{"id":"cd9b83db-741a-4649-b4e8-6cca41805638","arxiv_id":"1908.06653","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A new IceCube analysis pipeline for neutrinos from gamma-ray burst precursor flashes is described, with a public catalog of 172 precursors and predicted sensitivity better than previous limits.","lead":"This paper builds a new catalog of precursor flashes, brief brightness bumps before powerful cosmic explosions called gamma-ray bursts, and designs a way to look for neutrinos arriving at the same time. It does not report a detection yet, but it shows the planned search would be more sensitive than earlier IceCube limits.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Stacked-sensitivity projection appears to ignore overlap between pre-prompt time windows; at Δt≈10^5 s a single IceCube event can fall in multiple GRB windows and be counted multiple times, so the 'more sensitive for all Δt' claim is not yet supported.","rationale":"I read the paper as a methods and sensitivity-projection paper: the scientific payload is the public precursor catalog and the claim that an IceCube stacking search over pre-prompt windows would beat IC40/59 for every Δt. I considered the precursor-selection validation issue raised by the reader. It is a real weakness, but it is not the most load-bearing part of the central claim, because the universal [tp−Δt, tp] stacking search is explicitly designed to be applicable to all GRBs, independent of whether a precursor was identified. The claim that would falsify the headline is the sensitivity curve itself. That curve is built by treating each GRB as an independent time window; with 1527 bursts over 8 years and Δt up to 10^5 s, windows overlap substantially. The expected number of windows covering a random time is μΔt ≈ 0.6 at Δt = 10^5 s, so a single neutrino can be counted in more than one GRB term. Section 4 does not specify a joint-likelihood treatment of overlaps. This can bias the projected average-fluence sensitivity and makes 'more sensitive for all considered Δt' under-supported unless overlap effects are shown negligible. The reader's conditional verdict remains appropriate; I would not move to reject because this is a fixable, checkable technical assumption in a projection, not a fundamental flaw in the analysis concept.","tokens_in":6715,"tokens_out":16933,"duration_ms":179691,"concrete_test":"Using the public GRB catalog, generate the 1527 pre-prompt windows [tp−Δt, tp] for Δt = 10, 10^2, 10^3, 10^4, and 10^5 s. Simulate background-only IceCube events at the stated 6.7 mHz rate and signal events injected with a fixed average fluence. Compute the median sensitivity twice: once with the naive independent-window stacking likelihood implied by Eq. (4.1), and once with a joint likelihood that assigns each event to at most one window, e.g., a union-of-windows background model with a per-event source assignment. If the two average-fluence sensitivity curves differ by more than about 10% for any Δt ≥ 10^4 s, the claim that the analysis is more sensitive for all considered Δt must be qualified and Figure 3 corrected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in Section 4 and Figure 3 is a stacking projection over 1527 GRBs with Δt up to 10^5 s. For 1527 GRBs spread over roughly 8 years, the mean rate is about 0.52 per day, so at Δt = 10^5 s (about 1.16 days) the expected number of pre-prompt windows covering a random time is μΔt ≈ 0.6, and about 45% of all times fall in at least one window. The unbinned likelihood of Eq. (4.1) is written for a single GRB, and the text says the data 'can be combined in a stacking analysis' without specifying how overlapping windows are handled. If the stacked likelihood is built as a product of per-GRB likelihood terms, the same IceCube event is entered once for each window containing it. A neutrino from one GRB can then be counted as signal for another GRB whose [tp−Δt, tp] window happens to overlap, artificially increasing the expected signal for a given average fluence; background events in overlapping windows are likewise double counted. The previous IC40/59 analysis used short prompt-phase windows, so this issue does not affect the comparison line in the same way. Because no overlap check, joint-likelihood construction, or union-of-windows treatment is described, the blanket statement that the new analysis is more sensitive for all considered Δt is not supported at large Δt.