{"id":"fa5e73fd-b383-4e71-8699-c840f2820584","arxiv_id":"2506.03564","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"No repeat bursts were detected from 36 once-seen fast radio bursts in FAST follow-up, placing upper limits of about 10^-2.6 to 10^-0.22 repeats per hour.","lead":"Astronomers used the world's largest radio telescope to watch 36 fast radio bursts that had each flashed only once, and none flashed again during follow-up. The authors then calculated some of the tightest limits yet on how often such sources can repeat.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Rate limits hinge on extrapolating a Euclidean N(>S) ~ S^-1.5 law ~100x below the initial-burst threshold (Eq. 6); a flatter repeat-burst fluence law would make the claimed bounds optimistic by 3-10x, and the abstract's 'upper limits' are central estimates or medians, not confidence bounds.","rationale":"The observational core is credible: PRESTO-based single-pulse search at S/N > 7, visual inspection to S/N ~ 4, public FAST data (PT2021_0076), and the non-detection claim is standard and reproducible in principle. Independent support exists: one ML-selected target, FRB 20190110C, was confirmed as a repeater during the writing, and the rate derived for it is consistent with the quoted upper limits, which validates the general calibration of the method. The reader's CONDITIONAL verdict is therefore on the conversion step, not on the data. The load-bearing step is Eq. (6): it converts each telescope exposure to a common 1-Jy exposure through (S/S_0)^-1.5, and because FAST's threshold is about 100x deeper than CHIME's, the 10-minute FAST exposures are amplified by ~866x and typically dominate the denominator even after the P_acc coverage penalty. The headline limits therefore inherit the Euclidean slope assumption, extended two decades beyond the measured regime and to a burst population whose fluence law is unmeasured. The paper itself flags this premise in the Fig. 5 caption, so it is a known condition rather than a hidden error; a CONDITIONAL verdict asking for the alpha-dependence to be shown is the right response. I disagree with the reader on one technical point. The reader states that 'repeat bursts systematically fainter' would make the scaled upper limits optimistic. Fainter subsequent bursts imply a steeper cumulative fluence law, which increases FAST's deep-sensitivity leverage and would make the quoted limits conservative. The scenario that makes them optimistic is a flatter law, with subsequent bursts comparable to the initial burst. Either way the assumption is the hinge, and it should be quantified. A second, compounding issue the reader noted in passing is the labeling of the headline numbers. For Poisson, Table 3 reports the ML estimate with the 90% interval as error bars (the abstract quotes the central values); for Weibull, the paper explicitly reports the posterior median as the 'upper limit.' Proper 90% bounds are about 4x higher (Poisson) and 5-15x higher (Weibull 95th percentile). Fixing this is a reporting requirement independent of any model check. Since the non-detection itself is robust and the concern is bounded, I do not move the verdict: CONDITIONAL remains appropriate, with conditions (1) re-quote limits as proper confidence bounds and (2) quantify the dependence on the assumed fluence slope and repeat-burst properties.","tokens_in":22786,"tokens_out":38803,"duration_ms":404467,"concrete_test":"Recompute Eq. (6) and the Weibull posterior for all 36 sources with the fluence slope alpha = 1.0, 1.4, 1.5, and 2.0, and re-quote the limits as 90% confidence bounds (Kraft et al. upper limit for Poisson; 95th posterior percentile for Weibull) instead of central estimates or medians. Two checks settle the concern: (a) Does the most stringent source, FRB 20110523A (quoted 2.5e-3 hr^-1), stay below 1e-2 hr^-1 at alpha = 1.0? (b) Do the abstract's quoted ranges remain within a factor of about 4 of the 90% bounds? A complementary calibration: measure the first-to-subsequent-burst fluence slope of the CHIME/FRB Catalog 1 repeater sample and insert that slope into Eq. (6); if it differs from -1.5 by more than about 0.2, the headline numbers must be re-quoted as alpha-dependent.