{"id":"72875f72-373f-4a5f-8c17-99e0651ede03","arxiv_id":"2507.11977","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":0.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A conference proceedings that reviews recent CMS boosted Higgs searches and machine-learning jet taggers without adding a new measurement.","lead":"This proceedings paper summarizes recent CMS searches for boosted Higgs bosons, where the Higgs is produced with so much transverse momentum that its decay products merge into one large-radius jet. It highlights machine-learning tagging techniques and reports sensitivity results for H to WW, HH to bbVV, VH to bb, and gamma H production.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Boosted-Higgs results quoted in §§3–5 hinge on ML taggers shown only on CMS simulation; without data/MC closure for GloParT after Lund reweighting, the quoted significances and limits are not independently supported.","rationale":"The reader's weakest assumption — that ML tagger performance demonstrated only in CMS simulation transfers to data after Lund jet plane reweighting and control-region extrapolation — is exactly the point on which the quoted physics numbers rest. The proceedings gives no data/MC closure evidence, and the placeholder in Section 6 makes part of the claimed results unverifiable. This is a genuine limitation of the document as a standalone scientific statement. However, the paper is explicitly a proceedings summary of approved CMS notes; the appropriate place for the missing validation is in those notes, not necessarily in this summary. The absence of validation therefore does not by itself prove the quoted results wrong, but it does mean the reader cannot verify the central claims from the preprint. Since the reader's UNVERDICTED verdict already captures this non-verifiability, my concern does not move the verdict; it reinforces it. I do not claim the underlying CMS analyses are incorrect, only that this document does not provide the evidence needed to assess their correctness.","tokens_in":5438,"tokens_out":5857,"duration_ms":75619,"concrete_test":"Obtain CMS-PAS-HIG-24-008 and CMS-PAS-JME-23-001 and check whether GloParT/P(H1ℓ) is validated in data: compare tagger output distributions in a high-pT W/Z+jets control region before and after Lund jet plane reweighting, and verify closure by checking the data/MC ratio in bins of the multi-prong Lund plane. Then re-evaluate the expected H→WW significance with the tagger selection efficiency shifted by the measured data/MC residual in the signal region. If the expected significance moves by more than the quoted systematic uncertainty, the reported 1.76σ expectation is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claims — expected 1.76σ for boosted H→WW, μ_VH = 0.72+0.75−0.71, constraints on κ_2V and on γH production — all depend on the large-radius jet taggers (ParticleNet-MD, GloParT, and the fine-tuned P(H1ℓ)) performing in data as they do in simulation. The only performance evidence shown is Fig. 1, labeled 'CMS Simulation Preliminary'. Section 3 states that GloParT 'is trained in Monte Carlo (MC) simulation' and is calibrated with Lund jet plane reweighting, and Section 4 repeats this for HH→bbVV, but the text presents no data/MC comparison of tagger score distributions, no ROC curve evaluated in data, and no closure test for the reweighting. The QCD background is estimated by extrapolation from control regions, which further assumes the tagger's background rejection is correctly modeled after extrapolation. If simulation does not faithfully model the relevant jet substructure in the boosted phase space, or if the control-region extrapolation fails to capture a soft-drop-mass sculpting effect, the quoted observed significances, fitted signal strengths, and limits would shift. This is a validation gap, not an internal inconsistency. Section 6 is also unfinished, with the literal placeholder 'Figures we want to show in the paper' replacing the γH and light-quark Yukawa constraint plots, so those results cannot be checked from this text either.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This proceedings contribution from the CMS collaboration summarizes recent searches for boosted Higgs bosons. It describes ML-based large-radius jet tagging with ParticleNet-MD, GloParT, and a fine-tuned GloParT variant; the boosted H->WW search with 0-lepton and 1-lepton categories; the HH->bbVV search and its constraints on the kappa_2V coupling; the V(qq)H(bb) measurement with mu_VH = 0.72 (+0.75, -0.71); and first LHC constraints on gamma-H production and light-quark Yukawa couplings. The numerical results are taken from the cited CMS physics analysis summaries and prior publications; the paper contains no new derivation, fit, or data analysis.","tokens_in":5725,"tokens_out":4297,"duration_ms":50734,"significance":"If the quoted results are correct, they represent genuine experimental progress: the first dedicated CMS boosted H->WW search, a competitive kappa_2V constraint from HH->bbVV, the first dedicated CMS boosted VH->bb