{"id":"a5c50903-440f-4022-9865-842714b5771c","arxiv_id":"2501.14026","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A WISE 22 micron selection yields 23 luminous obscured quasars at z~2, and their composite SED is redder and brighter in the infrared than earlier type 2 quasar templates.","lead":"This paper reports 23 bright, dust-obscured quasars at redshifts 1 to 3.5, found by selecting strong 22 micron infrared emission and confirmed with Gemini and Keck spectroscopy. It builds an average infrared-to-ultraviolet spectrum for this rare population, which can be used to find more obscured supermassive black holes at the peak of galaxy growth.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sample purity is internally inconsistent: 3/23 targets have broad (FWHM>4000 km/s) lines yet are counted as type 2, so the 96% type-2 success rate and the Fig. 8 composite template are not established for a pure type-2 sample.","rationale":"The paper has real strengths: 23/24 spectroscopic confirmations, public photometry, a no-AGN control fit (Section 4.4) that strongly disfavors a pure starburst explanation (SFR_IR ~ 3e5 M_sun/yr), and an internally consistent torus-dominated IR interpretation. The reader's weakest assumption - sample purity - is the same concern I identify: it is both an internal inconsistency (Section 2.5 vs. abstract and Table A2) and load-bearing, because the central claims of 'efficient type-2 selection' and a 'type-2 SED template' depend on the sample being obscured. Other issues (the 1:1 obscured fraction used for L_bol and M_BH is quoted without derivation, and the selection function is deferred to a companion paper) are real but secondary: they affect derived luminosities, not the SED shape or the existence of a torus component. I therefore keep the reader's CONDITIONAL verdict, sharpening the required condition: the classification must be applied consistently and the composite redone for a pure narrow-line subsample.","tokens_in":40915,"tokens_out":6450,"duration_ms":55741,"concrete_test":"Reclassify all 23 targets using the paper's own Section 2.5 criterion: any object with a permitted line FWHM >2000 km/s (H-alpha or other) is non-type-2. Then: (a) recompute the claimed 96% success rate and the fraction at z>2; (b) rebuild the Fig. 8 composite photometry and mean torus luminosity (eq. 3) excluding the 3-4 broad-line objects; (c) test whether the lambda L_lambda(30 micron) scaling and the rest-UV/optical mean shift by more than the published error bars. If the composite is statistically unchanged, the SED template is robust but the sample is a mixed type-1/type-2 population; if it changes, the template is biased and must be regenerated from the narrow-line subsample only.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (Abstract, Section 5) that mid-IR selection efficiently isolates type-2 QSOs, and that Fig. 8 is a new empirical type-2 template, requires the 23 confirmed targets to be obscured (no broad permitted lines). The paper contradicts this requirement: Section 2.2 reports six initial reddened type-1 candidates from GNIRS (broad FWHM >2000 km/s), and Section 2.5 reports 3/23 objects with broad lines FWHM>4000 km/s, calling them 'LRD-like'; Table A2 nonetheless groups J0221+0050 and J2258-0022 as 'Type 2' and only marks J0213+0024 as 'Redden type 1'. This is an internal classification inconsistency, not merely a terminological choice: it inflates the claimed 96% success rate, contaminates the composite photometry (Fig. 8) and mean torus luminosity with broad-line (type-1) objects, and weakens the advertised efficiency for finding true type-2 QSOs. The definitional shift between FWHM>2000 (reddened type 1) and FWHM>4000 (still called type 2) is never reconciled.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a pilot sample of 24 WISE 22 micron selected, optically faint or red type-2 quasar candidates in Stripe 82, with Gemini/GNIRS and Keck/LRIS spectroscopy confirming 23. The authors compile rest-frame 0.1-10 micron photometry, fit each SED with AGNFITTER, normalize to rest-frame 30 micron, and build a composite SED. They report a mean torus luminosity L_torus ~ 10^46.84 erg/s, discuss three possible origins for the rest-UV/optical light, compare with previous type-2 SEDs and JWST little red dots, and argue that mid-IR selection is an efficient way to find luminous type-2 quasars at z>2.","tokens_in":41261,"tokens_out":10349,"duration_ms":87778,"significance":"If correct, this is an important bright-end benchmark for obscured accretion at cosmic noon: 23 spectroscopically confirmed luminous obscured QSOs with a new composite template, plus a clean no-AGN control showing that explaining the W4 flux by star formation alone requires an implausible SFR > 3e5 solar masses per year. The paper also includes full photometry tables and a transparent discussion of SED degeneracies. The main caveat is sample purity: the target population as defined and analyzed is not a pure type-2 sample, so the headline