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The quantitative claim, upper limits of about 10^-2.6 to 10^-0.22 hr^-1 (Poisson) and 10^-2.3 to 10^-0.25 hr^-1 (Weibull), is set for most sources not by the raw 10-minute FAST exposures but by the sensitivity-scaling factor in Eq. (6). With S_F ~ 0.011 Jy, (S_F/S_0)^-1.5 ~ 866, so the FAST term P_acc*T_F*(S_F/S_0)^-1.5 contributes roughly 15 hr of equivalent 1-Jy exposure per source even after the P_acc ~ 0.1 coverage penalty, typically dominating the prior CHIME/GBT term. The Fig. 5 caption states the key premise: 'We assume the properties of subsequent bursts are consistent with those of the initial burst.' The slope -1.5 is supported for the field population (CHIME alpha = -1.4 +/- 0.11), but it is extrapolated about two decades in fluence and applied to the repeating-burst population of these specific sources, for which the fluence law is unmeasured. The bias direction deserves precise statement. 'Repeat bursts systematically fainter than the initial burst' corresponds to a steeper cumulative law, which would give FAST's deep threshold more leverage and make the quoted limits conservative; the dangerous case is a flatter law, i.e., subsequent bursts comparable to the initial one. The most stringent endpoint is fragile: FRB 20110523A (S_P = 0.009 Jy; quoted 2.5e-3 hr^-1) rises to about 2.8e-2 hr^-1 if alpha = 1.0 is used in Eq. (6) instead of 1.5, moving the claimed low end of the range by over an order of magnitude. Compounding this, the abstract's numbers are not confidence bounds. The Poisson entries in Table 3 are ML estimates N_bursts/T_eff, with the Kraft et al. 90% upper bounds about 3.9x higher; for the Weibull model the paper states that 'the 50th percentile of the resulting r distribution is reported as the upper limit,' while the 95th percentiles are 5-15x higher. Both layers push the headline limits in the same optimistic direction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports FAST follow-up observations of 36 FRBs previously classified as non-repeating, selected using empirical criteria (Hashimoto et al. 2020a) and a UMAP-based machine-learning scheme (Chen et al. 2022). Each source was observed for 10 minutes; no bursts with S/N > 7 were detected, yielding a typical 7-sigma fluence limit of about 0.013 Jy ms. Combining these non-detections with prior CHIME, GBT, ASKAP, UTMOST, and Parkes observations, the authors derive upper limits on the repetition rate under Poisson and Weibull models, claiming ranges of roughly 10^-2.6 to 10^-0.22 hr^-1 (Poisson) and 10^-2.3 to 10^-0.25 hr^-1 (Weibull). One target, FRB 20190110C, was independently confirmed as a repeater during the preparation of the manuscript, and its measured rate is shown to be consistent with the derived limits.","tokens_in":23128,"tokens_out":10717,"duration_ms":110592,"significance":"If the non-detections are robust, the paper provides one of the largest samples of deep follow-up constraints on apparently non-repeating FRBs, with per-source sensitivity limits computed using the standard radiometer equation and a careful treatment of positional coverage through P_acc. The work also usefully validates the machine-learning classification scheme through the independent confirmation of FRB 20190110C, and the data are public. However, the quantitative repetition-rate limits are conditional on an untested assumption about the fluence distribution of repeat bursts, and the quoted 'upper limits' are posterior medians rather than confidence upper limits, so the headline numbers require reinterpretation.","major_comments":[{"comment":"The values labeled 'upper limits' are the 50th percentiles of the posterior distributions, as explicitly stated for the Weibull case ('the 50th percentile of the resulting r distribution is reported as the upper limit'). This is not a confidence upper limit; the 90% upper bounds in the same table are typically several times larger (e.g., FRB 20190129A has a median of 0.048 hr^-1 and a 90% upper bound of 0.188 hr^-1). The abstract's stated ranges are therefore central estimates, and the claim of 'one of the most stringent upper limits' is overstated unless genuine 90% or 95% percentile upper limits are reported.","section":"Section 5.2.2, Table 3, Abstract"},{"comment":"The sensitivity scaling assumes N(>S) proportional to S^-1.5 for repeat bursts from these specific sources, extrapolated about two decades below the original detection sensitivity. This slope is measured for the field population, not for the