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This ICRC 2019 proceedings paper describes work toward an IceCube search for neutrinos from GRB precursors. The authors construct a new, publicly available GRB catalog, analyze Fermi-GBM data to identify precursor emission episodes (reporting 172 precursors in 145 of 1843 bursts, about 8%), and develop an unbinned-likelihood analysis for coincident IceCube neutrinos. They present a sensitivity comparison with the earlier IC40/59 GRB-neutrino analysis, claiming that their proposed analysis is more sensitive for all time-window sizes Δt up to 10^5 s. The paper is a methods and projections contribution rather than a completed measurement.","tokens_in":6962,"tokens_out":5532,"duration_ms":53413,"significance":"If the sensitivity projection is correct, the proposed analysis would substantially improve on previous IceCube constraints on GRB precursor neutrino fluence, providing a direct test of choked-jet precursor models. The public GRB catalog and the automated precursor identification are potentially useful resources for the community. The paper is preliminary, however; the central sensitivity claim rests on an unspecified stacking procedure and an unpublished event selection, and the precursor selection is not quantitatively validated. These issues need to be addressed before the stated conclusions can be accepted.","major_comments":[{"comment":"The stacked sensitivity projection is not sufficiently defined to support the claim that the analysis is 'more sensitive for all considered time window sizes Δt'. Equation (4.1) gives the likelihood for a single GRB, and the text only says the data 'can be combined in a stacking analysis' without specifying how overlapping time windows [tp−Δt, tp] are treated. For 1527 GRBs at a mean rate of about 0.5 per day, a window of Δt = 10^5 s (about 1.16 days) covers a substantial fraction of the total observing time, so a single neutrino event can fall in the windows of several GRBs. If the stacked likelihood is a product of per-GRB terms, that event is counted multiple times, artificially altering both the signal and background expectations. The authors should present the joint likelihood, explain how overlapping windows are handled (e.g., by using the union of windows or a time-dependent background), and recompute the sensitivity, or alternatively justify that overlap effects are negligible over the plotted range. Without this, the central sensitivity claim is unsupported.","section":"Section 4, Eq. (4.1), Fig. 3"},{"comment":"The precursor identification relies on hard thresholds (30 Hz rate excess, at least two NaI detectors, 2 s quiescence, and a precursor-to-prompt fluence ratio below 1/3) that are adopted from a previous search, but no systematic variation of these thresholds is presented, and no uncertainty is attached to the reported 8% fraction (145/1843 GRBs). The only validation is the statement that 'roughly half' of the identified precursors were also seen in the previous Fermi-GBM search, which is qualitative and gives no overlap fraction with errors. Since the precursor-tagged analysis in Section 4 uses these episodes to define the neutrino search windows, a misclassification rate would directly affect the expected background and the final sensitivity. The authors should provide quantitative validation, such as injection tests or a detailed comparison with the catalog of [13], or explicitly state that the precursor sample is preliminary and not yet used for the sensitivity claims.","section":"Section 3"},{"comment":"The sensitivity calculation is not reproducible from the information given. The event selection is described only as 'an improved event selection' and the figures are labeled 'IceCube preliminary', with no reference to the underlying event sample, effective area, or background rate as a function of declination. In addition, the statement that λ≈2.3 corresponds to detecting a signal in 90% of background-free trials is a statistical claim that should be justified by a pseudo-experiment or an analytic calculation. Without these details, a reader cannot verify the claimed improvement over IC40/59. The authors should either provide a reference to a public description of the event selection or give the necessary detector response information in the paper.","section":"Section 4, Fig. 2"}],"minor_comments":[{"comment":"The fraction 145/1843 is 7.9%; the text rounds to 'about 8%'. Please quote the fraction with a Poisson uncertainty (e.g., 7.9% ± 0.6%) or state explicitly that the value is preliminary and not yet part of a full measurement.","section":"Section 3"},{"comment":"The caption does not explain the color coding of the Bayesian block histogram or the relationship between the top and bottom panels. Please add a sentence describing the yellow precursor bins and red prompt bins, and note that both panels show the same GRB.","section":"Figure 1"},{"comment":"The database is said to contain 'over six thousand GRBs', but Section 3 uses a sample of 1843 Fermi-GBM bursts. Please clarify the difference between the full catalog and the sample used for the precursor search, and explain why the IceCube analysis uses the latter.","section":"Section 2"},{"comment":"The Poisson prefactor (n_s+n_b)^N / N! is written in a compact form that may confuse readers; consider writing it as e^{-(n_s+n_b)} (n_s+n_b)^N / N! to make the normalization explicit.","section":"Eq. (4.1)"},{"comment":"The author list 'K. Murase and other' is incomplete; please use the full list or 'et al.'