{"id":"8759f037-b315-4989-9ed8-208189f55303","arxiv_id":"2607.22193","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"ATLAS reports the first boosted H→ττ cross-section measurements for Higgs pT > 300 GeV, with an observed (expected) significance of 3.8 (3.3) standard deviations, consistent with the Standard Model.","lead":"ATLAS measured how often Higgs bosons are produced with very high sideways momentum and decay into pairs of tau leptons, using LHC collision data from 2015 to 2024. The results agree with Standard Model predictions and give the first evidence for boosted Higgs production in the di-tau channel.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 3.8σ evidence hinges on the fake di-τ background shape in the signal region, which is transferred from the anti-ID region with only a flat normalization uncertainty and no BDT-score shape uncertainty or closure test.","rationale":"The reader identified the fake-factor transfer as the weakest assumption, and I agree. My stress-test sharpens this by noting that the observed Run 3 excess, which drives the combined 3.8σ significance, occurs in the dataset with the largest fake background. The fake factors are parameterized by only two kinematic/topology variables, not by the BDT-score fit variable, and the assigned 100% uncertainty is normalization-only. This is a concrete, checkable gap: a closure test in BDT-score bins using simulated QCD/W+jets events would either reveal a shape bias large enough to affect the central claim or confirm that the flat normalization uncertainty is sufficient. Because the paper does not provide such a test, I would condition acceptance on that additional validation. I am not rejecting the measurement: the analysis is internally consistent, the OMNI tagger and control regions are well described, and the concern may well resolve unfavourably for the critique. But for a result whose headline is an evidence-level significance, the shape of the dominant background in the fit variable should not be protected only by a normalization uncertainty.","tokens_in":62644,"tokens_out":6329,"duration_ms":74613,"concrete_test":"Using simulated QCD-multijet and W+jets events that pass the OS anti-ID selection, compare the nominal fake-factor SR prediction with the actual ID-region yield as a function of the final BDT score. If the ID/anti-ID ratio varies significantly with BDT score, apply the measured BDT-dependent correction as an alternate fake template and refit the inclusive signal strength. A shift in μ larger than the quoted fake-systematic contribution (~0.1–0.25) or a change in significance of more than ~1σ would demonstrate that the missing shape uncertainty is load-bearing; conversely, a negligible shift would validate the current treatment.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The inclusive significance is dominated by Run 3 (3.6σ observed vs 2.6σ expected), where the fake di-τhad background is explicitly larger because of higher pile-up and looser triggers (§9). The fake background in the signal region is estimated by taking events from the anti-ID (OMNI-fail) OS region and multiplying by fake factors parameterized only by subleading-subjet pT and track multiplicity (§6). The BDT-score shape of the fake background in the SR is therefore inherited from the anti-ID region, while the assigned systematic in §8 is a flat 100% normalization uncertainty on the total SR fake yield. If the ID/anti-ID ratio varies with the BDT score, the shape of the fake template is biased. Since the BDT score is the fit variable, a mild shape mismatch could mimic or suppress a Higgs-like excess at high BDT score and shift the fitted signal strength by more than the quoted normalization uncertainty. The paper presents no closure test of the fake-factor prediction in an OMNI-ID region, e.g., using same-sign events after subtracting real-τ processes, and no shape uncertainty on the transferred template.