spectroscopic epoch and normalize each spectrum independently; if the LRD continuum does vary at the roughly 2 percent level, the white-noise conclusion would need revision.","The same multi-epoch NIRSpec data could be searched for correlated narrow-line variability, for example in [OIII], which would discriminate between scattered or reprocessed emission and collisionally excited line origins in the dense-gas picture."],"forward_implications":["If LRDs genuinely vary as white noise, standard reverberation-mapping campaigns will not detect correlated H-alpha responses on monthly cadences, and black-hole masses for LRDs cannot be derived the usual way.","The flat structure function means longer baselines do not accumulate variability signal the way they do for normal AGNs; detecting real LRD variability will require much larger samples or rare individual objects, not simply longer monitoring.","Models that predict normal AGN-like ionizing-flux flicker, and hence roughly 6 percent monthly broad-line variability, are disfavored for the bulk of the LRD population, while dense-envelope or super-Eddington models that damp short-term variability are favored.","The enhanced Balmer decrement combined with weak variability points to collisional excitation or radiative transfer in dense gas, rather than dust reddening, as the origin of LRD line ratios.","Rare variable LRDs such as NX7607 may be transitional objects or misclassified reddened AGNs, and identifying such objects is a path toward understanding diversity within the population."],"supporting_citations":[{"why":"provides the maximum-likelihood sigma_0 estimator used to derive intrinsic H-alpha rms variability from pairwise flux differences.","marker":"(Y. Shen et al. 2019)"},{"why":"supplies SDSS-RM H-alpha light curves used to construct the ensemble structure-function baseline for normal AGNs.","marker":"(Y. Shen et al. 2024)"},{"why":"provides DR16Q spectral measurements used for the low-redshift Balmer-decrement comparison.","marker":"(Q. Wu & Y. Shen 2022)"},{"why":"reports a deficit of significant flux-difference detections in LRDs on roughly six-month timescales, extending the flat variability pattern beyond the NEXUS baseline.","marker":"(Z. Liu et al. 2026)"},{"why":"shows local LRD analogs have low, flat continuum and H-alpha variability over decades, anchoring the long-timescale end of the white-noise pattern.","marker":"(C. J. Burke et al. 2026)"},{"why":"documents one LRD with significant short-term H-alpha variability and quantifies roughly 30 percent systematic uncertainties used to caveat variability detections.","marker":"(E. Lambrides et al. 2026)"},{"why":"reports year-scale broad-line variability in the lensed LRD A2744-QSO1, included in the combined timescale comparison.","marker":"(L. J. Furtak et al. 2025)"},{"why":"provides difference-imaging photometry for the first three epochs that justifies the non-varying continuum normalization assumption.","marker":"(Z. Stone et al. 2025)"},{"why":"defines the sample-selection and classification criteria that separate LRDs from blue broad-line AGNs in the NEXUS survey.","marker":"(M.-Y. Zhuang et al. 2025)"}],"fun_headline_variants":["LRDs are quiet: H-alpha rms only ~4%","Little Red Dots: white-noise variability, ~4% rms","LRD flicker is flat, not red-noise: H-alpha ~4%","Quiet red dots: H-alpha variability ~4% rms"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the LRD continuum is steady over the observed months, so dividing each epoch spectrum by its local continuum only corrects slit losses—an assumption supported by difference-imaging photometry from the first three epochs, since if the continuum actually varies the normalization would cancel the associated line variability and could create the flat low-amplitude pattern the paper reports.","fun_headline_variants_meta":{"raw":{"variants":["LRDs are quiet: H-alpha rms only ~4%","Little Red Dots: white-noise variability, ~4% rms","LRD flicker is flat, not red-noise: H-alpha ~4%","Quiet red dots: H-alpha variability ~4% rms"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001218,"raw_usage":{"total_tokens":4911,"prompt_tokens":869,"completion_tokens":4042,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":3963}},"tokens_in":613,"tokens_out":4042,"duration_ms":30055,"temperature":1.0,"reasoning_tokens":3963,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T23:36:27.210729+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same or a larger LRD sample and normalize every NIRSpec epoch using independent difference-imaging photometry for all epochs rather than the spectral continuum. If the resulting H-alpha rms variability exceeds about 6 percent on rest-frame timescales below 100 days, or if the structure function rises with timescale instead of staying flat, the white-noise, weak-variability conclusion fails.","supporting_citations":[{"cited_title":"J., Stone, Z., Shen, Y., & Jiang, Y.-F","cited_arxiv_id":null,"evidence_quote":"shows local LRD analogs have low, flat continuum and H-alpha variability over decades, anchoring the long-timescale end of the white-noise pattern."}],"review_version":1}