{"id":"bc286b6e-bd6f-476c-8e6b-b0d54d5607e4","arxiv_id":"2505.20483","paper_version":3,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"The WWZ and ZH production rates match the standard model, with a combined significance of 4.5 standard deviations and first evidence at 13.6 TeV.","lead":"CMS measured how often proton collisions at the LHC produce two W bosons and one Z boson, using 200 inverse femtobarns of data at 13 and 13.6 TeV. The result matches standard model predictions and provides the first evidence that this triboson process occurs at the newer 13.6 TeV energy.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 4.5 sigma evidence hinges on unvalidated MC modeling of BDT score shapes; a control-region data/MC closure test of those shapes would settle whether the background extrapolation is biased.","rationale":"Good-faith reading: this is a mature CMS analysis with established methodology. The selection, control regions, and simultaneous fit follow the pattern of the 2020 VVV observation. The result is statistics-dominated (systematics increase uncertainty by less than 5 percent), so the main risk is a statistical fluctuation or a background shape mis-modeling. I find no internal inconsistency in the quoted numbers: the significance (4.5 observed, 5.0 expected) and the signal-strength uncertainty (+0.31/-0.28) are consistent with a non-parabolic profile likelihood in a low-count counting experiment; the asymmetry is in the expected direction. The reader's weakest assumption identifies the BDT generalization as the key risk. I agree with that identification but sharpen it: the scores do not need to be calibrated probabilities (the fit uses them as discriminants), but the MC-modeled distributions of the scores for the dominant backgrounds in the SR must be accurate. The paper does not show a data/MC comparison of the BDT output shapes in the CRs; only the aggregate CR yields and the post-fit SR bin counts are shown. Since the CRs are in different phase space (no mT2 cut for ZZ, b-jet required for tXZ), a shape error can hide in the SR bins. The proposed closure test would directly validate the transfer. This does not change my acceptance of the paper: the CMS internal review and the fit's nuisance parameters likely cover this risk, but the residual uncertainty is exactly in the claimed evidence level. Therefore the reader's ACCEPT verdict stands unchanged.","tokens_in":32172,"tokens_out":18548,"duration_ms":198926,"concrete_test":"Extract from the CMS analysis nTuples (or the HEPData record if event-level inputs are preserved) the events in the ZZ CR and the tXZ CR, apply the same BDT multiclassifier, and compare the three score distributions in data vs MC. For any bin with more than five events, compute the Poisson p-value for a data/MC discrepancy; then propagate the observed discrepancy as a nuisance shape shift into the simultaneous fit and report the change in the inclusive signal strength and significance. If the shift exceeds the quoted uncertainty, the central claim is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (inclusive signal strength 1.03 +0.31/-0.28, significance 4.5 sigma) is extracted from a fit to bins in the three BDT scores. The dominant backgrounds (ZZ, ttZ+tWZ) are normalized using control regions that are not binned in the BDT scores; their extrapolation into the signal-region bins is taken entirely from MC. The CRs use different phase space (no mT2 cut for ZZ; b-tagged jet required for tXZ), and the BDT output distributions for those backgrounds in the SR are not validated against data anywhere in the letter. If the MC mis-models any of the 27 BDT input variables (m(4l), m(W leptons), pTmiss, angular separations), the background shape across SR bins can be biased without changing the CR yields, shifting the fitted signal strength and significance. This is the most load-bearing condition because the quoted significance is only 1.3 sigma above the expected 5.0, and the Run 3 observed excess (3.8 vs 2.5 expected) is the new physics claim. The reader's 'true probabilities' framing is slightly too strong; the scores need not be calibrated probabilities, but their modeled distributions must match data.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports a measurement of WWZ and ZH production in the four-lepton final state using 138 fb^-1 at 13 TeV and 62 fb^-1 at 13.6 TeV. A BDT multiclassifier separates nonresonant WWZ, ZH (H->WW), and background, and a simultaneous binned likelihood fit is performed to signal regions binned in the three BDT scores along with dedicated control regions for ZZ and ttZ/tWZ. The inclusive WWZ+ZH signal strength is measured to be 1.03 +0.31/-0.28 with an observed (expected) significance of 4.5 (5.0) standard deviations. The WWZ and ZH processes are measured separately for the first time. The Run 3 data alone give an observed (expected) significance of 3.8 (2.5) standard deviations, which is presented as the first evidence for triboson production at 13.6 TeV. The results are consistent with SM predictions within uncertainties.","tokens_in":32431,"tokens_out":6997,"duration_ms":68786,"significance":"If correct, this is the most precise WWZ cross-section measurement to date and the first simultaneous separation of nonresonant WWZ from ZH production. The analysis is mature in its use of NLO MC, dedicated control regions, a full systematic treatment, and a well-documented statistical framework. The systematic uncertainties are much smaller than the statistical ones, and the result is consistent with the SM. The central claim, however, rests on the MC-modeled BDT score distributions for the dominant backgrounds, which are constrained only in normalization by the control regions; this point needs explicit validation before the significance claim can be fully accepted.","major_comments":[{"comment":"The