{"id":"7f6a0a59-2c97-4d95-be38-afa8a39fd9fc","arxiv_id":"2412.14937","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A muon-removal preprocessing step restores tau-hadron identification efficiency in boosted semileptonic tau-pair events to the isolated-tau level, as shown in simulation and validated with Z->tau tau data.","lead":"ATLAS physicists developed a way to clean up hadronic tau-lepton signals by deleting the overlapping muon's track and energy deposits before running tau identification, recovering nearly full efficiency for highly boosted tau pairs. The method is validated with Z boson decays in 140 fb^-1 of LHC data, and could sharpen future searches for new heavy particles that decay into pairs of tau leptons.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Untested merged topo-cluster failure mode: when the muon and tau_had deposits share a single calorimeter cluster, the energy-compatibility removal criterion leaves the cluster intact, so the claimed isolated-level efficiency recovery is not quantified for this category.","rationale":"The reader's weakest_assumption identifies the same load-bearing premise: unambiguous identification and removal of the muon's track and clusters without removing or leaving tau_had energy. My concern sharpens this to a specific, testable failure mode: the energy-compatibility criterion in Section 4.1 fails exactly when muon and tau_had energy are merged into a single topological cluster, which is likely in the boosted regime the paper targets. The aggregate efficiency plots and the 12%-precision Z validation do not resolve this category-specific risk, so the central claim remains conditional. I agree with the reader's CONDITIONAL verdict and do not recommend changing it. The proposed test isolates the merged-cluster category and would settle whether the concern lands. The paper's internal inconsistency about 3-prong background rejection ('no degradation' in the conclusion vs. 'decreases slightly' in Section 4.2) is noted but is not the most load-bearing issue for the central efficiency-recovery claim.","tokens_in":52620,"tokens_out":3892,"duration_ms":35623,"concrete_test":"In the G->HH->4tau simulation, use generator-level truth to split dR^gen,vis < 0.4 seed jets into two categories: (a) the muon truth deposits and the tau_had-vis truth deposits are contained in distinct topological calorimeter clusters, and (b) they share at least one cluster. Recompute the Fig. 3 reconstruction-plus-TauID efficiency curves separately for the two categories before and after muon removal. If category (b) shows a residual efficiency deficit larger than ~5% (relative to the isolated-tau baseline) at any working point, the 'efficiency recovered to isolated level' claim requires qualification; if no deficit remains, the merged-cluster concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the muon-removal step in Section 4.1: the ID track and calorimeter clusters associated with the muon are removed from the tau seed jet, but a calorimeter cluster is removed only if its energy is compatible with the expected muon energy loss. For highly boosted tau_mu tau_had pairs with dR < 0.4, the muon and the tau_had shower frequently overlap geometrically, so their energy deposits can be merged into the same topological cluster. Such a cluster has energy far above the few-GeV muon MIP loss, and the algorithm will therefore leave it in the seed jet, preserving muon contamination in the TauID RNN inputs. Conversely, a MIP-like cluster that also contains a soft tau_had component would be removed, carving away signal. The paper reports only the aggregate 'MuonRM' efficiency (~95%) and the overall efficiency recovery; it does not report the fraction of signal seed jets in which the muon calorimetric deposit is left unremoved because of merging, nor the residual cluster energy after cleaning. The Z->tau_mu tau_had validation has a 12% total uncertainty dominated by a 10% MC cross-section correction, so it cannot discriminate a 10-20% efficiency mis-modelling confined to the merged-cluster category. Thus the 'isolated-level efficiency' claim depends on the untested assumption that merged clusters are either rare or harmless.