repeating-burst population of these candidates. Since the FAST term P_acc T_F (S_F/S0)^-1.5 dominates the denominator for most sources, a flatter true slope (e.g., alpha = 1) would make the quoted limits less stringent by up to an order of magnitude; as a concrete example, the limit for FRB 20110523A rises from about 2.5e-3 hr^-1 to roughly 2.8e-2 hr^-1. Please present the limits as conditional on this assumption and include a robustness test varying alpha.","section":"Eq. (6) and Fig. 5 caption"},{"comment":"As written, Eq. (6) with N_bursts = 0 (the observed number of detections) yields r_scaled,joint = 0, yet Table 3 reports non-zero upper limits for all sources. The manuscript does not state what value of N_bursts (or which posterior quantile) is inserted into Eq. (6), making the central computation irreproducible. Please clarify the statistical procedure (e.g., drawing N_bursts from the Kraft et al. 1991 posterior) and distinguish it from the nominal N_bursts = 0.","section":"Section 5.2.1, Eq. (6)"}],"minor_comments":[{"comment":"The word 'frquency' should be 'frequency'.","section":"Section 4, paragraph 2"},{"comment":"The Weibull rate range is given as 'about 10^-2.3 to 10^0.25 hr^-1', but the abstract and Table 3 imply the upper end is 10^-0.25 hr^-1; please correct the sign.","section":"Section 5.2.2, final paragraph"},{"comment":"The sentence 'These are indicated by the grey, blue, and red triangles in Figs. 5 and 6. These are indicated by the blue and red triangles...' is duplicated; remove the repetition.","section":"Table 3, header"},{"comment":"Several typos should be corrected: 'observec' (Section 2.1), 'implicit constrants' and 'Neverthless' (Section 6.1), 'repeaterss' (Section 6.1), and 'smaple' (Section 7).","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the statistical labeling: reporting the 50th percentile of the posterior as an 'upper limit' is unconventional, and the abstract's numbers are not conservative. I would urge the editor to require the authors to report actual confidence upper limits (e.g., 90th or 95th percentile) or to clearly label the values as median rates. The dependence of the results on the assumed burst fluence slope (alpha) also needs a robustness test, since the FAST term in Eq. (6) dominates for most sources."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:2506.03564. The observational core is solid: the authors followed up 36 non-repeating FRB candidates with FAST, 10 minutes each, and found nothing above S/N 7. The search is documented carefully, the sensitivity limits are per-source, and the data are public. That is a useful dataset—the largest such sample to date, about five times the size of the Arecibo sample in Good et al. (2023). The P_acc correction for FAST's partial coverage of the positional uncertainty region is a legitimate improvement to the scaled-rate formalism. And the fact that one ML-selected source, FRB 20190110C, later turned out to be a CHIME-confirmed repeater is a nice external validation of the Chen et al. (2022) classification.\n\nNow the soft spots, in proportion. The headline repetition-rate limits are not as firm as the abstract implies. The Poisson numbers in Table 3 are maximum-likelihood estimates, with the Kraft 90% bounds roughly a factor of 4 higher; the Weibull numbers are the 50th percentiles of the posterior, which the paper itself calls 'upper limits.' Calling a median an upper limit is misleading, and anyone quoting the abstract's 10^-2.6–10^-0.22 range as a rigorous bound would be overstating the result. Second, Eq. (6) scales FAST's sensitivity by (S_F/S_0)^-1.5, pushing a fluence law measured for the CHIME population about two decades below the initial-burst threshold. If repeat bursts from these sources follow a flatter cumulative fluence distribution, the limits become optimistic by a factor of a few to ten. The assumption is stated—'consistent with the initial burst'—but its effect on the conclusions is not explored. A sensitivity scan in alpha would fix this.