.","section":"Reference [9]"},{"comment":"The abstract says the paper will 'discuss the implications in case a significant signal is found', but the text contains no such discussion. Either add a short outlook or adjust the abstract.","section":"Abstract / Section 5"}],"recommendation":"major_revision","confidential_remarks":"The main issue is the undefined stacking procedure in the sensitivity comparison; if the authors can show that overlapping time windows are handled correctly (or that their effect is negligible), the paper would be a reasonable methods note for a conference proceedings. The precursor validation and the unpublished event selection are secondary but should be addressed for completeness. The paper is preliminary and would benefit from a clear statement that the sensitivity numbers are estimates pending a more detailed analysis."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real deliverable here is the public catalog: 172 precursor episodes across 145 GRBs from 2011–2018, extracted from Fermi-GBM with a reproducible Bayesian-block procedure, plus the GRBweb database and parsing tools. That is a concrete, citable contribution that others in the field can use. The sensitivity projection for a stacked IceCube search is also a reasonable next step, and the paper is honest that no neutrino search has been run yet.\n\nThe methods are standard and clearly described. The precursor definition follows the previous GBM search, which is a sensible anchor. The background characterization and Bayesian-block segmentation are standard and do not look problematic. The single-GRB sensitivity argument for Δt < 1000 s is sound: with background essentially zero, the required number of signal events is independent of window size.\n\nNow the soft spots. The precursor selection depends on thresholds—30 Hz excess, two-detector coincidence, 2 s quiescence, one-third fluence ratio—that are arbitrary and not varied. The paper reports 8% with no error bar and admits only half the precursors overlap with the prior search; that alone suggests the selection is not yet well validated. The sensitivity projection also relies on an unpublished IceCube event selection, so the absolute numbers cannot be checked.\n\nThe bigger issue is the stacked sensitivity at large Δt. The likelihood in Eq. (4.1) is written for a single GRB, and the text says data 'can be combined in a stacking analysis' without explaining how overlapping pre-prompt windows are handled. With 1527 GRBs over 8 years, at Δt = 10^5 s the windows frequently overlap; a single neutrino event could fall in multiple GRB time windows and be counted multiple times if the stacking is a simple product of per-GRB likelihood terms. That would artificially inflate the expected signal for a given fluence. The previous IC40/59 analysis used window sizes of seconds, so this problem was negligible there. The paper gives no overlap check or union-of-windows treatment, so the claim that the new analysis is more sensitive for all considered Δt is not yet supported.\n\nFor small Δt, the comparison is probably fine. For large Δt, it is genuinely uncertain. This is fixable: show how the stacked likelihood treats overlapping windows, run pseudo-experiments, and validate the precursor catalog against independent data.\n\nWho should read this? GRB and neutrino folks who want the precursor catalog or plan to do a precursor search. The catalog itself is worth a citation even if the sensitivity claim needs work. I would send this to peer review, but only after the overlap issue is addressed. It is a conference proceedings, so the bar is lower, but the advertised sensitivity claim should not stand as is.","headline":"Useful public GRB precursor catalog and a plausible sensitivity projection, but the blanket 'more sensitive for all Δt' claim needs an overlap check before it can be trusted.","tokens_in":7554,"tokens_out":1578,"would_cite":false,"duration_ms":18328,"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":"GRB precursor neutrinos may be the real signal, and a new IceCube search targets them","keywords":["gamma-ray bursts","neutrino astronomy","GRB precursors","IceCube","Fermi-GBM","unbinned likelihood","choked jet","time window"],"falsifier":"Re-run the same precursor-identification algorithm on randomized Fermi-GBM background intervals, shifting the light curves so no real burst is present, and count how often the 30 Hz, two-detector, 2 s quiescence rule flags a precursor; if the false-positive rate is comparable to the reported 8% occurrence rate, the precursor sample is dominated by background fluctuations and the neutrino windows are not tied to real emission. A second decisive check would compute the stacked IceCube sensitivity with the same 1527 bursts but replacing the precursor windows with randomly shifted windows of equal length; a similar sensitivity curve would show the time information carries no