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents cross-section measurements of highly boosted Higgs boson production decaying to pairs of hadronically decaying tau leptons, using 140 fb^-1 of Run 2 data at 13 TeV and 162 fb^-1 of Run 3 data at 13.6 TeV. Events are selected with a dedicated large-radius-jet di-tau reconstruction and the OMNI transformer-based tagger. Signal and background are separated with a two-class BDT, and events are further categorized into VBF and three ggF pT bins using a four-class BDT. Backgrounds are estimated from simulation, with the Z->tau tau normalization constrained in control regions and the misidentified di-tau background estimated with a data-driven fake-factor method. A profile-likelihood fit yields an observed (expected) significance of 3.8 (3.3) sigma for inclusive H->tau tau production with pT(H)>300 GeV, a combined signal strength of mu=1.16 (+0.42/-0.25 stat, +0.23/-0.27 syst), and STXS and differential cross-section results consistent with the SM within uncertainties.","tokens_in":62962,"tokens_out":3449,"duration_ms":39677,"significance":"If the result holds, this is the first evidence for boosted H->tau tau production and opens a previously inaccessible high-pT phase space for Higgs coupling measurements. The analysis is methodical: it uses a dedicated reconstruction and tagger, three-fold cross-validated BDTs, Z->tau tau control regions to anchor the dominant irreducible background and the in-situ tau energy scale, and a complete inventory of experimental and theoretical systematics. The observed and expected significances are consistent, and the cross-section results are presented in both STXS and differential forms. The paper also provides a useful comparison with the CMS boosted H->tau tau measurement and the ATLAS boosted H->bb measurement.","major_comments":[{"comment":"The fake di-tau background template in the signal region is obtained by applying fake factors, derived in same-sign and anti-ID regions, to opposite-sign anti-ID events, and the fit variable is the two-class BDT score. The only systematic assigned to this background is a flat 100% normalization uncertainty on the total SR yield (§8). However, §8 states that this uncertainty is intended to account for 'possible differences in the templates of the fake di-tau background between the ID and anti-ID regions.' A flat normalization uncertainty cannot cover BDT-score shape differences. No closure test is shown in an OMNI-passing region (e.g., same-sign events after subtracting real-tau contributions), and no shape uncertainty is assigned to the fake template. Since the BDT score is the discriminating variable and the fake background is larger in Run 3 (§9), a shape mismatch between anti-ID and I","section":"§6, §8"},{"comment":"The quoted evidence for the boosted H->tau tau signal is driven by the Run 3 dataset (3.6 sigma observed vs 2.6 sigma expected), while Run 2 shows a deficit (1.7 sigma observed vs 2.1 sigma expected). The Run 3 excess is therefore the main support for the combined claim. Given the larger fake background in Run 3 and the shape-transfer concern above, it would strengthen the paper to show the stability of the signal strength under alternative fake-template definitions or under a BDT-score shape variation. This is not a request to change the results, but it would make the evidence claim more robust.","section":"§9, Table 8"}],"minor_comments":[{"comment":"The offline selections for the tau trigger in 2015/2016/2017/2018 differ (p_T(sj1) > 250 GeV vs 200 GeV in 2022–2024). It would be helpful to state explicitly that these thresholds are justified by trigger plateaus; the text only says that 'events are assured to fall well within the efficiency plateau.'","section":"§5, Table 3"},{"comment":"The fake-factor parameterization is described as a function of subleading-subjet pT and track multiplicity. It is not clear whether the fake factor is also parameterized by the prong multiplicity of the leading subjet; the text says 'number of tracks within each subjet,' but the implementation is not detailed. Please clarify.","section":"§6"},{"comment":"Minor typo: 'The yield for the fake di-tau background in in the Run 3 dataset' should read 'in the Run 3 dataset.'","section":"§9"},{"comment":"The axis label in Figure 2(b) reads 'R^gen_tauhad-vis tauhad-vis [GeV]'; the quantity is dimensionless (Delta R), so the unit should be removed.","section":"Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The reader's report recommended acceptance, but I believe the fake-template shape issue is a legitimate load-bearing concern. The paper explicitly claims that the 100% normalization uncertainty covers template differences between ID and anti-ID regions, which it does not. A closure test or shape systematic is needed to support the 3.8 sigma evidence claim. This is fixable within the manuscript's scope and does not require invalidating the analysis; hence major_revision rather than reject."