ZZ and tXZ control regions are defined without binning in the BDT scores, so the fit constrains only the integrated normalizations of these backgrounds while their shapes across the signal-region BDT bins are taken entirely from MC simulation. The paper does not provide a data/MC closure test of the BDT output distributions in a background-enriched region, nor does it include an explicit shape systematic for the BDT templates. Since the observed (expected) combined significance is 4.5 (5.0) sigma and the Run 3 excess is 3.8 sigma against an expected 2.5 sigma, the result is sensitive to a possible bias in the background shape. The authors should add a BDT-score-binned data/MC validation in a control region, assign a shape uncertainty to the BDT templates, or demonstrate the robustness of the fitted signal strengths and significances under a reweighting of the BDT distributions.","section":"Signal regions and control regions (page 4)"}],"minor_comments":[{"comment":"The statement that the softmax-transformed scores 'can be interpreted as probabilities' is imprecise; they are not calibrated posterior probabilities, though calibration is not required for the analysis. Suggest rewording to 'relative scores that sum to one'.","section":"Multiclassifier description (page 4)"},{"comment":"Reference [43] contains an internal inconsistency: 'ACAT 2017' with dates 'April 23-27, 2007' and location 'Amsterdam'. The workshop was ACAT 2007; please correct the year in the workshop name.","section":"References"},{"comment":"The caption would benefit from a brief definition of the 'WWZ-like' and 'ZH-like' bin labels directly in the caption, rather than only in the main text, to make the figure self-contained.","section":"Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The analysis is sound and the result is of high interest for PRL. The main concern is the lack of explicit validation of the BDT score shapes for the dominant backgrounds, which is load-bearing for the significance claim. This is a fixable issue: adding a control-region closure test or a shape systematic would address it. The Run 2 WWZ-only signal strength of 0.52 with large uncertainty and the Run 3 inclusive 1.74 with 3.8 sigma observed versus 2.5 expected are worth watching, but they are consistent with statistical fluctuations given the uncertainties."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a mature CMS precision measurement. The genuinely new parts: the first simultaneous separation of nonresonant WWZ from ZH (H->WW) via a BDT multiclassifier, the first evidence for triboson production at 13.6 TeV, and the most precise WWZ cross-section measurement to date. The results are consistent with the SM: combined signal strength 1.03 +0.31/-0.28, observed significance 4.5 vs expected 5.0.\n\nThe analysis is well executed. The event selection is clear, the leading backgrounds (ZZ and ttZ/tWZ) get dedicated control regions, and the systematic treatment is thorough. The statistical uncertainties dominate, so the result does not depend on delicate systematics. The paper also stays honest: it reports the Run 3 evidence as 3.8 sigma observed against 2.5 expected, and the HEPData record makes the numbers usable.\n\nThe stress-test concern about BDT score shapes is legitimate but I think it is a minor omission rather than a flaw. The control regions are not binned in the BDT scores, so the shapes of the background scores in the signal region are taken from MC. A closure test of those shapes in data would have been nice to see. However, the fit includes the usual MC-shape uncertainties, and the observed significances are not so close to the threshold that a small shape bias would flip the conclusion. The Run 2 and Run 3 significances sit on opposite sides of their expectations (2.9 vs 4.4 and 3.8 vs 2.5), but that is well within statistical fluctuations for the event counts involved.\n\nThis paper deserves a serious referee. It is the kind of result that will be cited as the reference for triboson production for the next few years. I would cite it in my own work. For a reading group I'd rate it maybe: it's a good case study in modern LHC analysis technique, but not conceptually deep enough to be required reading.\n\nRecommendation: send to peer review.","headline":"A mature CMS precision measurement that delivers the first WWZ/ZH separation and first 13.6 TeV triboson evidence; the main soft spot is the lack of an explicit data/MC closure test of BDT score shapes, but that is a minor omission, not a fatal flaw.","tokens_in":32988,"tokens_out":2979,"would_cite":true,"duration_ms":30344,"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 measures WWZ and ZH production in four-lepton final states at 13 and 13.6 TeV, reporting the most precise WWZ measurement to date and the first evidence for triboson production at 13.6 TeV, with an inclusive signal strength of…","keywords":["WWZ production","triboson production","ZH production","four-lepton final state","boosted decision tree","signal strength","standard model"],"falsifier":"Inspect the multiclassifier output scores in background-dominated control regions: if the data distributions of the three scores do not match the simulated prediction within uncertainties, the calibration of the scores as probabilities is invalidated.","tokens_in":31972,"feed_emoji":"⚛️","tokens_out":5855,"duration_ms":54942,"temperature":0.7,"pith_summary":"This paper measures how often proton-proton collisions produce a W boson pair together with a Z boson, a rare process that tests the standard model's self-interactions of gauge bosons. It reports the most precise measurement of WWZ production to date and the first evidence for this triboson process at the new 13.6 TeV collision energy. For the first time, it separates the nonresonant WWZ process from ZH