search, and the first LHC constraints on gamma-H production. The paper is a compact, well-referenced compilation of these results, and the cited numbers are internally consistent with the corresponding CMS notes. Its principal value is as a proceedings summary; the quantitative support for the claims resides in the cited CMS analyses, not in this manuscript. The paper also has an unfinished placeholder in Section 6, so as submitted it is not a complete article.","major_comments":[{"comment":"The two right-hand panels of Fig. 3 contain the literal placeholder text 'Figures we want to show in the paper' in place of the gamma-H cross-section constraints and the light-quark Yukawa coupling constraints. Section 6 refers to these as 'shown in Fig. 3', so the headline claims of the section cannot be checked from the manuscript. Please replace the placeholders with the final plots, or remove the specific claims and refer the reader to CMS-PAS-HIG-23-011.","section":"Section 6, Fig. 3"},{"comment":"The only performance evidence for the taggers that underpin the quoted significances, signal strengths, and limits is a CMS simulation ROC curve. Section 3 states that GloParT is trained in Monte Carlo simulation and calibrated with Lund jet plane reweighting, but the paper provides no data/MC comparison of tagger score distributions, no control-region closure test, and no reference to where these validation studies are documented in the cited CMS notes. Since all quantitative results in Sections 3-5 depend on these taggers, please explicitly point to the validation sections of CMS-PAS-HIG-24-008, CMS-PAS-HIG-24-017, and CMS-PAS-JME-23-001, or add a short validation summary.","section":"Sections 2-4, Fig. 1"},{"comment":"The text reports an observed significance of 0 and expected significance of 1.76 sigma for the combined boosted H->WW search, but it does not give the corresponding observed upper limit on the cross section. For a proceedings summary this is a completeness issue, but it is worth stating the observed 95% CL limit from CMS-PAS-HIG-24-008 so that readers can compare the sensitivity with the quoted expected significance.","section":"Section 3"}],"minor_comments":[{"comment":"There is a typo in the sentence 'second-strongest CMS constraints on the the quartic VVHH coupling modifier': the word 'the' is repeated.","section":"Section 4"},{"comment":"The legend of the middle panel lists 'H WW* qq vs. QCD' twice; one of the entries should presumably denote a different final state, such as the 1-lepton or 3-prong topology. Please correct the legend.","section":"Fig. 1 (middle)"},{"comment":"The phrase 'the combined mu_VH is measured (expected) to be' is slightly ambiguous because the expected value is a fit result under the SM hypothesis rather than an experimental measurement. Please phrase it as 'observed (expected under the SM) value' for consistency with Section 3.","section":"Section 5"},{"comment":"The claim 'first gamma-H results at the LHC' appears in the text without an immediate citation to CMS-PAS-HIG-23-011; the reference is listed in the bibliography but not cited at the point of the claim.","section":"Section 6"},{"comment":"The paper would benefit from a sentence at the end of the Introduction stating that all results are taken from the cited CMS notes and that the figures are CMS preliminary material, since the body text otherwise reads as if the quoted numbers are derived in this paper.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"This is a conference proceedings summary by a CMS collaboration member, and the central results are all from external CMS notes. The main blocking issue is the unfinished placeholder content in Section 6 and the lack of any explicit validation pointer for the ML taggers. For this type of article, I would not require full analysis details, but the placeholder must be replaced and the validation references should be made explicit before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a proceedings write-up, not a research paper. It has zero new measurements, but it is a clean, honest summary of four recent CMS boosted-Higgs analyses, and the numbers it quotes are traceable to the cited notes. The main problem is that the posted text is unfinished: Section 6 contains a literal placeholder, 'Figures we want to show in the paper,' which should never have been submitted.\n\nCredit where due: the paper gives a useful overview of the ML taggers (ParticleNet-MD, GloParT, fine-tuned P(H1ℓ)) and the analyses they enable: first dedicated CMS boosted H→WW search, second-strongest κ2V constraint from HH→bbVV, first dedicated boosted VH→bb search, and first LHC constraints on γH production. Those are real experimental firsts, but they belong to the PAS notes, not to this proceedings.