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes a new procedure, called τμ_had, for reconstructing hadronically decaying tau leptons in highly boosted τμτhad pairs, where the muon and the tau-hadron visible decay products overlap within a single anti-kt R=0.4 seed jet. The method removes the muon's inner-detector track and calorimeter clusters from the seed jet, when their energy is compatible with muon energy loss, and then re-runs the standard tau reconstruction and TauID RNN on the cleaned jet. Efficiency studies in simulated G→HH→4τ events show that the reconstruction and identification efficiencies are recovered to near the isolated-tau level for all TauID working points. The method is validated using Z→τμτhad events in 140 fb−1 of 13 TeV ATLAS data; the ratio of observed to predicted yields in the signal region is 0.97 ± 0.12, with a total uncertainty of 12%.","tokens_in":52834,"tokens_out":12089,"duration_ms":98477,"significance":"The paper addresses a real bottleneck for BSM searches with boosted tau pairs. Its strengths are the direct efficiency evaluation with generator-level truth, the ROC-curve comparison with the standard algorithm, and the data validation in a Standard Model process. If the claimed efficiency recovery holds, the method offers a substantial gain in signal sensitivity for channels such as G→HH→4τ. The validation is fit-for-purpose as a consistency check, though its precision is limited by the 10% Z+jets cross-section correction.","major_comments":[{"comment":"The cluster-removal step removes a calorimeter cluster only if its energy is compatible with the expected muon energy loss. For ΔR < 0.4, the muon and the tau-hadron shower can be merged into a single topological cluster whose energy is far above the muon's MIP loss, in which case the algorithm leaves the cluster in the seed jet; the converse failure (removing a cluster that also contains soft tau-hadron energy) is also possible. The paper does not report how often either case occurs or the residual energy after cleaning, yet the abstract attributes the efficiency recovery to the removal of muon information. Please add a quantitative assessment of the merged-cluster category—e.g., the fraction of signal seed jets with an unremoved muon cluster and the mean residual cluster energy—or otherwise demonstrate that this category does not compromise the claimed isolated-level efficiency.","section":"4.1"},{"comment":"The reported data/prediction ratio of 0.97 ± 0.12 is dominated by the 10% Z+jets cross-section correction applied to the simulation, so the validation has limited power to expose a 10-20% efficiency mis-modelling localized in the merged-cluster category described above. The paper should state this limitation explicitly and, if possible, quantify the validation's sensitivity to the muon-removal efficiency, for instance by comparing the SR and SRstd0 yield ratios.","section":"5.4"}],"minor_comments":[{"comment":"The phrase 'raised to the level expected for an isolated τhad' is stronger than the 95% MuonRM efficiency and the small residual differences visible in Figures 3 and 5; consider using 'nearly' or quoting the residual efficiency difference.","section":"Abstract and 4.2"},{"comment":"The sentence 'The measurement precision for the charge and kinematic properties of the visible τhad system is similarly recovered' should specify that the η and pT resolutions are meant (as shown in Figure 7), since charge mis-assignment is not directly addressed there.","section":"4.2"},{"comment":"The labels 'SRstd0' and 'SRstdtight' use a subscript zero that is not defined in the table caption; a one-line explanation in the caption would improve readability.","section":"Table 2"},{"comment":"When quoting the background subtraction of 223 ± 5 events, please state that the uncertainty is statistical only, consistent with the notation used elsewhere in the table.","section":"5.4"},{"comment":"The y-axis label 'Fraction of events / 0.016' is unusual; consider 'Fraction of events per 0.016' for clarity.","section":"Figure 1"}],"recommendation":"major_revision","confidential_remarks":"This is a competent ATLAS performance paper. The major revision is triggered by the need to quantify the merged-topological-cluster failure mode and to temper the validation claim. With those additions, the paper would be acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the muon-removal preprocessing is a genuinely new, simple, and effective trick, and the paper backs it with careful MC studies and a sensible Z->tau_mu tau_had data validation. The central claim—that removing the muon restores tau_had identification efficiency to the isolated level—holds up, with caveats.\n\nWhat is new: Ref. [10] handled the boosted both-hadronic di-tau case with a BDT; here they instead remove the muon's ID track and calorimeter clusters from the tau seed jet before re-running the standard TauID RNN. The MC efficiency curves (Figures 3-6) show near-complete recovery across working points, for 1- and 3-prong, and the ROC curves show an order-of-magnitude gain. The pile-up and pT stability checks are reassuring. The Z validation is a good choice, and the data/MC agreement in RNN score, pT_col, and collinear mass (Figures 11-13) is genuinely good: 0.97 ± 0.12 in the SR.