production, where a Higgs boson decays to two W bosons, in a single simultaneous fit. The measured inclusive signal strength is 1.03 +0.31/-0.28 times the standard-model prediction, with an observed significance of 4.5 standard deviations (5.0 expected).","feed_headline":"Most precise WWZ measurement yet, first at 13.6 TeV","feed_subtitle":"A combined fit gives a signal strength of 1.03 with 4.5-sigma significance.","key_machinery":"The key machinery is a boosted decision tree multiclassifier, trained on simulated events, that assigns each four-lepton event three scores (WWZ, ZH, background) normalized by a softmax function to sum to one. Events are binned in these scores to build signal regions, and a simultaneous maximum-likelihood fit extracts the two signal strengths together with background normalizations, with control regions constraining the dominant ZZ and ttZ/tWZ backgrounds.","core_discovery":"The central claim is that the combined WWZ and ZH production rate in the four-lepton final state is consistent with the standard model, with signal strength 1.03 +0.31/-0.28 and observed significance 4.5 standard deviations (5.0 expected). This constitutes the most precise measurement of WWZ production to date and the first evidence for triboson production at 13.6 TeV. The measurement also establishes the first simultaneous extraction of nonresonant WWZ and ZH (H to WW) contributions from the same data, using a multiclassifier that assigns each event a probability for each of the two signals and for background.","pith_inferences":["If the multiclassifier separation holds as data accumulate, WWZ alone can be used to constrain anomalous quartic gauge couplings without contamination from the Higgs-mediated ZH diagram.","The same multiclassifier approach could be applied to other triboson final states (e.g., WZZ or ZZZ) to separate resonant and nonresonant contributions.","A direct test of the BDT calibration would be to repeat the extraction using a cut-based signal region; agreement would confirm the scores' probability interpretation, while disagreement would flag simulation mismodeling.","The larger Run 3 signal strength (1.74 +0.71/-0.60) compared with Run 2 (0.75 +0.34/-0.29) is consistent with the standard model within uncertainties, but the two eras' central values differ by more than one standard deviation; a dedicated study of run-dependent acceptance or background modeling would clarify whether this is statistical."],"forward_implications":["The measured inclusive signal strength of 1.03 +0.31/-0.28 makes this the most precise WWZ measurement to date.","The 3.8-sigma observed significance at 13.6 TeV provides the first evidence for triboson production at that energy.","The first simultaneous separation of nonresonant WWZ from ZH allows each process to be compared individually with the standard model; the two are consistent within their current uncertainties.","Because statistical uncertainties dominate (systematics add less than 5% to the total uncertainty), more data will directly shrink the measurement's uncertainty.","The separation of ZH from nonresonant WWZ is robust: fixing the ZH rate to its measured value changes the WWZ signal strength by less than 1%.","The inclusive signal strength across both energies is consistent with the standard-model prediction of 1.0 within the quoted uncertainties."],"supporting_citations":[{"why":"Prior observation of three-gauge-boson production that this analysis extends.","marker":"[3]"},{"why":"Independent observation of VVZ production that motivates the current study.","marker":"[5]"},{"why":"Previous measurement of ZH signal strength used for comparison and for the cross-check that fixes the ZH rate.","marker":"[6]"},{"why":"Next-to-leading-order cross section calculation that provides the standard-model denominator for WWZ.","marker":"[8]"},{"why":"Recommended Higgs-boson production cross sections at 13.6 TeV used for the ZH prediction.","marker":"[36]"},{"why":"Software toolkit used to train the boosted decision tree multiclassifier.","marker":"[43]"},{"why":"Statistical analysis tool used for the simultaneous maximum-likelihood fit.","marker":"[44]"}],"fun_headline_variants":["First evidence for triboson production at 13.6 TeV","CMS: WWZ/ZH signal strength 1.03, 4.5 sigma","WWZ/ZH production matches standard model at 13.6 TeV","First simultaneous extraction of WWZ and ZH signals","Most precise WWZ measurement, 4.5-sigma evidence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The multiclassifier's three scores are treated as true probabilities of WWZ, ZH, and background for real events, which requires the simulated training samples to model the 27 kinematic input variables accurately; if the simulation is wrong, the separated signal strengths and the claimed WWZ-ZH distinction could be biased.","fun_headline_variants_meta":{"raw":{"variants":["First evidence for triboson production at 13.6 TeV","CMS: WWZ/ZH signal strength 1.03, 4.5 sigma","WWZ/ZH production matches standard model at 13.6 TeV","First simultaneous extraction of WWZ and ZH signals","Most precise WWZ measurement, 4.5-sigma evidence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000723,"raw_usage":{"total_tokens":3260,"prompt_tokens":980,"completion_tokens":2280,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":2185}},"tokens_in":596,"tokens_out":2280,"duration_ms":16200,"temperature":1.0,"reasoning_tokens":2185,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:53:13.787909+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inspect the multiclassifier output scores in background-dominated control regions: if the data distributions of the three scores do not match the simulated prediction within uncertainties, the calibration of the scores as probabilities is invalidated.","supporting_citations":[],"review_version":1}