\n\nNone of this undermines the central empirical result: no repeats were seen, and the sample is large. It does mean the paper's advertised upper limits should be read as model-dependent estimates with wide error bars. The paper deserves a serious referee; a journal should send it out, asking for a relabeling of the Weibull/Poisson quantities, a sensitivity analysis on the flux-scaling index, and a sentence or two on why 25 of 69 ML targets were observed. It would be a solid contribution to the FRB repetition-rate literature once those items are addressed.","headline":"A clean and useful non-detection campaign whose headline rate limits are model-dependent and loosely labeled.","tokens_in":23908,"tokens_out":2931,"would_cite":true,"duration_ms":32054,"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":"Deep FAST follow-up of 36 apparently non-repeating FRBs finds no repeat bursts, placing the most stringent upper limits yet on their repetition rates.","keywords":["fast radio bursts","FRB repetition rates","non-repeating FRBs","FAST telescope","Poisson process","Weibull distribution","UMAP classification","upper limits"],"falsifier":"Detecting a burst above S/N 7 from any of the 36 sources in another 10-minute FAST exposure of comparable sensitivity would directly contradict the core non-detection result, as would measuring the fluence distribution of known repeater bursts and finding a slope steeper than $-1.5$.","tokens_in":22520,"feed_emoji":"📡","tokens_out":5601,"duration_ms":61005,"temperature":0.7,"pith_summary":"This paper asks whether FRBs currently classified as non-repeating are truly single bursts or simply repeaters whose later bursts have been missed. The authors selected 36 likely misclassified candidates using empirical luminosity-duration relations and an unsupervised machine-learning method, then observed each for 10 minutes with FAST. No burst above a signal-to-noise ratio of 7 was detected, corresponding to a typical 7-$\\sigma$ fluence limit of about 0.013 Jy ms. Combining these non-detections with prior observations yields upper limits on repetition rates of about $10^{-2.6}$ to $10^{-0.22}$ per hour under a Poisson process and about $10^{-2.3}$ to $10^{-0.25}$ per hour under a Weibull process. If the limits hold, many apparent non-repeaters are not rapidly active repeaters, sharpening the observational distinction between the two FRB classes.","feed_headline":"36 apparently non-repeating FRBs stay silent in FAST follow-up","feed_subtitle":"Ten-minute FAST exposures set repetition-rate limits below once per 20 hours, over five times more sources than earlier limits.","key_machinery":"The central object is the scaled joint repetition rate, $r_{\\rm scaled,joint} = N_{\\rm bursts} / [T_P(S_P/S_0)^{-1.5} + P_{\\rm acc} T_F(S_F/S_0)^{-1.5}]$, which converts non-detections from telescopes with different sensitivities into a common per-hour rate. It scales exposure times by the Euclidean source-count slope $\\alpha = -1.5$ and by $P_{\\rm acc}$, the probability that FAST's 19 beams cover the source's positional uncertainty region. A parallel Weibull analysis uses the scaled interval $\\Delta_{\\rm scaled} = \\Delta(S/S_0)^{-1.5}P_{\\rm acc}$ inside the Bayesian formalism for burst-interval distributions.","core_discovery":"The paper establishes that 36 FRBs selected as likely hidden repeaters do not produce a detectable burst during 10-minute FAST exposures, and that their absence of bursts can be converted into the tightest joint upper limits on FRB repetition rates reported so far. Using a sensitivity-scaled repetition-rate formula, the authors find typical upper limits near $10^{-1.3}$ hr$^{-1}$, about a factor of three tighter than prior Arecibo-based limits and based on a sample five times larger. One machine-learning-selected source, FRB 20190110C, was independently confirmed by CHIME as a repeater during this work, supporting the preselection strategy. The paper also reports that, under a Weibull model, the burst rate shape parameter $k$ is poorly constrained but tends below 1 for many sources, hinting at possible temporal clustering in FRB activity.","pith_inferences":["If repeat bursts are systematically fainter than the first detected burst, the assumed $N(>S)\\propto S^{-1.5}$ fluence scaling makes the reported upper limits optimistic; a measured fluence distribution of known repeaters' bursts would test this directly.","Because FAST's 19 beams cover only about 10% of the positional uncertainty for these sources, some missed repeats could be due to positional mismatch rather than true quiescence; wider-field monitoring would separate these possibilities.","The same empirical-plus-machine-learning target selection could be applied to the expanded CHIME