{"id":"e6419b31-e94b-450e-9398-a8bb39538f48","arxiv_id":"2501.15254","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Twelve faint quasar candidates show soft X-ray flux changes by factors over 10 across 20 years of XMM-Newton data, extending extreme-variability quasar searches to r ~ 22.","lead":"Using a random forest classifier on X-ray, optical, and infrared data, the authors identify 12 quasar candidates whose soft X-ray brightness changed by more than a factor of 10 over two decades of XMM-Newton observations. A generalist reader might care because the sample pushes studies of extreme quasar variability to fainter optical magnitudes and hints at how rare very large X-ray outbursts are.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"RASS non-detections cannot support the rarity claim: all 12 candidates have bright-state fluxes below the ~1e-13 RASS limit, so non-detection is expected even without variability.","rationale":"The paper's central claim has two parts: a sample of 12 highly variable quasar candidates, and a conclusion from RASS non-detections that >100x variability is extremely rare. The reader's weakest_assumption focuses on the first part, classifier contamination of faint, unconfirmed sources. I agree that is a real risk: the random forest is trained on SDSS/LAMOST spectroscopically identified sources and applied to faint sources with no spectra, and Section 6 explicitly says spectra are needed to confirm quasar nature. However, the second part is more load-bearing in a logical sense, because it fails even if every one of the 12 is a genuine quasar. The paper itself states RASS sensitivity is ~1e-13 erg/s/cm2, while Table 5 shows that all 12 bright states are below this value. Therefore the RASS non-detections are consistent with no variability at all and cannot be used to bound the frequency of >100x variables. This is not a matter of current consensus; it is an internal logical gap. A concrete computational test using RASS exposure maps and the published XMM fluxes would settle the point. I would keep the verdict CONDITIONAL because the sample itself is still a useful candidate list pending spectroscopy, but the rarity claim should be removed or substantially weakened. Since the reader already reached CONDITIONAL, the verdict is unchanged.","tokens_in":15838,"tokens_out":7912,"duration_ms":75099,"concrete_test":"For each of the 12 candidates, compute the RASS 0.1-2.4 keV count-rate upper limit from the ROSAT PSPC exposure maps at the source position, and convert the XMM 0.2-2 keV bright-state flux from Table 5 to the RASS band assuming the same power-law model (Gamma=1.7, NH=3e20). If the predicted RASS count rate is below the RASS bright-source detection threshold (~0.05 ct/s) for all 12 objects, then the non-detection probability is near unity regardless of variability, demonstrating that the paper's rarity conclusion is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing flaw is not only the absence of spectroscopic confirmation; it is that the RASS non-detections cannot support the headline inference of rarity. Section 5 states that none of the 12 candidates is detected by ROSAT and concludes that quasars with >100x variability are 'exceedingly rare.' But RASS has a flux sensitivity of ~1e-13 erg/s/cm2 (Voges et al. 2000), while Table 5 gives XMM 0.2-2 keV fluxes in units of 1e-14 erg/s/cm2. The brightest state in the table is 9.41e-14 (J122809.3+435358), and most objects never exceed ~5e-14. All 12 are therefore fainter than the RASS limit even in their bright states. A source with variability factor >100 would need a bright-state flux more than 100 times its faint-state flux; even at the selection floor of 5e-15 used in Section 4.2, that bright state would be >5e-13 and RASS-detectable. The selected candidates show no such bright states, so RASS non-detection is exactly what would be observed for non-variable sources. The inference would require a sample of quasars with XMM bright-state fluxes above the RASS limit, not a sample whose bright states are all below it. Thus the rarity claim is logically unsupported independent of classification accuracy.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript applies a random forest classifier to 4XMM-DR13 sources cross-matched with SDSS DR18 and AllWISE photometry, using a training sample of spectroscopically classified stars, galaxies, and quasars. After Gaia proper-motion filtering and exclusion of FIRST counterparts, the authors select 12 quasar candidates that changed their 0.2–2 keV flux by a factor of at least 10 between XMM-Newton epochs, extending to r-band magnitudes around 22. The paper further reports that none of the 12 candidates is detected in ROSAT and concludes that quasars exhibiting variations of more than two orders of magnitude are extremely rare.","tokens_in":16120,"tokens_out":6347,"duration_ms":54657,"significance":"The sample construction is careful: the training set is built from SDSS and LAMOST spectra, the classifier is