{"id":"e580d962-b14f-43df-a8d4-7ef6f1b9547c","arxiv_id":"2501.01310","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A systematic search of 12 years of Fermi-LAT data identifies 40 jetted AGN with linear or quadratic long-term gamma-ray flux trends, mostly with multiplicative oscillations.","lead":"Astronomers sifted 12 years of Fermi satellite data on 3,308 active galaxies and found 40 whose gamma-ray brightness steadily rose, fell, or curved over about a decade. This is the first systematic list of such long-term brightness trends, which may point to pairs of supermassive black holes or changes inside the jets.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The R2>=75% criterion is applied to the seasonal_decompose smoothed trend (period=40), not to the raw 28-day fluxes; the 40-source count therefore needs to be re-verified on the raw light curves.","rationale":"The reader's CONDITIONAL verdict is appropriate. The central claim requires that each of the 40 sources has a statistically meaningful long-term trend in the measured gamma-ray flux. The weakest link is the definition of 'significant' in Section 3.3: R2>=75% is evaluated on the seasonal-decompose trend, not on the original binned fluxes. This is not a purely cosmetic issue. For stochastic red-noise variability, which the paper itself discusses in Section 4.2, a 40-bin moving average smooths low-frequency fluctuations into a curve that can be fit by a line or parabola with high R2, while the raw points have much lower explained variance. The F-test for linear trends is performed on this same smoothed curve, so its p-values are not calibrated for the true residual noise. The validation simulations inherit this smoothed-trend definition: Section 4.4's FPDR measures how often the pipeline declares a trend in simulated no-trend light curves, but because 'declares a trend' means 'passes the smoothed-trend R2 criterion', a low FPDR only shows that the pipeline is self-consistent, not that the R2 threshold corresponds to a true trend. The paper is honest about several limitations, including the decreased detection rate under pink noise (49.2% for linear trends) and the flare sensitivity results, and the visual examples do suggest genuine trends in some sources. The correct response is not to reject the work but to require a raw-data refit and a multiple-testing correction before the 40-source catalog is treated as established. Hence the reader's CONDITIONAL verdict stands unchanged.","tokens_in":24689,"tokens_out":7718,"duration_ms":82201,"concrete_test":"Refit each of the 40 light curves with linear and quadratic models directly to the original 28-day binned fluxes (with upper limits replaced as in Section 3.1), computing R2 and F-test p-values on the raw data. Count how many of the 40 retain R2 >= 75% and p <= 0.01; if the count is materially below 40 or the raw-data R2 values are systematically far below the Table 2 values, the 40-source claim is inflated by the smoothing step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.3 computes R2 on the trend component returned by statsmodels seasonal_decompose (period=40, Multiplicative; the green line in Fig. 5), then fits a line or parabola to that smoothed curve. For a red-noise light curve, a 40-bin moving average can produce a smooth, monotonically curved trace that a line or quadratic fits with R2 >= 75% even though the raw flux is consistent with no trend. The R2 values reported in Table 2 therefore measure the fit to a smoothed surrogate, not the fraction of the source's observed variance explained by the trend, and the F-test on the smoothed fit uses residuals that are artificially smooth and autocorrelated. The validation in Section 4.4 applies the same pipeline to simulated light curves and so cannot validate this premise; it only shows that the pipeline is internally consistent. The paper's own FPDR, 0.47%, implies about 7 expected false positives among the 1492 sources, and no multiple-testing correction is applied, so even taking the pipeline at face value the count of 40 is not false-positive controlled. The visual examples (PG 1553+113, 3C 84, BL Lac) and the injection-recovery simulations indicate that some real trends exist, which is why this is not a rejection, but the exact list and the 40-object claim are not yet established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a systematic search for long-term (~10 yr) linear and quadratic flux trends in the gamma-ray light curves of jetted AGN. From 3308 sources in 4FGL-DR2, a variability-index cut and an upper-limit-fraction cut select 1492 sources with 28-day binned light curves spanning 2008-2020. The detection pipeline compares Lomb-Scargle periodograms with and without linear detrending (requiring a >20% shift of the dominant peak), extracts a smooth trend with statsmodels seasonal_decompose (period=40, multiplicative), fits a