catalogue to grow the monitored sample and push typical repetition-rate limits below $10^{-2}$ hr$^{-1}$.","Simultaneous multi-telescope follow-up would calibrate the sensitivity-scaling assumption, since a burst detected by two telescopes with different sensitivities would directly measure the relevant fluence distribution."],"forward_implications":["If the limits hold, a typical non-repeating FRB in this sample repeats less than once per 20 hours, meaning many apparent non-repeaters are not rapidly active sources.","The joint limits, based on five times more sources than prior Arecibo follow-up, tighten constraints on FRB repetition rates by a factor of about 3.","The confirmation of FRB 20190110C as a repeater after selection by the machine-learning method supports UMAP-based preselection as an efficient way to find true repeaters.","Under a Weibull model, the posterior central values of the shape parameter $k$ tend below 1 for many sources, hinting at burst clustering, although $k$ remains poorly constrained in the absence of detected bursts."],"supporting_citations":[{"why":"Supplies the empirical $L_\\nu$--$w_{\\rm int}$ separation between repeaters and non-repeaters used to select the FRBCAT empirical targets.","marker":"Hashimoto et al. (2020a)"},{"why":"UMAP machine-learning classification that identifies non-repeaters likely to be hidden repeaters; the source of the CHIME_ML target list.","marker":"Chen et al. (2022)"},{"why":"First CHIME FRB catalogue providing target properties, positional uncertainties, and the $\\alpha=-1.5$ source-count slope used in sensitivity scaling.","marker":"CHIME/FRB Collaboration et al. (2021)"},{"why":"Arecibo follow-up framework for Poisson-scaled repetition rates and exposure/sensitivity scaling, which this paper extends with FAST data and a positional-coverage factor.","marker":"Good et al. (2023)"},{"why":"Weibull-distribution Bayesian formalism for burst intervals, adapted here to compute Weibull upper limits on repetition rates.","marker":"Oppermann et al. (2018)"},{"why":"Sensitivity-limit equation used to convert each telescope's signal-to-noise threshold and system parameters into a fluence limit.","marker":"Cordes & McLaughlin (2003)"},{"why":"Bayesian confidence intervals for small burst counts, used for the Poisson repetition-rate error bars.","marker":"Kraft et al. (1991)"},{"why":"Measured burst rate of FRB 20121102A used in the Monte Carlo simulations that set the optimized 10-minute exposure time per source.","marker":"Li et al. (2021)"}],"fun_headline_variants":["Tightest FRB repetition limits from 36 silent FAST targets","36 FRBs stay quiet, yielding tightest repetition-rate limits","No bursts from 36 FRBs in FAST follow-up, limits tighten","FAST silence on 36 FRBs sets new repetition-rate bounds","10-minute FAST stares find no repeats from 36 FRBs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The upper limits assume that any repeat bursts have the same brightness distribution as the original burst, scaling as $N(>S)\\propto S^{-1.5}$ down to FAST's much deeper sensitivity, and that their widths and spectral shapes are similar to the first burst; if repeat bursts are typically fainter, the true repetition rates could be higher than reported.","fun_headline_variants_meta":{"raw":{"variants":["Tightest FRB repetition limits from 36 silent FAST targets","36 FRBs stay quiet, yielding tightest repetition-rate limits","No bursts from 36 FRBs in FAST follow-up, limits tighten","FAST silence on 36 FRBs sets new repetition-rate bounds","10-minute FAST stares find no repeats from 36 FRBs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000978,"raw_usage":{"total_tokens":4171,"prompt_tokens":982,"completion_tokens":3189,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":598,"completion_tokens_details":{"reasoning_tokens":3099}},"tokens_in":598,"tokens_out":3189,"duration_ms":27682,"temperature":1.0,"reasoning_tokens":3099,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:01:12.785965+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Detecting a burst above S/N 7 from any of the 36 sources in another 10-minute FAST exposure of comparable sensitivity would directly contradict the core non-detection result, as would measuring the fluence distribution of known repeater bursts and finding a slope steeper than $-1.5$.","supporting_citations":[],"review_version":1}