power.","tokens_in":6464,"feed_emoji":"🔭","tokens_out":7481,"duration_ms":72167,"temperature":0.7,"pith_summary":"This paper argues that the precursor stage of a gamma-ray burst, and not the prompt flash, may be the dominant time for high-energy neutrino production, and that IceCube's earlier GRB searches looked in the wrong window. To enable a search in the right window, the authors build an up-to-date public GRB catalog, identify precursor flashes in Fermi-GBM data from 2011 through 2018, and design an unbinned maximum-likelihood IceCube analysis over those precursor time windows. They find 172 precursors in 145 of 1843 bursts, about 8 percent, with roughly half of the overlapping events matching an earlier Fermi-GBM precursor search. The central quantitative claim is that their proposed stacked search, using 1527 GRBs, is more sensitive to the average neutrino fluence from GRB precursors than the previous IC40/59 IceCube analysis for every considered time window size.","feed_headline":"New IceCube search would tighten GRB precursor neutrino limits","feed_subtitle":"A background-free window under 1000 s and five times more bursts would beat the old limits at every window size.","key_machinery":"The load-bearing machinery is a two-stage pipeline. Stage one defines the target: Fermi-GBM light curves are background-fitted with a linear model, rebinned into Bayesian blocks (variable-width histograms that optimize bin edges to separate signal from background), and any bin whose rate exceeds the background by 30 Hz in at least two NaI detectors, is separated from the prompt episode by at least 2 s of quiescence, and carries less than one third of the prompt fluence is tagged as a precursor. Stage two quantifies the neutrino sensitivity: an unbinned Poisson maximum-likelihood ratio with spatial, energy, and temporal probability densities and a flat signal time profile inside the window. The key simplification is that for a single GRB with $\\Delta t < 1000$ s the expected number of background events in the window is $\\ll 1$, so the search is background-free and the required signal is a fixed roughly 2.3 events, independent of $\\Delta t$.","core_discovery":"The paper claims that high-energy neutrinos from GRB precursors have not yet been searched for, and that a search is now feasible. It presents a new public GRB catalog, a Fermi-GBM precursor sample of 172 episodes in 145 bursts, and an unbinned likelihood analysis for IceCube that combines spatial, energy, and temporal information. For a single GRB with a time window shorter than 1000 s, the expected number of background events is far below one, so the search is essentially background-free and the fluence sensitivity does not depend on the window size. In a stacked analysis over 1527 GRBs, the authors show their projected sensitivity to the average neutrino fluence is better than that of the earlier IC40/59 analysis for all time windows considered, primarily because of the full detector geometry, improved event selection, and roughly five times more bursts.","pith_inferences":["An implication left implicit is that the same window-selection logic can be applied to subthreshold GRB-like bursts already stored in the database; if precursors occur without a prompt trigger, those neutrinos would otherwise be missed.","Because the precursor selection flags only gamma-ray-bright flashes, any choked-jet precursor that produces neutrinos without a detectable gamma-ray excess is invisible to this search; a neutrino follow-up of all GBM triggers, not just flagged precursors, would test that population.","The flat time profile in the likelihood ignores the internal light-curve shape of each precursor; folding in the observed time structure could increase sensitivity for long windows while also testing the choked-jet picture, but this is a modification the paper does not make.","The same unbinned likelihood with per-event time windows could be applied to other transient classes with known pre-main-emission activity, such as repeating fast radio bursts, if precursor-like gamma-ray counterparts are identified."],"forward_implications":["For time windows shorter than 1000 s, the single-GRB search is essentially background-free, so a coincident neutrino would be significant and the fluence limit does not depend on how the window size is chosen.","The stacked search over 1527 GRBs would place average per-GRB fluence limits below the IC40/59 limits for every window size plotted, making it the strongest available constraint on GRB precursor neutrino emission.","The public catalog and precursor table let other instruments or follow-up programs reuse the same windows for independent searches.","Choked-jet and other precursor-emission models that predict neutrino fluence above the projected sensitivity can be directly confirmed or excluded by this analysis."],"supporting_citations":[{"why":"Earlier survey establishing that a small fraction of GRBs show precursor emission