{"id":"e5996c82-3fa1-4f50-9686-88409196e49d","arxiv_id":"2511.08212","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A displaced-vertex search for compressed-spectrum long-lived particles excludes top squarks up to about 1100 GeV and wino-like neutralinos up to about 550 GeV, the strongest limits on these benchmark models.","lead":"Using 100 fb^-1 of 2017-2018 LHC data, CMS searched for long-lived particles decaying a few centimeters from the collision point, focusing on the hard-to-reach 'compressed' regime with a small mass gap and soft decay products. No signal was found; the search sets the strongest limits to date on two supersymmetric coannihilation benchmark models.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Alpha_p>0.2 is said to be optimized on SR event yields; if those were observed yields, the background-only p-value and 95% CL exclusions are overfitted. The paper needs to show the selection was frozen before unblinding.","rationale":"The reader's verdict is CONDITIONAL and already flags the alpha_p threshold issue among its two required points, so my read does not move the verdict. However, the reader's weakest_assumption field points to the transfer-factor pT_miss extrapolation, whereas I judge the alpha_p threshold optimization to be the single more load-bearing concern: if the optimization used observed SR yields, the statistical coverage of the exclusion claim itself is compromised, not just the background estimate. The transfer-factor extrapolation is a legitimate secondary concern, and the reader's request for a pT_miss-dependence check remains worthwhile. I therefore mark agreement as partial and leave the verdict at CONDITIONAL: the paper should resolve the alpha_p optimization question before the exclusion ranges are taken at face value.","tokens_in":41377,"tokens_out":8607,"duration_ms":102105,"concrete_test":"Request the internal analysis log/commit showing that the alpha_p>0.2 requirement and the SR definitions were fixed before any observed SR event counts were available. Then perform a blinded re-optimization: choose alpha_p in {0.1, 0.2, 0.3} using only signal simulation and control-region data, and recompute the CLs limits. If the observed 95% CL exclusion ranges (stop 400–1100 GeV; neutralino 220–550 GeV) shift materially across thresholds, the reported ranges are not robust; if the threshold was set after unblinding, the analysis should be redone with a pre-fixed threshold.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4 states: \"This threshold is optimized to maximize the search sensitivity based on the event yields in the SRs and is not directly derived from Fig. 4.\" If \"event yields in the SRs\" means the observed yields used in the final background-only fit, this is a circular step: the SR-defining alpha_p cut would be chosen using the same data that are then tested for the background-only hypothesis. The central claim—observed 95% CL exclusions of stop masses below 400–1100 GeV and wino-like neutralino masses below 220–550 GeV—depends on the nominal coverage of the CLs procedure. A threshold tuned to optimize sensitivity on observed SR yields can inflate the apparent exclusion and make the quoted ranges not robust. The ambiguity is not resolved by the text: it says only that the threshold is \"not directly derived from Fig. 4\" and provides no optimization record. Even if the intended meaning was expected yields from simulation, the wording should be corrected. This concern is more load-bearing than the transfer-factor extrapolation identified by the reader: the transfer-factor method has at least an explicit validation in orthogonal planes, whereas the alpha_p optimization has no such check.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a CMS search for long-lived particles (LLPs) in proton-proton collisions at 13 TeV using 100 fb^-1 of 2017-2018 data, targeting displaced vertices with low-momentum tracks, large missing transverse momentum, and an ISR jet. It interprets the search in two compressed-spectrum SUSY coannihilation benchmarks: a stop NLSP and a bino-wino NLSP with mass splittings of 12-25 GeV. The background is estimated entirely from data via a transfer-factor method that relates track-count planes and applies S_vtx_xy fractions measured at low pT_miss to the high-pT_miss signal regions. After a maximum-likelihood fit under the background-only hypothesis, the observed yields are consistent with the prediction (reported global p-value 0.5), and 95% CL limits are set excluding stop masses below 400-1100 GeV and bino-wino neutralino masses below 220-550 GeV depending on model parameters. The paper claims the most stringent limits to date for these two models and the first LHC sensitivity to such compressed LLP signatures using displaced