success rate and the composite template need revision.","major_comments":[{"comment":"The paper characterizes all 23 confirmed targets as type 2 QSOs (title, abstract, Sec. 2.4), but Sec. 2.5 states that 3/23 have broad emission lines with FWHM > 4000 km/s and Fig. 4 presents two 'LRD-like' objects (J0221+0050, J2258-0022) with broad H-alpha. Table A2 nonetheless labels J0221+0050 and J2258-0022 as 'Type 2' and only J0213+0024 as 'Redden type 1'. Under the paper's own footnote-1 definition (FWHM > 2000 km/s and red continuum), broad H-alpha objects cannot be type 2 without further justification, and the definitional shift between FWHM > 2000 and FWHM > 4000 is never reconciled. This is load-bearing: the claimed ~96 per cent type-2 success rate, the composite photometry in Fig. 8, and the new 'type 2 QSO SED model' are all constructed from a mixed sample. The authors should reclassify the three broad-line objects, report the type-2 success rate for the pure sample, and re-derive the composite SED and mean torus luminosity with the reclassified sample, or explicitly present them as an obscured-plus-partially-obscured AGN template.","section":"Sec. 2.5 and Table A2"},{"comment":"The bolometric luminosities and Eddington-limit black hole masses quoted in the abstract and Sec. 3.5 assume an obscured fraction R = L_torus / L_bol = 0.5. The justification in Sec. 3.5 cites the paper's own number-density estimate in Sec. 2 (n_type2 ~ 0.55 deg^-2 versus n_type1 ~ 0.65 deg^-2) together with literature values. That estimate is not fully independent: it is derived from the same WISE W4 selection and torus-SED assumptions used in the SED fitting, so setting R from it and then computing L_bol = L_torus / R contains an element of circularity. The text says the R=1 case is extreme but does not give the resulting L_bol and M_BH ranges. Please report results for both R=0.5 and R=1, or otherwise demonstrate insensitivity, and either replace the in-house number counts with an external obscured-fraction constraint or state explicitly that the quoted L_bol and M_BH scale as R^-1.","section":"Sec. 3.5, Eqs. (4) and (7)"},{"comment":"The census of reddened type 1 objects is internally inconsistent: Sec. 2.2 reports six reddened type 1 QSO candidates from GNIRS (broad lines with FWHM > 2000 km/s), while Sec. 2.5 concludes that the reddened type 1 fraction is only 1/23, using a different definition (footnote 1) that requires no narrow permitted UV lines. Because the two LRD-like objects have broad H-alpha, they are the key to this discrepancy. The paper should tabulate, for all 23 targets, the observed line species, FWHMs, and available spectral coverage so the reader can verify the final classification. This is particularly important because J0213+0024, the sole reddened type 1, lacks an LRIS spectrum and therefore cannot be checked against the footnote-1 criterion.","section":"Sec. 2.2 and Sec. 2.5"}],"minor_comments":[{"comment":"The redshift range stated in the abstract (0.88-2.99) disagrees with z=3.49 for J0150+0056 in Table A2, and Sec. 2.5 calls J0112-0016 (z=2.99) the highest-redshift target; please harmonize the abstract, Fig. 2, and the text.","section":"Abstract and Tables A1/A2"},{"comment":"Because the primary SED runs set PRIOR_AGNfraction=True and turn_on_AGN=True, state explicitly that the hot-torus conclusion is robust to removing these priors, or provide a control run without the AGN prior; the no-AGN test in Sec. 4.4 addresses a different question about whether the IR can be explained without any AGN component.","section":"Sec. 3.4 and Sec. 4.4"},{"comment":"The formula for the error on the mean appears garbled in the text, with misplaced radical symbols; please reset the equation.","section":"Eq. (6)"},{"comment":"Column headers such as 'log like' and 'log z' are not defined, and the units of SFR_IR are not specified; please add a caption or explanatory note.","section":"Tables B1 and B2"},{"comment":"The Keck/LRIS spectra are described as being published in a later paper; for a paper whose central classification claim depends on these spectra, line-width measurements or reduced spectra for all 23 objects should be included as supplementary material, even if the full spectral analysis is deferred.","section":"Data Availability and Sec. 2.5"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about sample purity is real and central; I would not accept the paper in its present form because Section 2.5 undermines the title and abstract. After reclassification and recomputation, the remaining science (torus-dominated IR, composite SED, LRD comparison) is likely defensible. I also note that the selection and first spectra appeared in Ishikawa et al. (2023), so the incremental novelty of the SED analysis should be clarified in the introduction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nRead the Stripe82 type 2 paper. The core product is a rest-frame 0.1–10 μm composite SED of 23 luminous mid-IR-selected quasars at z~2, built by scaling photometry to the fitted 30 μm luminosity. That composite and the LRD comparison are genuinely new, and the no-AGN control fit is good evidence: without the torus, the IR demands SFR ~3e5 Msun/yr, which is absurd. The torus-dominated conclusion holds up for the sample as a whole.