\n\nSoft spots: the tagger performance evidence is entirely from CMS simulation (Fig. 1, 'CMS Simulation Preliminary'), with no data/MC closure test shown for the Lund jet plane reweighting that calibrates GloParT. Quoted significances and limits rest on the taggers performing in data as they do in simulation. That is a genuine validation gap, but it is a gap in the underlying analyses, not a flaw in this summary document. For a proceedings, it is acceptable to point to the notes instead of reproducing the validation. The placeholder in Section 6 is the real quality-control issue; also, the γH constraints and light-quark Yukawa results cannot be checked from the text because the plots are missing.\n\nBottom line: this paper is for readers who want a compact pointer to the current CMS boosted-Higgs program. It deserves a serious referee in the sense that a proceedings editor should check that the quoted results match the cited notes and should demand removal of the placeholder before publication. It is not a contribution that advances the literature by itself.","headline":"A clean but unfinished proceedings summary of four real CMS boosted-Higgs analyses; the placeholder in Section 6 is the only hard flaw.","tokens_in":6207,"tokens_out":2178,"would_cite":false,"duration_ms":22643,"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":"Backed by graph-network and transformer jet taggers, CMS reports new dedicated searches in four boosted-Higgs topologies, including first LHC limits on gamma-H production.","keywords":["boosted Higgs bosons","jet substructure","large-radius jets","machine-learning taggers","ParticleNet-MD","GloParT","Higgs pair production","gamma-H production"],"falsifier":"Measure the background efficiency of ParticleNet-MD on a data control sample (for example QCD multijet events) at a fixed signal efficiency in the same phase space as Figure 1, such as $450 < p_T < 600$ GeV and $90 < m_{SD} < 140$ GeV; if the data efficiency exceeds the simulation ROC curve by more than the quoted systematic uncertainty, the expected significances such as the $1.76\\,\\sigma$ for boosted $H\\to WW$ would not be reproduced.","tokens_in":5239,"feed_emoji":"⚛️","tokens_out":15153,"duration_ms":150935,"temperature":0.7,"pith_summary":"These proceedings claim that the boosted-Higgs regime at the LHC has become experimentally accessible, and that CMS has now produced dedicated searches in four channels in which a Higgs boson is highly boosted and its decay products merge into a single large-radius jet. The key enablers are two deep-learning taggers: ParticleNet-MD, a graph neural network for $X(qq)$ jets, and GloParT, a transformer covering 37 classes including $X(VV)$ and top pairs. With them, CMS reports an expected significance of $1.76\\,\\sigma$ for boosted $H\\to WW$ (observed 0), a constraint on the quartic coupling modifier $\\kappa_{2V}$ from $HH\\to bbVV$, $\\mu_{VH}=0.72^{+0.75}_{-0.71}$ for $V(qq)H(bb)$, and first LHC constraints on $\\gamma H$ production. Several of these are first dedicated searches in their final states; each quantitative result originates in a separate CMS analysis summarized here.","feed_headline":"Neural-network jet taggers power first boosted-Higgs searches at CMS","feed_subtitle":"They are dedicated CMS searches in four boosted-Higgs channels, with first LHC limits on gamma-H.","key_machinery":"The machinery that carries the argument is the combination of soft-drop groomed large-radius jets with two learned taggers: ParticleNet-MD, a graph neural network that sorts jets into eight classes including $X(qq)$, and GloParT, a transformer that sorts jets into 37 classes including $X(VV)$ and top pairs. The groomed jet mass $m_{SD}$ is the observable used for signal extraction, and the taggers are trained in simulation, with GloParT calibrated to data using Lund jet plane reweighting (a data-driven correction to simulated jet substructure) and, for the one-lepton $H\\to WW$ channel, fine-tuned into a dedicated model $P(H_{1\\ell})$ that improved expected significance by nearly 70%. These objects carry the argument by suppressing the QCD multijet background by orders of magnitude at fixed signal efficiency.","core_discovery":"The central discovery reported is that a dedicated boosted-Higgs search program is now running at CMS, enabled by large-radius jets and modern deep-learning taggers. The paper presents four analyses: $H\\to WW$ in all-hadronic and one-lepton categories, $HH\\to bbVV$, $V(qq)H(bb)$, and $\\gamma H$. Its headline numbers are an observed (expected) significance of 0 ($1.76\\,\\sigma$) for $H\\to WW$; the second-strongest CMS constraint on $\\kappa_{2V}$ from $HH\\to bbVV$; $\\mu_{VH}=0.72^{+0.75}_{-0.71}$ and $\\mu_{VZ}=0.09\\pm0.63$ for the boosted $VH$ search; and first LHC limits on $\\sigma_{\\gamma H}$ using $H\\to bb$ and $H\\to4\\ell$ decays. The paper's claim is that these are the first dedicated searches in their final states and that the taggers are what make that sensitivity possible.","pith_inferences":["Editorial inference: The near-70% expected-significance gain from fine-tuning GloParT on the one-lepton $H\\to WW$ signature suggests that signal-specific transfer learning may be the fastest near-term sensitivity lever for rare boosted-Higgs final states.","Editorial inference: Because GloParT and ParticleNet-MD are generic classifiers of 37 