vertices.","tokens_in":41476,"tokens_out":3892,"duration_ms":41963,"significance":"If the result holds, it closes a previously unexplored corner of LLP parameter space: compressed coannihilation scenarios with small mass splittings that have limited sensitivity in earlier displaced-vertex searches. The analysis is careful in several respects: the background estimate is data-driven with explicit transfer-factor equations; the SR-excluded material map is derived from data; the validation is performed in orthogonal planes; per-region discrepancies are reported transparently; and the tracking/vertexing efficiency is calibrated with K0S decays. These are substantive strengths. The main caveat is the optimization of the alpha_p selection: the text states it was optimized on 'event yields in the SRs' without clarifying whether those yields were observed or expected, and this ambiguity is load-bearing for the quoted exclusions. The transfer-factor closure also deserves quantitative support. With those points resolved, the result would meet the standard for a high-energy physics search paper.","major_comments":[{"comment":"The text states: 'This threshold is optimized to maximize the search sensitivity based on the event yields in the SRs and is not directly derived from Fig. 4.' If 'event yields in the SRs' means the observed yields that are later fit under the background-only hypothesis and used for CLs limits, then the alpha_p > 0.2 requirement is a selection on the data being tested, which can bias the background-only p-value and inflate the quoted exclusion ranges in Figs. 10-11. The statement that the threshold is 'not directly derived from Fig. 4' does not resolve this. Please state explicitly whether the optimization used expected signal and background from simulation, blinded control regions, or observed SR yields; if the latter, the selection is circular and the limits should be rederived with a frozen/blinded threshold. The ambiguity must be removed before the central exclusion claim can be asse","section":"Section 4, alpha_p optimization"},{"comment":"The transfer factors are measured in the 200 < pT_miss < 400 GeV range and applied at pT_miss > 700 GeV, relying on the asserted independence of pT_miss and S_vtx_xy and on track-count ratios being pT_miss-independent. The validation in Fig. 7 is described only qualitatively ('For most of the regions... agree within statistical uncertainties'), and the paper does not report a quantitative closure test or a validation p-value. The observed departures in the search planes (tight B/D: 3 and 5 observed vs. 7.7 and 9.2 predicted; medium B: 98 vs. 78.8) are acknowledged but not propagated into the limit coverage. Please provide a quantitative pT_miss-dependence check for the transfer factors, a validation p-value for the orthogonal planes, and an explicit statement of how the observed deviations affect the CLs expected/observed limits.","section":"Section 5, Eqs. (3)-(11) and Fig. 7"},{"comment":"The K0S-based efficiency systematic is derived after normalizing simulation to data at small Lxy, so the 10-11% envelope covers the Lxy-shape dependence but by construction absorbs the absolute reconstruction efficiency. The paper does not separately quantify the absolute normalization uncertainty in the signal efficiency. Since the quoted cross-section limits scale directly with signal efficiency, please state what part of the absolute efficiency is constrained by simulation only and what additional uncertainty (if any) is assigned to that absolute normalization.","section":"Section 6, K0S calibration"}],"minor_comments":[{"comment":"The decay width is given in units of cm^-1, while cτ is quoted in length units. Please clarify the natural-unit convention and the relation cτ = B/Γ, including factors of c if needed.","section":"Eq. (1)"},{"comment":"The statement that pT_miss and S_vtx_xy are 'statistically independent' and that S_vtx_xy shapes are consistent across pT_miss ranges is not backed by a figure or a numerical test. Please include the relevant distribution comparison or a quantitative test statistic.","section":"Section 5, independence statement"},{"comment":"The global p-value of 0.5 is quoted without specifying which regions enter the test statistic or whether nuisance parameters are profiled. Please define the statistic and the region set used for this p-value.","section":"Section 7, p-value"},{"comment":"The text notes that data vertices tend to have larger alpha_p than simulation because of pileup mismodeling. Since alpha_p is explicitly used in the optimization, please comment on how this data/simulation discrepancy affects the choice of the alpha_p threshold and the robustness of the assumed background model.","section":"Section 4, vertex alpha_p discrepancy"}],"recommendation":"major_revision","confidential_remarks":"The paper