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the headline: this is the first ATLAS measurement of boosted H→ττ cross sections above pT(H)=300 GeV, and the OMNI tagger is a genuine technical advance — a point-edge transformer using track-level inputs that gives about a factor of ten better background rejection than the earlier BDT. The analysis is careful: trigger plateaus, overlap removal, three-fold cross-validated BDTs for signal-background separation, and Z→ττ control regions that anchor both the in-situ tau energy scale and the dominant Z→ττ+jets background. The observed significance of 3.8σ against an expected 3.3σ is consistent, and the measured signal strength μ=1.16 (with quoted stat and syst uncertainties) is compatible with the SM at the 69% level.\n\nThe real soft spot is the fake di-τ background. In the Run 3 data the fake background is larger, and it is estimated by taking events from the anti-ID (OMNI-fail) OS region and multiplying by fake factors parameterized only by subleading-subjet pT and track multiplicity. The BDT-score shape of the fake template is therefore inherited from the anti-ID region, and the systematic assigned is a flat 100% normalization uncertainty. The paper says this uncertainty is intended to cover template differences too, but a normalization uncertainty does not by itself cover a shape mismatch in the fit variable. There is no explicit closure test in an OMNI-ID region and no shape uncertainty on the transferred template. This is a legitimate referee question and I would want to see it addressed, either with a closure test or a shape systematic, before fully trusting the signal strength.\n\nThat said, I do not think this is a load-bearing flaw. The 100% normalization uncertainty is fairly conservative, the Z→ττ control regions constrain the most important pieces, and the observed vs expected significance are close (3.8 vs 3.3). The Run 2 deficit in the pT>650 GeV bin is within statistical expectations, and the combined upper limit of 4.5×SM (expected 6.5) is not restrictive. The paper also does not provide public data or code, which is normal for ATLAS but limits independent verification.\n\nWho is this for? Anyone working on boosted Higgs measurements, tau identification, or STXS interpretations. It deserves a serious peer review, and with a reasonable request for the fake-background closure test, it should be published. I would cite it in my own work if I were active in this area.","headline":"First ATLAS boosted H→ττ measurement is solid, but the fake-background shape transfer needs a closure test.","tokens_in":63468,"tokens_out":3241,"would_cite":true,"duration_ms":34510,"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 reports cross-section measurements of Higgs boson production with transverse momentum above 300 GeV in the di-tau final state, finding evidence (3.8 sigma) consistent with the Standard Model.","keywords":["boosted Higgs","hadronic tau decays","di-tau reconstruction","cross-section measurement","simplified template cross-sections","gluon-gluon fusion","vector-boson fusion","proton-proton collisions"],"falsifier":"Recompute the fake di-tau background with fake factors parameterized additionally in the two-class BDT score and refit the signal; if the predicted background shape in the signal region shifts by more than the assigned 100% normalisation uncertainty, the fitted signal strength and the 3.8 sigma significance would change.","tokens_in":62522,"feed_emoji":"⚛️","tokens_out":5879,"duration_ms":61361,"temperature":0.7,"pith_summary":"The paper aims to establish that Higgs bosons produced with transverse momentum above 300 GeV can be measured through their decay into a pair of hadronically decaying tau leptons, a regime where the two taus merge into one jet and standard reconstruction fails. Using a dedicated boosted di-tau reconstruction and a machine-learning tagger, it reports cross-section measurements in this previously inaccessible phase space. The inclusive signal has an observed (expected) significance of 3.8 (3.3) standard deviations, with a signal strength of 1.16, compatible with the Standard Model at the 69% level. Differential and simplified-template cross-sections are also reported for gluon-gluon fusion and vector-boson fusion production, all consistent with Standard Model predictions.","feed_headline":"Boosted Higgs in tau pairs seen at 3.8 sigma","feed_subtitle":"Measurement at Higgs momenta above 300 GeV matches Standard Model and opens a new high-energy window.","key_machinery":"The load-bearing object is the boosted di-tau candidate: a large-radius