and 8 jet classes, the same trained architectures could be transferred to boosted decays of other heavy resonances (for example, $W$, $Z$, top, or exotic particles) with only modest fine-tuning, making this search program a template for a broader boosted-object program.","Editorial inference: The quoted significances and limits assume that tagger performance measured in simulation survives the Lund-plane calibration in data; a public data-vs-simulation closure test of the ROC curves in the exact phase space of Figure 1 would either confirm or undermine that assumption."],"forward_implications":["The $H\\to WW$ search is statistics-limited: with observed significance 0 and expected $1.76\\,\\sigma$, adding HL-LHC data is a direct route to first evidence.","The $HH\\to bbVV$ analysis gives a new, independent bound on the quartic coupling modifier $\\kappa_{2V}$ that can be combined with the stronger $HH\\to4b$ channel.","The first dedicated boosted $V(qq)H(bb)$ search measures $\\mu_{VH}=0.72^{+0.75}_{-0.71}$, consistent with the standard model, so no enhancement of vector-boson couplings appears in this channel.","The first LHC constraints on $\\gamma H$ production, although statistics-limited, open a new production mode where an observation would be clear evidence of beyond-standard-model physics.","The planned first ParticleNet $bb$ triggers and improved GloParT taggers are expected to bring significant sensitivity gains in the HL-LHC era."],"supporting_citations":[{"why":"Supplies the boosted H->WW analysis, including the 1.76 sigma expected significance and the fine-tuned GloParT model P(H1l).","marker":"[11]"},{"why":"Supplies the HH->bbVV analysis and the kappa_2V constraint reported here.","marker":"[7]"},{"why":"Supplies the boosted V(qq)H(bb) analysis and the mu_VH measurement.","marker":"[17]"},{"why":"Supplies the first gamma-H cross-section constraints and light-quark Yukawa limits.","marker":"[20]"},{"why":"Documents the ParticleNet-MD tagger performance for H->bb identification in boosted topologies.","marker":"[5]"},{"why":"Supplies the GloParT transformer architecture and its deployment for boosted H->WW tagging.","marker":"[6,7]"},{"why":"Supplies the Lund jet plane reweighting method used to calibrate the GloParT tagger to data.","marker":"[12,13]"}],"fun_headline_variants":["CMS neural nets open first dedicated boosted-Higgs searches","First gamma-H limits from CMS boosted-Higgs searches","CMS boosted-Higgs: null WW, first gamma-H limits","Deep-learning taggers: key to CMS's four boosted-Higgs analyses","CMS harnesses neural nets for boosted-Higgs searches"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the machine-learning jet taggers, trained on simulated jets, classify real proton-proton jets with the same efficiency and background rejection after the data-driven substructure calibration and control-region extrapolations.","fun_headline_variants_meta":{"raw":{"variants":["CMS neural nets open first dedicated boosted-Higgs searches","First gamma-H limits from CMS boosted-Higgs searches","CMS boosted-Higgs: null WW, first gamma-H limits","Deep-learning taggers: key to CMS's four boosted-Higgs analyses","CMS harnesses neural nets for boosted-Higgs searches"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000793,"raw_usage":{"total_tokens":3423,"prompt_tokens":802,"completion_tokens":2621,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":418,"completion_tokens_details":{"reasoning_tokens":2534}},"tokens_in":418,"tokens_out":2621,"duration_ms":23236,"temperature":1.0,"reasoning_tokens":2534,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:56:06.842399+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the background efficiency of ParticleNet-MD on a data control sample (for example QCD multijet events) at a fixed signal efficiency in the same phase space as Figure 1, such as $450 < p_T < 600$ GeV and $90 < m_{SD} < 140$ GeV; if the data efficiency exceeds the simulation ROC curve by more than the quoted systematic uncertainty, the expected significances such as the $1.76\\,\\sigma$ for boosted $H\\to WW$ would not be reproduced.","supporting_citations":[{"cited_title":"SearchforHiggsbosonproductionathightransversemomentuminthe WW* decay channel in proton-proton collisions at√𝑠 = 13 TeV","cited_arxiv_id":null,"evidence_quote":"Supplies the boosted H->WW analysis, including the 1.76 sigma expected significance and the fine-tuned GloParT model P(H1l)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the boosted V(qq)H(bb) analysis and the mu_VH measurement."},{"cited_title":"Search for𝛾H production and constraints on the Yukawa couplings of light quarks to the Higgs boson","cited_arxiv_id":null,"evidence_quote":"Supplies the first gamma-H cross-section constraints and light-quark Yukawa limits."},{"cited_title":"Performanceofheavy-flavourjetidentificationinboostedtopologiesin proton-proton collisions at√𝑠 = 13 TeV","cited_arxiv_id":null,"evidence_quote":"Documents the ParticleNet-MD tagger performance for H->bb identification in boosted topologies."}],"review_version":1}