validated with cross-validation and external catalogs (SIMBAD, NED), light curves and X-ray images are visually inspected, and a comparison sample of known quasars is selected with identical criteria. The public machine-readable table and GitHub repository are useful assets. If spectroscopically confirmed, the 12 candidates would extend the known population of extreme soft-X-ray variable quasars to the optically faint end. However, two load-bearing issues remain: the ROSAT-based rarity inference is not supported by the data presented, and the reliability of the classifier for the faintest, uncataloged sources is not quantified. The paper itself acknowledges in Section 6 that spectra are needed to confirm the quasar nature of the candidates.","major_comments":[{"comment":"The claim that ROSAT non-detections imply that quasars with variability factors of 100 or more are 'exceedingly rare' is not supported by the data. All 12 candidates have bright-state 0.2–2 keV fluxes below 1e-13 erg/s/cm2 (Table 5; the largest is 9.41e-14 for 4XMM J122809.3+435358), while the paper itself quotes the RASS sensitivity as a few times 10^-13 erg/s/cm2. Non-detection in RASS is therefore expected even if these objects are non-variable. Constraining the occurrence rate of >100x variability would require a sample whose bright-state fluxes are above the RASS limit, or a statistical model of RASS detection probability that accounts for the sensitivity and epoch. Please remove or substantially qualify this inference.","section":"Abstract; Section 5, Figures 5–6"},{"comment":"Table 5 reports the XMM-Newton and Chandra soft X-ray fluxes without uncertainties, and the text only states that candidates with 'large flux uncertainties' were rejected after visual inspection. Since the factor-of-10 variability claim is the central result, the reader needs the individual flux uncertainties (and ideally detection significances) for every epoch to verify that the flux ratios are statistically robust, especially for the faintest measurements near the 5e-15 erg/s/cm2 threshold.","section":"Section 4.2, Table 5"},{"comment":"The random forest is trained on spectroscopically confirmed sources that are systematically brighter in X-rays than the 12 candidates (Figure 4), and none of the 12 has a spectrum. The paper acknowledges that spectra are needed, but to support the claim that the sample extends the highly variable quasar population, it should provide a quantitative estimate of classifier performance in the flux/magnitude regime of the candidates—for example, flux-bin precision and recall on the held-out test set, or the random-forest probability P_QSO for each of the 12 objects, which is currently omitted from Table 5.","section":"Section 4.1, Figure 4, Section 6"}],"minor_comments":[{"comment":"The text refers to 'the relatively deeper observations in the 4XMM-DR3 catalog'; this should be 4XMM-DR13.","section":"Section 4.2, last paragraph"},{"comment":"The proper-motion cut is described as log(fPM0) ≤ −4 in the prose and as log(fPM0) < −4 in the following sentence; please harmonize the inequality.","section":"Section 4.1"},{"comment":"The sentence 'The optimal value is log 152 for max_features' is unclear; please specify whether this is log2(152) or another convention.","section":"Section 3.4"},{"comment":"The relation among the 100,183 XMM-WISE-SDSS sources, the 23,501 training sources, and the 76,682 sources actually classified by the random forest is implicit; a brief sentence stateing that only the unclassified remainder is subject to the classifier would improve clarity.","section":"Section 2.1 and Section 3"}],"recommendation":"major_revision","confidential_remarks":"This is a candidate-sample paper with a sound and reproducible selection pipeline, but the secondary rarity inference built on ROSAT non-detections is logically unsupported and should be revised. The lack of flux uncertainties in Table 5 is a straightforward fix. The paper fits the journal's scope; after the major points are addressed it could be acceptable, though the authors should consider whether the rarity claim can be salvaged at all with the current data."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper delivers what it says on the tin: a machine-learning classification of 4XMM-DR13 sources, and a hand-vetted sample of 12 quasar candidates with soft X-ray variability factors above 10, reaching r~22. That is genuinely new—previous work on extreme X-ray variable quasars was mostly limited to i<20.2. The sample construction is careful: the authors re-run a known Random Forest pipeline, apply Gaia proper-motion cuts to remove stars, cross-check against NED, and visually inspect light curves and X-ray images. They also ship the candidate table on GitHub, which is a plus.