jet (radius 1.0) seeded by the merged hadronic tau pair, inside which two smaller subjets (radius 0.2) are identified as the individual tau decays. Identification uses the OMNI tagger, a point-edge transformer (a graph neural network combined with transformer layers) trained on subjet and per-track features; at the 85% efficiency working point it rejects QCD multi-jet background about ten times better than the previous boosted-decision-tree tagger. The signal is extracted with a two-class boosted decision tree, events are classified into simplified-template cross-section categories with a four-class BDT, and the dominant","core_discovery":"The central discovery is that the H to tau tau process is present in the highly boosted regime: combining 140 inverse femtobarns at 13 TeV and 162 inverse femtobarns at 13.6 TeV, events with Higgs transverse momentum above 300 GeV show an excess over background-only expectations with 3.8 sigma observed (3.3 sigma expected). The measured inclusive cross-section, the simplified-template cross-sections, and the differential results in three gluon-gluon-fusion bins and one electroweak qqH bin are consistent with Standard Model predictions. This constitutes evidence for Higgs production in a phase space previously inaccessible in the di-tau channel.","pith_inferences":["If this result holds, the same boosted di-tau reconstruction should extend searches for heavy resonances decaying to tau pairs, where merged tau decays currently limit sensitivity.","The 100% normalisation uncertainty on the fake-tau background, with no explicit tagger-score shape term, means a shape mismatch in the tagger score could shift the central signal strength; this could be tested by re-deriving fake factors in bins of the tagger score.","Extrapolating the expected significance of 3.3 sigma, full Run 3 statistics could push this channel alone toward the 5 sigma observation threshold.","The Run 3 sample has a larger fake background than Run 2, suggesting future iterations may need tighter triggers or tagger optimisation; the observed run-to-run compatibility is an early indication of how that scaling behaves."],"forward_implications":["The analysis establishes the pT_H greater than 300 GeV di-tau phase space as accessible, with an observed excess of 3.8 sigma over background only.","The simplified-template and differential results give measured cross-sections for gluon-gluon fusion in three Higgs-transverse-momentum bins and for electroweak qqH production, all compatible with the Standard Model.","The Run 2 and Run 3 measurements are mutually compatible, allowing the two data sets to be combined in one likelihood fit.","The measured signal strength of 1.16 (plus 0.42 over minus 0.25 statistical, plus 0.23 over minus 0.27 systematic) is compatible with the Standard Model at the 69% level."],"fun_headline_variants":["Boosted Higgs in tau pairs: 3.8 sigma observed","ATLAS sees boosted Higgs decaying to tau pairs","Higgs at high momentum found in tau decays","Evidence for boosted Higgs in di-tau final state","3.8 sigma excess for Higgs in boosted tau channel"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing assumption is that the fake-factor method, which transfers same-sign, tagger-failing events to the opposite-sign signal region using only subleading-subjet pT and track multiplicity, predicts both the normalisation and the shape of the misidentified-tau background in the signal region.","fun_headline_variants_meta":{"raw":{"variants":["Boosted Higgs in tau pairs: 3.8 sigma observed","ATLAS sees boosted Higgs decaying to tau pairs","Higgs at high momentum found in tau decays","Evidence for boosted Higgs in di-tau final state","3.8 sigma excess for Higgs in boosted tau channel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000343,"raw_usage":{"total_tokens":1775,"prompt_tokens":852,"completion_tokens":923,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":845}},"tokens_in":596,"tokens_out":923,"duration_ms":9767,"temperature":1.0,"reasoning_tokens":845,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T05:28:58.213953+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the fake di-tau background with fake factors parameterized additionally in the two-class BDT score and refit the signal; if the predicted background shape in the signal region shifts by more than the assigned 100% normalisation uncertainty, the fitted signal strength and the 3.8 sigma significance would change.","supporting_citations":[],"review_version":1}