is a well-executed CMS search with a transparent data-driven background estimate, but the alpha_p optimization statement is a genuine blinding concern. I would ask the collaboration to provide the optimization record or reword the text to make clear that the threshold was fixed before looking at the SR yields. The transfer-factor closure issue is less severe but should be quantified. Recommendation: major_revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a legitimate new result. The genuinely new ingredient is using well-measured low-momentum tracks (pT > 0.5 GeV) with a retuned inclusive vertex finder to get sensitivity to displaced decays with mass gaps 12-25 GeV, where the earlier ATLAS/CMS DV searches lost sensitivity. They demonstrate the efficiency gain, and the resulting 95% CL exclusions on stop and wino-like neutralino NLSPs are the strongest in those benchmark models. The analysis is careful: data-driven transfer-factor background, orthogonal validation planes, material map excluding signal regions, and a 10-11% vertex-efficiency systematic from KS decays. The observed yields are consistent with background (global p=0.5), and the paper is transparent about the downward fluctuations in tight B/D that strengthen the observed limits relative to expected.\n\nThe main soft spot is the alpha_p threshold. Section 4 says it is 'optimized to maximize the search sensitivity based on the event yields in the SRs.' That phrasing is ambiguous. If the SR yields were the observed yields, the threshold is tuned on the data that are then tested, and the quoted exclusions would be overfitted. The paper needs to state clearly that the alpha_p threshold (and the SR-defining pT_miss and S_vtx_xy cuts) were fixed before looking at the yields in the final signal regions. This is more important than the transfer-factor extrapolation.\n\nThe transfer-factor method itself is reasonable but has the usual extrapolation caveat: track-count ratios and S_vtx_xy shapes are measured at 200-400 GeV and applied above 700 GeV. They argue independence and show validation planes, but the validation is not perfect (tight B/D low, medium B high). The deviations are within fluctuations, so I would not call it a flaw, but I'd want to see a closure test at intermediate pT_miss or a note about the size of the shifts. The KS calibration normalizes away the absolute vertex efficiency at small Lxy and is measured on alpha_p <0.2 vertices while the search uses alpha_p >0.2; that's a minor concern, and the b-jet cross-check helps.\n\nThis paper is for experimental colleagues in LLP/SUSY searches and for theorists mapping coannihilation parameter space. If the alpha_p wording is fixed, it is a solid paper. I'd send it to a serious referee; it closes a genuinely unprobed corner and the methodology is largely standard.","headline":"New compressed LLP exclusion from CMS using low-pT displaced tracks; solid analysis, but the alpha_p optimization wording must be clarified before the limits are fully credible.","tokens_in":42183,"tokens_out":3739,"would_cite":true,"duration_ms":38127,"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":"A search for long-lived particles in compressed supersymmetry excludes top squarks to 1100 GeV and wino-like neutralinos to 550 GeV.","keywords":["long-lived particles","displaced vertices","low-momentum tracks","top squark","wino-like neutralino","compressed supersymmetry","missing transverse momentum","transfer factor method"],"falsifier":"Measure, in a dedicated high-pTmiss control sample with zero good tracks, the fraction of events that would have S_vtx_xy > 20 and compare it with the low-pTmiss proxy used in the paper; a statistically significant upward trend with pTmiss would mean the tight-plane background predictions are biased low, and the quoted exclusions would shrink.","tokens_in":41036,"feed_emoji":"⚛️","tokens_out":3155,"duration_ms":32092,"temperature":0.7,"pith_summary":"This paper reports a search for long-lived particles that decay inside the detector to a displaced vertex with low-momentum tracks, accompanied by large missing transverse momentum and an initial-state radiation jet. The analysis targets supersymmetric coannihilation scenarios where the mass difference between the long-lived particle and the invisible lightest particle is only 12–25 GeV, a region previous searches had difficulty reaching. Using 100 fb−1 of 13 TeV proton-proton collision data, the observed event yields agree with a data-driven background prediction, and the search excludes top squarks with masses below 400–1100 GeV and wino-like neutralinos below 220–550 GeV, depending on model parameters. The paper also introduces a transfer-factor method that predicts backgrounds in multiple signal regions purely from data, and reports the most stringent limits to