\n\nThe main caveat, which the authors themselves state, is that none of the 12 have spectroscopic confirmation. That makes them candidates, not quasars. This is a fair limitation and not fatal, but it should be flagged more prominently. The bigger problem is the ROSAT argument in Section 5. The authors note that none of the 12 are detected in RASS and then suggest that quasars with variability above a factor of 100 are 'exceedingly rare.' That inference is not supported. Table 5 gives XMM fluxes in units of 1e-14 erg/s/cm2; the brightest state among the 12 is 9.41e-14, and most are below 5e-14. The RASS sensitivity is about 1e-13 erg/s/cm2. So even in their bright states, all 12 are fainter than the RASS limit. Non-detection in RASS is exactly what you'd expect for steady sources at these fluxes. The rarity claim would require a sample with bright-state fluxes above the RASS boundary; this sample has none. The stress-test note is right, and the authors should either rephrase this as a simple statement about sensitivity or drop the quantitative rarity claim.\n\nThere are also internal inconsistencies that need cleaning: the abstract quotes 53,992 quasar candidates and 10,210 SIMBAD matches, while the body reports 52,486 and 8,410. That kind of mismatch should have been caught in proof. And Table 5 lists fluxes without uncertainties, which matters for a paper whose whole point is flux ratios.\n\nWho is this for? Astronomers working on X-ray variability and AGN demographics. As a candidate catalog, it's useful. As a claim about the rarity of order-of-magnitude variability, it's not ready. I'd send it back for a revision that fixes the numbers, adds flux uncertainties, and rewrites the ROSAT paragraph to say 'below the RASS limit' without drawing a statistical conclusion. The core sample deserves to be published, but the current framing overreaches.","headline":"Useful candidate sample, but the ROSAT-based rarity claim doesn't survive contact with the flux limits.","tokens_in":16706,"tokens_out":2970,"would_cite":false,"duration_ms":25835,"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":"This paper claims to identify 12 quasar candidates in 4XMM-DR13 whose soft X-ray flux changed by at least a factor of 10 between XMM-Newton epochs, extending extreme X-ray variability to optically faint quasars around r ~ 22, and argues…","keywords":["quasars","X-ray variability","random forest","4XMM-DR13","soft X-ray","Gaia proper motion","ROSAT","AGN"],"falsifier":"Obtain optical spectra of the 12 candidates: if a majority lack broad emission lines or other quasar signatures, the selection is contaminated. Alternatively, a targeted search of deeper all-sky X-ray data (e.g., eROSITA) for optically faint quasars with >100-fold soft X-ray variability would directly test the claimed rarity.","tokens_in":15638,"feed_emoji":"🔭","tokens_out":5951,"duration_ms":48471,"temperature":0.7,"pith_summary":"The paper claims that a random forest classifier trained on spectroscopically confirmed stars, galaxies, and quasars can pick out faint, previously unclassified quasar candidates from the XMM-Newton serendipitous catalog, and that this approach finds 12 candidates whose soft X-ray flux (0.2-2 keV) changed by at least a factor of 10 between XMM-Newton epochs. These candidates are optically faint, with SDSS i-band magnitudes around 22, extending the known population of extremely X-ray-variable quasars to the faint end. The paper further argues that because none of the 12 were detected in the ROSAT All-Sky Survey, quasars with variability amplitudes above about 100 are extremely rare. The importance is that such extreme variability, previously seen mostly in bright AGNs, may also occur in faint quasars that have been missed by shallower surveys.","feed_headline":"12 faint quasar candidates swing 10x in X-rays","feed_subtitle":"The finds reach i-band magnitude 22; the ROSAT silence hints >100-fold swings are rare.","key_machinery":"The random forest classifier uses 15 features: X-ray flux in 0.2-12 keV, four XMM hardness ratios (hr1-hr4), X-ray-to-optical flux ratio, SDSS r magnitude and colors (u-g, g-r, r-i, i-z), WISE magnitudes W1 and W2, and infrared colors (z-W1, W1-W2). SMOTE oversampling balances the training set of 23,501 spectroscopically classified sources (16,826 quasars, 3,800 stars, 2,875 galaxies). Stellar contaminants are removed by a Gaia proper-motion probability cut using the bivariate normal of pmra and pmdec, and extreme variability is defined by the ratio of soft-band fluxes between XMM epochs exceeding 10.","core_discovery":"The central discovery is a sample of 12 quasar candidates in the 4XMM-DR13 catalog that each changed their 0.2-2 keV flux by a factor of at least 10 between XMM-Newton observations spanning roughly 20 years. The selection pipeline first classifies 100,183 X-ray sources with optical and mid-infrared counterparts into quasar, galaxy, and star candidates using random forest; after a Gaia proper-motion cut to remove stars and a FIRST cut to remove radio-loud objects, it isolates sources with at least two clean XMM detections, a soft-band flux ratio greater than 10, and a minimum flux of 5e-15 erg $cm^{-2}$ $s^{-1}$, followed by visual inspection