line or parabola to that extracted trend and requires R^2 >= 75% with an F-test (p <= 0.01) on the slope, and classifies accompanying oscillations as additive or multiplicative. The result is a catalogue of 40 sources (32 linear, 8 quadratic, ~80% with multiplicative oscillations), an estimated 0.47% false-positive detection rate, and extensive injection-recovery tests under white, pink, red, and bending-power-law noise and under injected flares. Physical interpretations in terms of SMBH binaries and jet-intrinsic processes are discussed.","tokens_in":24955,"tokens_out":12333,"duration_ms":114194,"significance":"If the catalogue is correct, this is a valuable first census of long-lived monotonic gamma-ray trends in AGN and a useful target list for periodicity searches. The strengths are the scale and reproducibility of the processing (standard public packages throughout), the injection-recovery experiments of Sections 4.2-4.3, and the transparent internal diagnostics: Section 4.4 states that about 7 of the 1492 sources are expected to pass the pipeline spuriously, and Section 4.3 openly reports that flaring contamination suppresses trend recovery almost completely. Against this, the headline claim of 40 objects rests on an R^2 criterion computed on a smoothed surrogate of each light curve rather than on the observed fluxes (Section 3.3, Figures 5 and A1), and the paper's own FPDR implies that several of the 40 are likely spurious under the pipeline as built. Some of the flagged sources (e.g., PG 1553+113, 3C 84, BL Lac) are convincing from the displayed light curves, so the direction of the result is plausible, but the exact list and the count of 40 are not yet established.","major_comments":[{"comment":"The R^2 selection is applied to the smoothed trend component returned by seasonal_decompose, not to the 28-day binned light curve. The Figure 5 caption states that the red line is 'the fit of the green line', where the green line is the trend extracted by seasonal_decompose (period=40, multiplicative); the R^2 values quoted in the panels and in Table 2 therefore quantify how well a line or parabola describes an already-smoothed, noise-reduced curve, not the fraction of observed flux variance explained by the trend. For a red-noise time series, a 40-bin moving-average trend can be smooth, extended, and monotonic-looking even when the raw flux has no sustained monotonic component, so the R^2>=75% filter can admit such sources and reject others whose raw scatter is large. The paper's own Table 3 is consistent with this reading: the R^2 stage passes 100% of simulated linear-trend light curves under every noise type, which would be implausible if R^2 were evaluated against the noisy simulated fluxes. The F-test on the slope (p<=0.01) is evaluated on the same smooth, autocorrelated residuals and does not carry its nominal significance. The 40-source list should be re-derived by fitting the linear/quadratic model directly to the 28-day fluxes, or at least complemented by a table of raw-data R^2 values for all 40 candidates.","section":"Section 3.3, Figures 5 and A1, Table 2"},{"comment":"The paper's own FPDR, 0.47%, implies about 7 expected false positives among the 1492 sources, and no multiple-testing or false-discovery control is applied anywhere in the workflow. Since the pipeline reports 40 detections, about 17% of the list is expected to be spurious by the paper's calibration; the abstract's statement that the analysis 'identified 40 jetted AGN that exhibit long-term trends' is therefore stronger than the analysis supports. The authors should either report a corrected significance or an expected number of false positives per source (e.g., from the FPDR simulations) so that the reliability of the individual entries in Table 1 can be assessed.","section":"Section 4.4"},{"comment":"The validation simulations inherit the smoothed-trend framework of Section 3.3, so they test internal consistency rather than the premise that the extracted trend represents the true astrophysical signal. Additionally, the slope ranges used to generate the simulated light curves in Section 4.2 are drawn from the detected values in Table 2, making the quoted recovery efficiencies conditional on the parameter ranges the search is meant to establish. Neither point is circular in the definition of the target (no result is defined through a fitted parameter), but both limit the evidential weight of the 'total detection' rates; re-running the recovery and FPDR simulations with the raw-data R^2 criterion would directly test the load-bearing assumption.","section":"Sections 4.2 and 4.4"},{"comment":"Upper-limit bins are replaced with the flux that maximizes the likelihood for that bin, a censoring treatment whose effect on fitted slopes is not modeled in the simulations of Sections 4.2-4.4. If such bins cluster in one part of the light curve, the substituted