\n\nThe soft spot is sample purity, and it is real. The paper calls all 23 confirmed targets type 2, but Section 2.5 reports 3/23 with broad lines FWHM>4000 km/s (J0221+0050 and J2258-0022 are LRD-like, and J0213+0024 is classified reddened type 1). Table A2 groups the first two as Type 2. That is internally inconsistent with the standard type 2 definition used throughout. It inflates the 96% success rate and means the Fig 8 composite is not a pure type 2 template; it is a composite of obscured and partially obscured/broad-line AGN. This is addressable: relabel and recompute the composite with and without the broad-line objects. If the torus-dominated shape survives (likely), the main astrophysical point stands, but the efficiency claim for finding true type 2s needs modifying.\n\nOther issues are minor. The abstract redshift range (0.88–2.99) disagrees with the full text and Table 1, which includes one object at z=3.49. The obscured fraction R=0.5 is asserted from number densities not derived in this paper, so the L_bol and M_BH numbers carry an unquantified systematic. Fine for orientation, not for precision.\n\nWho should read it: people working on obscured AGN demographics, SED templates, and the LRD connection. It deserves a serious referee—the data work and empirical anchor are solid, and the classification flaw is fixable.\n\nRecommendation: engage, but push for a clean split of the sample by emission-line width before the composite is used as a pure type 2 template.","headline":"A useful z~2 composite SED for luminous mid-IR-selected obscured quasars, with a real sample-purity caveat: the '23 type 2' count includes objects with broad lines.","tokens_in":41806,"tokens_out":2870,"would_cite":true,"duration_ms":25904,"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":"A 22-micron colour cut isolates luminous type 2 quasars at cosmic noon with 96 percent spectroscopic success.","keywords":["type 2 quasars","obscured AGN","mid-infrared selection","WISE","spectral energy distribution","dust torus","little red dots","Stripe 82"],"falsifier":"Observe the H-alpha line profiles of a larger W4-selected sample: the paper predicts that more than 90 percent will be narrow-line type 2 objects, so a broad-line fraction above roughly 25 percent in about 50 candidates would falsify the claim that this mid-IR cut isolates type 2 quasars specifically.","tokens_in":40726,"feed_emoji":"🔭","tokens_out":6237,"duration_ms":52651,"temperature":0.7,"pith_summary":"The paper reports a pilot survey that uses the WISE 22 μm band, combined with an optically faint or very red colour cut, to hunt for luminous type 2 quasars at z > 2 in SDSS Stripe 82. Spectroscopic follow-up confirms 23 of 24 candidates as quasars at z = 0.88–3.49, with 12 above z > 2, a roughly 96 percent success rate that is much higher than earlier Spitzer-based selections. From multi-band photometry the authors build a composite rest-frame 0.1–10 μm spectral energy distribution and show that the infrared emission is dominated by a hot dust torus with a mean torus luminosity of about $10^{46}$.84 erg/s. They argue that the composite SED is a new empirical template for this bright, obscured population, and that its shape below 1 μm resembles JWST 'little red dots', suggesting those faint high-redshift objects could be lower-luminosity analogues of these quasars.","feed_headline":"22-micron cut finds 23 luminous obscured quasars","feed_subtitle":"Red WISE selection confirms them at 96 percent and ties them to JWST little red dots.","key_machinery":"The machinery is the combination of a 22-micron (W4) flux and colour cut with Bayesian SED fitting that decomposes each object's photometry into a hot dust torus, an accretion disk, stellar emission, and cold dust. The torus component is treated as the robust fit and used to normalize every object to a common rest-frame 30 μm luminosity; this normalization builds the composite rest-frame 0.1–10 μm SED, with the UV/optical decomposition left degenerate among scattered light, starlight, and a reddened disk.","core_discovery":"The central discovery claim is that a simple mid-infrared selection — W4 flux > 5 mJy, i.e. 12.62 < W4 < 14.62 AB mag, plus r > 23 or r − W4 > 8.38 — isolates the most luminous obscured (type 2) quasars at cosmic noon with high efficiency. On the authors' reading, the 23 confirmed objects all have AGN-dominated SEDs: the hot dust torus supplies essentially all the rest-frame >3 μm light with L_torus = $10^{46}$.84 erg/s, while the rest-frame UV/optical light is degenerate among scattered light, host starlight, and a reddened accretion disk. The paper therefore positions the resulting composite photometry as a template for finding more type 2 quasars at