{"id":"39607aef-856e-4a61-856d-610647ad4d85","arxiv_id":"2504.12141","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"XGBoost applied to simulated same-sign and opposite-sign dilepton events from WR decays in the inverse seesaw left-right model gives projected heavy neutrino mass reaches up to 17.1 and 19.5 TeV at a 100 TeV collider.","lead":"Using machine learning (XGBoost) on simulated collider events, this paper predicts how far future hadron colliders could search for heavy Majorana neutrinos in the inverse seesaw extension of the left-right symmetric model. A smart generalist might read it to see how far next-generation proton colliders (14, 27, and 100 TeV) could push searches for lepton-number violation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed 19.5 TeV SS reach is not supported by the signal cross-section scale: near m_N = m_WR the WR->ℓN phase space suppresses the rate by ~10^-3, leaving O(1) signal events at 30 ab^-1.","rationale":"The reader's weakest assumption was the near-degeneracy/interference condition, which the authors explicitly acknowledge in the conclusion. My stress-test identifies a different and, in my view, more load-bearing concern: the claimed high-mass reach may be numerically impossible given the phase-space suppression of WR -> ℓ N near m_N = m_WR. The reported cross sections (Tables 3/4) stop at 2 TeV, so the endpoint claims cannot be checked from the paper. If the generated high-mass signal yields are as estimated, the significance formula gives Z << 2. This is not a disagreement with the model or with ML methodology; it is a request to show the event counts behind the headline numbers. I therefore keep the verdict as CONDITIONAL: acceptance should require a demonstration of the endpoint sensitivity with explicit N_S and N_B values, or the headline mass limits should be revised downward. The proposed test is a single re-simulation with the authors' own setup, which would settle whether the 17.1/19.5 TeV reaches are real.","tokens_in":24999,"tokens_out":27753,"duration_ms":266371,"concrete_test":"Regenerate pp -> W_R -> ℓ N -> ℓℓ jj at sqrt(s) = 100 TeV for m_N = 17.1 and 19.5 TeV with m_WR = 20 TeV using the same MadGraph/DELPHES setup, apply the Eq. (3.1) cuts and the Table 6 ML thresholds, and compute N_S and Z = N_S/√(N_S+N_B) for L = 30 ab^-1. Also compare the generated cross sections with the narrow-width expectation σ(pp->W_R) × BR(W_R->ℓN) using the (1-x)^2(1+x/2) phase-space factor. If N_S < 5 at either mass, the headline 2σ reach is not supported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The headline numbers (17.1 TeV OS, 19.5 TeV SS at 100 TeV) sit at the kinematic endpoint m_N = m_WR = 20 TeV. The paper reports signal cross sections only at m_N = 1.5 and 2 TeV (Tables 3 and 4), not at the endpoint. For a fixed WR mass, the two-body width WR -> ℓ N is phase-space suppressed by P(x) = (1-x)^2 (1+x/2) with x = m_N^2/m_WR^2. From m_N = 2 TeV (P ≈ 0.985) to m_N = 19.5 TeV (P ≈ 0.0036), the expected signal rate falls by a factor ≈ 3.7e-3. Taking the SS eµ signal at m_N = 2 TeV as ~3e-6 pb (Table 4) and allowing a factor 2-3 acceptance gain for less-collimated jets, the m_N = 19.5 TeV signal yield at L = 30 ab^-1 is N_S ≲ 1. The surviving SM background after the Table 6 ML cuts is O(100) events (e.g., W±W±jj at 9.3e-3 pb times 0.07% efficiency, plus W±Zjj), so Z = N_S/√(N_S+N_B) requires N_S ≈ 20 for Z = 2. With N_S < 1, the SS claim fails by more than an order of magnitude. The OS 17.1 TeV point has P ≈ 0.099 relative to m_N = 2 TeV, giving N_S ~ 45 before the ML threshold; the OS background after Table 6 cuts is ~10^4 events, so Z ≈ 0.45 unless a much stricter threshold is demonstrated. The paper provides no event yields or significance values at these endpoint masses, so the sensitivity contours in Figs. 29 cannot be checked. This is a concrete numerical gap, separate from the acknowledged interference condition.","agreement_with_reader":"disagree"},"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-16T12:38:19.717934+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}