values can bias the estimated trend slope; the trend sample's low upper-limit fraction (median 1.8%, Figure 2) mitigates this risk but does not eliminate it. A robustness check recomputing the fits with upper limits either removed or replaced by the 95% upper-limit values would establish that the slopes and the membership of the 40-source list do not depend on this choice.","section":"Section 3.1 and Figure 1"}],"minor_comments":[{"comment":"Section 4.2 contains the typo 'liner-trend detection' (should be 'linear-trend'), and Section 4.1 refers to 'Table 110' where Table 1 is meant; the table captions in the text also render as 'T able 1' and 'T able 2' with an erroneous space.","section":"Section 4.2 and 4.1"},{"comment":"The printed uncertainties for AP Librae ('pm 4x105') and B2 1520+31 ('pm 2x104') appear to be missing the minus sign on the exponent (presumably 10^-5 and 10^-4); the same formatting issue affects several other entries where exponents are rendered without signs.","section":"Table 2"},{"comment":"The last panel is labeled 'PKS 2345 16' with a missing minus sign, and several panels show y-axis tick labels like '5 0 5 10' that suggest negative flux values without explanation in the text.","section":"Figure A1"},{"comment":"The sentence giving units has a parenthesis error ('are 10^-8 ph cm^-2 s^-1 days^-2), for...') and does not state the 10^-8 factor for the units of c, which would be helpful for readers using Table 2.","section":"Section 3.3"},{"comment":"The text reads 'Therefore, The remaining 8 sources' with an incorrect capital letter, and the paragraph beginning 'Analyzing the oscillatory behavior within this sample' is a sentence fragment.","section":"Section 4.1"},{"comment":"The abstract says the trends span approximately 10 years while the data span 12 years; because seasonal_decompose drops roughly 40 bins at the edges (noted in the Figure 5 caption), the epoch over which the trend is actually constrained should be stated explicitly.","section":"Abstract and Section 3.3"},{"comment":"The choice period=40 in seasonal_decompose (about 3.1 years) determines which timescales are assigned to the 'seasonal' component and which remain in the 'trend'; a brief justification of this choice or a robustness test with another period would strengthen the methodology.","section":"Section 3.3"}],"recommendation":"major_revision","confidential_remarks":"The main gap is fixable within the scope of a data-analysis paper: recompute R^2 and the F-test on the 28-day binned fluxes for the 40 candidates and for the simulations, and either apply a multiple-testing correction or present per-source false-positive probabilities. If the authors decline to do this, the 40-source claim should be softened to a candidate list. I see no citation or novelty problem, though the 'first known sample' wording should acknowledge the cited single-object trend detections. The paper's scope fits the journal well; conditional on the re-analysis, I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is the first systematic census of decade-long gamma-ray trends across 1,492 variable Fermi-LAT jetted AGN, and the additive/multiplicative oscillation classification is new. The 40-source catalog is a useful target list. But the headline count is not yet established: the R2 >= 75% criterion is applied to the seasonal_decompose trend line, not to the 28-day binned flux, and the 0.47% FPDR implies roughly 7 expected false positives with no multiple-testing correction. The paper's own simulations show the pipeline is internally consistent, but they do not test that premise because they use the same smoothed-trend framework.\n\nWhat is good: the sample is large, the pipeline is clearly described, and the authors are unusually honest about limitations. The flare-injection test, which shows detection rates collapsing when bright flares are present, is a real caveat stated up front. Known trend sources (PG 1553+113, 3C 84, BL Lac) all appear, which gives confidence that some of the 40 are real. The paper also cites prior individual-object detections and does not claim the trend phenomenon itself is new; the newness is the systematic search and the oscillation-type classification.\n\nWhere it is soft: the smoothed-surrogate R2 is the load-bearing choice for the 40 count, so it is more than cosmetic. The F-test p-values are computed on residuals that are artificially smooth and autocorrelated. There is no code or data release, so the table cannot be checked without rerunning the pipeline. Drawing simulation slope ranges from the detected values is a mild circularity for recovery efficiency, though not for the existence of the candidate list. These are all fixable in a revision.