higher redshift, and as evidence that the bright-end obscured:unobscured ratio near z ~ 2 is roughly 1:1.","pith_inferences":["Editorial inference: because three of the 23 confirmed objects have broad lines (FWHM > 4000 km/s) and one is classed as a reddened type 1, a cleaner type 2 template would exclude these; the published composite may blend obscured and partially obscured objects, and re-fitting without them would test how much the red IR-optical color depends on the pure type 2 subsample.","Editorial inference: the scattered-light scenario (0.5–1 percent scattering) implies that these quasars should show broad scattered lines in polarized rest-UV light; spectropolarimetry of the brightest targets would test whether scattering or a reddened disk dominates the UV.","Editorial inference: applying the same W4 + color selection in larger area surveys and measuring the type 2 luminosity function at z = 2–4 would directly connect this pilot to the JWST little-red-dot counts and settle whether the obscured fraction rises toward high redshift."],"forward_implications":["If the selection is as efficient as reported, extending it to wider or deeper W4 coverage should multiply the known luminous z > 2 type 2 population, which currently numbers only a few dozen.","The composite SED, being redder and about 21 times brighter in torus emission than older low-redshift type 2 templates, should improve photometric redshift and bolometric luminosity estimates for obscured quasars found in future wide-area surveys.","A roughly 1:1 bright-end type 2:type 1 ratio at z ~ 2 would mean the missing luminous obscured population is not large at these luminosities, constraining the obscured fraction.","If little red dots are fainter higher-redshift analogues, then hot dust torus emission in LRDs cannot be ruled out, and SED-based LRD classification must allow for a torus component."],"supporting_citations":[{"why":"Supplies the WISE-based mid-IR selection method and initial bolometric luminosity estimates for these targets.","marker":"Ishikawa et al. 2023"},{"why":"Provides the hot dust torus templates used to fit the IR emission and derive L_torus.","marker":"Stalevski et al. 2016"},{"why":"Provides the Bayesian SED fitting machinery that decomposes the photometry into AGN and galaxy components.","marker":"Calistro Rivera et al. 2016"},{"why":"Supplies the stellar population templates used to test host galaxy light in the SED fits.","marker":"Bruzual & Charlot 2003"},{"why":"Provides the accretion disk big-blue-bump template used to model reddened and scattered UV/optical light.","marker":"Temple et al. 2021"},{"why":"Supplies the low-redshift type 2 AGN composite SED that this paper compares against and finds redder.","marker":"Hickox et al. 2017"},{"why":"Provides the earlier Spitzer-selected type 2 quasar sample whose low z > 2 fraction motivates the new selection.","marker":"Lacy et al. 2013"},{"why":"Provides the JWST little-red-dot sample whose composite photometry is compared to the type 2 QSO SED.","marker":"Akins et al. 2024"},{"why":"Supplies the Hot DOG SED used as a comparison point for the bright torus component.","marker":"Fan et al. 2016"}],"fun_headline_variants":["IR selection uncovers 23 hidden quasars at cosmic noon","WISE 22-micron finds buried quasars, echoing JWST red dots","Luminous type-2 quasars found via dust glow, not starlight","Obscured quasar census at z~2: 23 new from mid-IR","Mid-IR trick reveals 23 quasars hidden behind dust"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The efficiency claim depends on the classification of all 23 confirmed targets as type 2 quasars, but three show broad lines with FWHM above 4000 km/s and one is called a reddened type 1 QSO; if those are genuinely broad-line objects, the sample is a mix of obscured and partially obscured AGNs rather than a pure type 2 sample.","fun_headline_variants_meta":{"raw":{"variants":["IR selection uncovers 23 hidden quasars at cosmic noon","WISE 22-micron finds buried quasars, echoing JWST red dots","Luminous type-2 quasars found via dust glow, not starlight","Obscured quasar census at z~2: 23 new from mid-IR","Mid-IR trick reveals 23 quasars hidden behind dust"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000857,"raw_usage":{"total_tokens":3885,"prompt_tokens":1274,"completion_tokens":2611,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":890,"completion_tokens_details":{"reasoning_tokens":2512}},"tokens_in":890,"tokens_out":2611,"duration_ms":15528,"temperature":1.0,"reasoning_tokens":2512,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:29:04.568743+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Observe the H-alpha line profiles of a larger W4-selected sample: the paper predicts that more than 90 percent will be narrow-line type 2 objects, so a broad-line fraction above roughly 25 percent in about 50 candidates would falsify the claim that this mid-IR cut isolates type 2 quasars specifically.","supporting_citations":[],"review_version":1}