episodes; motivates the precursor classification cuts.","marker":"[6]"},{"why":"Earlier precursor characterization and the one-third-fluence criterion used to separate precursor from prompt episodes.","marker":"[7]"},{"why":"Choked-jet model in which a stalled jet produces the precursor and enhances hadronic neutrino emission; the physical motivation for searching precursors.","marker":"[8]"},{"why":"Previous Fermi-GBM precursor search whose selection criteria this analysis follows and against which it compares overlapping detections.","marker":"[13]"},{"why":"Bayesian blocks algorithm used to build the variable-bin light curves from which precursor excesses are identified.","marker":"[14]"},{"why":"IC40/59 IceCube GRB neutrino search whose average fluence limits are the baseline this analysis claims to surpass.","marker":"[17]"},{"why":"IceCube unbinned maximum-likelihood GRB neutrino search that supplies the likelihood structure and signal/background probability densities.","marker":"[5]"},{"why":"IceCube detector description that defines the event sample, filtering, and reconstruction underlying the sensitivity calculation.","marker":"[16]"}],"fun_headline_variants":["IceCube launches first dedicated search for GRB precursor neutrinos","GRB precursor neutrinos get their own IceCube hunt","Background-free IceCube search targets GRB precursor neutrinos","New IceCube analysis beats old GRB precursor neutrino limits","IceCube to probe GRB precursor neutrinos in background-free window"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The neutrino search is only as reliable as the precursor definition: a precursor is whatever passes the 30 Hz, two-detector, at-least-2-s-quiescence, less-than-one-third-fluence cuts, and if that tag actually picks up background fluctuations or the beginning of the prompt emission, the neutrino time windows are placed at the wrong moments.","fun_headline_variants_meta":{"raw":{"variants":["IceCube launches first dedicated search for GRB precursor neutrinos","GRB precursor neutrinos get their own IceCube hunt","Background-free IceCube search targets GRB precursor neutrinos","New IceCube analysis beats old GRB precursor neutrino limits","IceCube to probe GRB precursor neutrinos in background-free window"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000767,"raw_usage":{"total_tokens":3366,"prompt_tokens":878,"completion_tokens":2488,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":2404}},"tokens_in":494,"tokens_out":2488,"duration_ms":18661,"temperature":1.0,"reasoning_tokens":2404,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:38:16.011195+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same precursor-identification algorithm on randomized Fermi-GBM background intervals, shifting the light curves so no real burst is present, and count how often the 30 Hz, two-detector, 2 s quiescence rule flags a precursor; if the false-positive rate is comparable to the reported 8% occurrence rate, the precursor sample is dominated by background fluctuations and the neutrino windows are not tied to real emission. A second decisive check would compute the stacked IceCube sensitivity with the same 1527 bursts but replacing the precursor windows with randomly shifted windows of equal length; a similar sensitivity curve would show the time information carries no power.","supporting_citations":[{"cited_title":"Burlon et al., ApJL 685 (Sept., 2008) L19","cited_arxiv_id":null,"evidence_quote":"Earlier survey establishing that a small fraction of GRBs show precursor emission episodes; motivates the precursor classification cuts."},{"cited_title":"Lazzati, MNRAS 357 (Feb, 2005) 722","cited_arxiv_id":null,"evidence_quote":"Earlier precursor characterization and the one-third-fluence criterion used to separate precursor from prompt episodes."},{"cited_title":"Wang and P","cited_arxiv_id":null,"evidence_quote":"Choked-jet model in which a stalled jet produces the precursor and enhances hadronic neutrino emission; the physical motivation for searching precursors."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Previous Fermi-GBM precursor search whose selection criteria this analysis follows and against which it compares overlapping detections."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Bayesian blocks algorithm used to build the variable-bin light curves from which precursor excesses are identified."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"IC40/59 IceCube GRB neutrino search whose average fluence limits are the baseline this analysis claims to surpass."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"IceCube unbinned maximum-likelihood GRB neutrino search that supplies the likelihood structure and signal/background probability densities."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"IceCube detector description that defines the event sample, filtering, and reconstruction underlying the sensitivity calculation."}],"review_version":1}