\n\nWho should read it: anyone working on AGN variability, periodicity searches, or SMBH binary candidates. It deserves a serious referee; the central idea is sound and the shortcomings are addressable. The revision should redo the fits on the raw binned fluxes, apply an FDR correction, and release the pipeline.","headline":"A genuinely useful first systematic census of long-term gamma-ray trends in jetted AGN, but the 40-source count rests on R-squared measured on a smoothed trend and needs re-verification on raw fluxes.","tokens_in":25589,"tokens_out":2233,"would_cite":true,"duration_ms":22689,"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 search of Fermi-LAT light curves identifies 40 jetted active galactic nuclei with significant decade-long gamma-ray flux trends—32 linear and 8 quadratic—and finds that in about 80% of them the variability amplitude grows…","keywords":["gamma-ray astronomy","active galactic nuclei","blazars","long-term trends","Fermi-LAT","light curves","quasi-periodic oscillations","variability"],"falsifier":"Take the 40 claimed trend sources and fit the same linear or quadratic models directly to the raw 28-day flux values, replacing upper limits by the paper's own likelihood-maximizing substitution or by a survival analysis, and recompute R² and the F-test p-value without any seasonal-decomposition smoothing; if a substantial fraction fail R² ≥ 0.75 on the raw data, the smoothed-trend criterion rather than an astrophysical trend would be doing the selection. A complementary check is to inject synthetic trends of known slope into real non-trending Fermi light curves and compare recovery rates when the fit is evaluated on raw versus smoothed data.","tokens_in":24437,"feed_emoji":"📈","tokens_out":5931,"duration_ms":52529,"temperature":0.7,"pith_summary":"The paper tries to establish that long-term monotonic trends in gamma-ray flux are not confined to a handful of famous blazars but are a measurable, systematic phenomenon: after searching 12 years of Fermi-LAT light curves of 3,308 jetted active galactic nuclei, it identifies 40 objects whose decade-long emission follows a significant linear or quadratic trend. If true, this gives a first statistical sample for studying the physical origin of such trends—candidates range from supermassive black hole binaries to jet-internal processes—and it warns that future searches for quasi-periodic oscillations must detrend or otherwise account for these slow changes. The paper also establishes a classification of oscillations into additive and multiplicative types, finding that roughly 80% of the trend sources have oscillation amplitudes that scale with the trend.","feed_headline":"40 jetted AGN show decade-long gamma-ray trends","feed_subtitle":"First systematic sample with 32 linear and 8 quadratic flux trends found in 12 years of Fermi-LAT data.","key_machinery":"The detector is a two-stage pipeline. First, a Lomb-Scargle periodogram run on both the original and linearly detrended light curve preselects sources whose dominant frequency shifts by more than 20%, flagging possible trends. Then seasonal decomposition with a 40-bin period and multiplicative model extracts a smooth trend line; a linear or quadratic fit must reach R² ≥ 0.75, with an F-test (p ≤ 0.01) excluding a constant model. The pipeline also classifies oscillation behavior by regressing the flux of high-flux peaks against time, labeling oscillations additive when the slope is consistent with zero and multiplicative otherwise.","core_discovery":"A systematic pipeline—Lomb-Scargle periodograms with and without detrending to flag frequency-domain changes, then seasonal decomposition to extract the smooth trend, then linear or quadratic fits with R² ≥ 0.75 and an F-test for nonzero slope—selects 40 of the 1,492 variable jetted AGN studied as showing long-term gamma-ray trends over roughly 10 years: 32 linear and 8 quadratic. Roughly 80% of these sources show multiplicative oscillations, where oscillation amplitude grows or shrinks with the trend, and the selected sources have very few upper limits (median 1.8%). The paper claims this is the first known sample of gamma-ray emitters with long-term trends of this nature and interprets the trends as possibly arising from supermassive-black-hole binary dynamics or from intrinsic jet phenomena such as magnetic reconnection or geometric Doppler-factor changes.","pith_inferences":["The pipeline's reliance on a fixed 40-bin seasonality and an R² ≥ 0.75 threshold may select trends that are smooth by construction; an independent check using raw-data fits or Bayesian model comparison could reveal whether the true incidence of long-term trends is higher or lower than 40 in 1,492.","The additive-versus-multiplicative classification, based on peak fluxes above 40–60% of maximum, could be extended to a full amplitude-trend correlation analysis using the entire light curve, which would test whether 'multiplicative' is a sharp dichotomy or a continuum.","If multiplicative oscillations are common, the physical interpretation