date for these two models.","feed_headline":"Compressed stop and wino masses excluded to 1100 and 550 GeV","feed_subtitle":"CMS uses displaced vertices from low-momentum tracks to close a gap in long-lived particle searches.","key_machinery":"The key mechanism is a purely data-driven background estimation based on transfer factors. Events are divided into four 'planes' by the number of good tracks in the displaced vertex (0, 1, 2, or ≥3), and within each plane into regions by missing transverse momentum (pTmiss > 700 GeV) and vertex displacement significance (S_vtx_xy > 20). Transfer factors—ratios of event counts between planes—are measured at low pTmiss (200–400 GeV) and applied to high-pTmiss control regions to predict backgrounds in the signal regions; the S_vtx_xy distribution is taken from low-pTmiss data as a proxy. The signal reconstruction relies on the inclusive vertex finder tuned to accept low-momentum tracks with lar","core_discovery":"The paper's central claim is that long-lived particles with small mass splittings can be efficiently detected by reconstructing displaced vertices from tracks with transverse momentum as low as 0.5 GeV, even when the decay products are soft. In the top squark NLSP model, the search excludes top squark masses less than 400–1100 GeV, and for the bino-wino NLSP model it excludes wino-like neutralino masses less than 220–550 GeV, with the exact bound depending on the mass difference, the mass, and the lifetime of the long-lived particle. The observed limits are the strongest reported for these two coannihilation scenarios.","pith_inferences":["The same technique could be applied to non-SUSY hidden-sector models that produce compressed spectra, such as dark photon or exotic Higgs decays, where final-state particles are soft.","A testable extension: measure the S_vtx_xy and track-count ratios as a function of pTmiss in the 0-good-track control plane with the full Run 3 dataset; any rising trend would indicate the transfer-factor prediction underestimates high-pTmiss backgrounds.","The 10–11% track/vertex reconstruction systematic, inherited from a K0S control sample, might be reducible with better detector simulation, which would sharpen the exclusions.","The low-pTmiss transfer factors assume no correlation between pTmiss and vertex multiplicity; if pileup conditions differ in later data-taking, the extrapolation should be revalidated."],"forward_implications":["The compressed stop coannihilation parameter space with Δm = 12–25 GeV is closed for stop masses up to 400–1100 GeV, depending on the branching fraction.","The bino-wino coannihilation scenario with wino-like neutralino masses below 220–550 GeV is excluded for lifetimes around 0.2–200 mm.","The transfer-factor background method can be applied to future LLP searches that need multiple exclusive signal regions with limited simulation.","The result provides a target for Run 3 and future collider searches: extending the same low-momentum displaced-vertex strategy to higher masses and smaller mass gaps."],"fun_headline_variants":["CMS excludes compressed stops up to 1100 GeV with soft-track vertices","Low-momentum tracks reveal long-lived particles in CMS data","CMS finds strongest limits for compressed stop and wino NLSPs","Displaced vertices with soft tracks close long-lived particle gap"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The background prediction relies on the assumption that the ratio of event counts between planes with different numbers of good tracks, and the fraction of events with high vertex displacement significance, are the same at low missing transverse momentum (200–400 GeV) as at high missing transverse momentum (>700 GeV).","fun_headline_variants_meta":{"raw":{"variants":["CMS excludes compressed stops up to 1100 GeV with soft-track vertices","Low-momentum tracks reveal long-lived particles in CMS data","CMS finds strongest limits for compressed stop and wino NLSPs","Displaced vertices with soft tracks close long-lived particle gap"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000537,"raw_usage":{"total_tokens":2447,"prompt_tokens":805,"completion_tokens":1642,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":1578}},"tokens_in":549,"tokens_out":1642,"duration_ms":11096,"temperature":1.0,"reasoning_tokens":1578,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T22:54:36.821361+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure, in a dedicated high-pTmiss control sample with zero good tracks, the fraction of events that would have S_vtx_xy > 20 and compare it with the low-pTmiss proxy used in the paper; a statistically significant upward trend with pTmiss would mean the tight-plane background predictions are biased low, and the quoted exclusions would shrink.","supporting_citations":[],"review_version":1}