of light curves and images. The 12 survivors have SDSS i-band magnitudes out to about 22, making them fainter than previously known highly variable quasars. The paper also reports that none of the 12 are detected in ROSAT, and given ROSAT's flux limit this implies that quasars with variability factors above 100 are extremely rare.","pith_inferences":["If any of the 12 turn out to be flaring M dwarfs or X-ray binaries, the rarity claim would need to be revised downward; this is testable by spectroscopy.","The ROSAT non-detection may simply reflect that these quasars were in a low state during the 1990s; the true rate of >100-fold events could be higher than inferred if such events are short-lived.","Because the training set is dominated by bright quasars, the classifier's precision at the faint end is uncertain; future spectroscopic follow-up of a random subset of the 52,486 candidates would quantify the contamination rate."],"forward_implications":["The population of quasars with extreme soft X-ray variability (factor >=10) extends to i-band magnitude about 22, not just the bright AGNs previously catalogued.","The ROSAT non-detections imply that variability amplitudes above 100 are very rare among these faint quasars, placing a constraint on models of changing accretion or obscuration.","The random forest pipeline can be applied to future XMM-Newton data releases to enlarge the sample and refine the rarity estimate.","The 12 candidates are prime targets for spectroscopy and multiwavelength monitoring to distinguish between changing accretion rate and obscuration as the driver."],"supporting_citations":[{"why":"Supplies the 4XMM-DR13 catalog, its median soft X-ray flux, and the assumed spectral model for converting count rates to fluxes.","marker":"N. A. Webb et al. 2020"},{"why":"Demonstrates random forest classification of 4XMM sources and provides the matching radii and feature set used here.","marker":"Y. Zhang et al. 2021"},{"why":"Provides the Gaia proper-motion probability cut algorithm used to remove stellar contaminants from quasar candidates.","marker":"Y. Fu et al. 2021, 2024"},{"why":"Supplies the comparison sample of AGNs with soft X-ray flux drops by about a factor of 10 from ROSAT/XMM data.","marker":"D. Bi et al. 2015"},{"why":"Gives the Chandra-based upper limit on the occurrence rate of extreme X-ray variability in radio-quiet quasars, framing the rarity claim.","marker":"J. D. Timlin et al. 2020"},{"why":"Provides the ROSAT catalog and its flux sensitivity limit used to infer that >100-fold variability is extremely rare.","marker":"W. Voges et al. 2000"}],"fun_headline_variants":["12 faint quasar candidates show 10x X-ray swings","Machine learning finds 12 faint quasars varying 10x in X-rays","Extreme X-ray variability seen in 12 faint quasar candidates","Rare 10x X-ray changes in 12 faint quasar candidates","20-year X-ray variability of 10x in 12 faint quasars"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The random forest classifier was trained on relatively bright, spectroscopically confirmed quasars, and none of the 12 final candidates has a confirming spectrum; if several are actually stars or galaxies, the sample and the rarity inference collapse.","fun_headline_variants_meta":{"raw":{"variants":["12 faint quasar candidates show 10x X-ray swings","Machine learning finds 12 faint quasars varying 10x in X-rays","Extreme X-ray variability seen in 12 faint quasar candidates","Rare 10x X-ray changes in 12 faint quasar candidates","20-year X-ray variability of 10x in 12 faint quasars"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00122,"raw_usage":{"total_tokens":5083,"prompt_tokens":1077,"completion_tokens":4006,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":693,"completion_tokens_details":{"reasoning_tokens":3910}},"tokens_in":693,"tokens_out":4006,"duration_ms":28196,"temperature":1.0,"reasoning_tokens":3910,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:28:01.616036+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Obtain optical spectra of the 12 candidates: if a majority lack broad emission lines or other quasar signatures, the selection is contaminated. Alternatively, a targeted search of deeper all-sky X-ray data (e.g., eROSITA) for optically faint quasars with >100-fold soft X-ray variability would directly test the claimed rarity.","supporting_citations":[{"cited_title":"2021, MNRAS, 503, 5263","cited_arxiv_id":null,"evidence_quote":"Demonstrates random forest classification of 4XMM sources and provides the matching radii and feature set used here."},{"cited_title":"D., III, Brandt, W","cited_arxiv_id":null,"evidence_quote":"Gives the Chandra-based upper limit on the occurrence rate of extreme X-ray variability in radio-quiet quasars, framing the rarity claim."},{"cited_title":"2000, yCat, 9029, 0","cited_arxiv_id":null,"evidence_quote":"Provides the ROSAT catalog and its flux sensitivity limit used to infer that >100-fold variability is extremely rare."}],"review_version":1}