shifts: the same mechanism that drives the trend also modulates the variability amplitude, a pattern naturally predicted by geometric Doppler-factor changes but less naturally by a simple accretion-rate trend.","A direct extension would apply the same pipeline to the next Fermi-LAT catalog release, and to optical and radio light curves of the same 40 objects, to test whether the gamma-ray trends persist and whether they are achromatic."],"forward_implications":["If correct, the 40 sources provide the first large sample for testing physical models of long-term gamma-ray trends, including supermassive-black-hole binary lump-driven accretion modulation and jet-internal mechanisms.","Roughly 80% of trend sources show multiplicative oscillations; future periodicity searches in these objects must account for a trend-dependent oscillation amplitude or risk masking genuine quasi-periodic oscillations.","The strong preference for linear (32) over quadratic (8) trends, and the low upper-limit fraction among selected sources, suggests trends are best seen in well-sampled light curves; real trends in sources with many upper limits may remain hidden.","The flare-injection simulations show that bright, long-duration flares can destroy trend detection, implying the 40-source sample is a lower bound and additional trend sources may be found once flaring epochs are modeled or removed."],"supporting_citations":[{"why":"Supplies the 4FGL-DR2 catalog that defines the 3,308-source sample and the variability-index values used in pre-selection.","marker":"Ballet et al. 2020"},{"why":"Provides the fermipy package used for Fermi-LAT binned analysis and the production of 28-day light curves.","marker":"Wood et al. 2017"},{"why":"Gives the test-statistic definition and significance thresholds adopted for source detection in the ROI fits.","marker":"Mattox et al. 1996"},{"why":"Establishes the prototypical periodic-behavior source PG 1553+113, which the paper uses to illustrate trend detection and to motivate the search.","marker":"Ackermann et al. 2015"},{"why":"Defines the upper-limit handling and detrending practices that the pipeline inherits for building the 1,492-source working sample.","marker":"Penil et al. 2020"},{"why":"Provides the R² thresholds (25%, 50%, 75% as weak, moderate, substantial) that justify the paper's R² ≥ 75% selection criterion.","marker":"Hair et al. 2011"},{"why":"Supplies the method for simulating light curves with matched power spectral density and probability density function used to estimate the false-positive detection rate.","marker":"Emmanoulopoulos et al. 2013"},{"why":"Presents the magnetic-reconnection model proposed as one physical origin for the observed long-term trends.","marker":"Giannios 2013"}],"fun_headline_variants":["40 jetted AGN show decade-long gamma-ray trends","40 AGN reveal long-term gamma-ray trends over 10 years","First systematic catalog of 40 AGN with flux trends","Decade-long trends in 40 AGN: linear and quadratic","Fermi-LAT detects 40 AGN with persistent gamma-ray trends"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole selection rests on the premise that the smooth trend extracted by the seasonal decomposition, and the R² computed from that smoothed line, faithfully represents the true astrophysical trend; the validation simulations use the same smoothing, so they never test whether a source that passes on the smoothed line would also pass on the raw 28-day light curve.","fun_headline_variants_meta":{"raw":{"variants":["40 jetted AGN show decade-long gamma-ray trends","40 AGN reveal long-term gamma-ray trends over 10 years","First systematic catalog of 40 AGN with flux trends","Decade-long trends in 40 AGN: linear and quadratic","Fermi-LAT detects 40 AGN with persistent gamma-ray trends"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000638,"raw_usage":{"total_tokens":2952,"prompt_tokens":968,"completion_tokens":1984,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":1895}},"tokens_in":584,"tokens_out":1984,"duration_ms":12738,"temperature":1.0,"reasoning_tokens":1895,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:29:43.529290+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the 40 claimed trend sources and fit the same linear or quadratic models directly to the raw 28-day flux values, replacing upper limits by the paper's own likelihood-maximizing substitution or by a survival analysis, and recompute R² and the F-test p-value without any seasonal-decomposition smoothing; if a substantial fraction fail R² ≥ 0.75 on the raw data, the smoothed-trend criterion rather than an astrophysical trend would be doing the selection. A complementary check is to inject synthetic trends of known slope into real non-trending Fermi light curves and compare recovery rates when the fit is evaluated on raw versus smoothed data.","supporting_citations":[],"review_version":1}