\n\nSoft spots, in proportion. The headline efficiency recovery is measured in simulation; the Z validation carries a 12% total uncertainty dominated by a 10% external Z+jets cross-section correction, so its power to expose a mis-modeled efficiency confined to a sub-category is limited. The stress-test concern about merged topo-clusters is legitimate: the algorithm removes a cluster only if its energy is compatible with the muon's MIP loss, so a merged muon+tau_had cluster would be left in the seed jet. The paper does not report how often this happens or what residual energy remains. That said, the fact that the overall efficiency recovery in MC is essentially complete suggests such cases are rare or benign in simulation; the residual worry is detector simulation accuracy for cluster merging, which the Z validation can only partially close. Minor: the conclusion says 'no degradation' in background rejection, but their own Figure 8 shows a slight decrease for 3-prong; this is a wording issue.\n\nWho this is for: anyone doing BSM searches with boosted semileptonic tau pairs, especially high-mass resonances decaying to HH -> 4tau. The method is likely to become standard in ATLAS. The paper deserves a serious referee; I'd recommend acceptance with minor revisions, asking the authors to quantify the merged-cluster category and soften the 'no degradation' claim.","headline":"Muon-removal preprocessing is a genuinely useful new trick for boosted semileptonic tau pairs; the central efficiency-recovery claim holds, with a legitimate but non-fatal caveat about merged clusters.","tokens_in":53426,"tokens_out":4013,"would_cite":true,"duration_ms":32182,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["14.60.Fg"],"model":"deepseek-v4-flash","headline":"The ATLAS collaboration claims that removing the muon's track and calorimeter deposits from the seed jet of a highly boosted tau-mu tau-had pair restores hadronic-tau identification efficiency to the isolated-tau level, and validates this…","keywords":["tau lepton reconstruction","boosted tau pairs","muon removal","hadronic tau identification","TauID RNN","Z boson validation","ATLAS detector","LHC"],"falsifier":"Measure the $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ yield ratio in a larger data set (for example the full LHC Run 3) with a total uncertainty below 5%: if the observed-to-predicted ratio moves away from unity, or if the TauID score distribution of muon-removed jets in data disagrees with isolated-tau jets from $Z\\to\\tau\\tau$ events with the same reconstructed transverse momentum, the assumption that the cleaned jet behaves as an isolated tau would be falsified.","tokens_in":52355,"feed_emoji":"⚛️","tokens_out":5608,"duration_ms":46377,"temperature":0.7,"pith_summary":"This paper tries to establish that a simple preprocessing step recovers the sensitivity to highly boosted semileptonic tau pairs, where one tau decays to a muon and two neutrinos and the other to hadrons plus a neutrino. In such events the muon often lands inside the seed jet of the hadronic tau, and the standard tau identification algorithm sees the muon as contamination and fails. The new procedure removes the muon's inner-detector track and calorimeter clusters from the seed jet and then re-runs the standard tau reconstruction and a recurrent-neural-network kaon identifier. In simulation, the identification efficiency returns to the level of an isolated tau for both one-prong and three-prong decays. In data, a validation using $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ events from 140 $\\mathrm{fb}^{-1}$ of 13 TeV collisions finds the observed yield agrees with the prediction within 12%.","feed_headline":"Muon removal restores tau tagging for boosted pairs","feed_subtitle":"ATLAS recovers isolated-tau efficiency at all working points and matches Z->tau_mu tau_had data within 12% uncertainty.","key_machinery":"The central mechanism is the muon-removal re-reconstruction: for each tau seed jet that contains a Medium-working-point muon, the muon's inner-detector track is removed and calorimeter clusters are removed only when their energies match the expected muon energy loss, and then the standard tauhad reconstruction and the TauID recurrent neural network are re-run on the cleaned jet. The muon's minimum-ionising behaviour and isolation-independent reconstruction make the subtraction clean enough that the RNN receives a jet that looks like an isolated tauhad.","core_discovery":"The central claim is that the hadronic tau candidate reconstructed after muon removal, denoted $\\tau^{\\mu\\backslash}_{\\mathrm{had}}$, recovers the tauhad identification efficiency of the standard ATLAS TauID algorithm to the level expected for an isolated tauhad, across all working points, for both one-prong and three-prong decays. Muon removal keeps only the inner-detector track and calorimeter clusters associated with a Medium-working-point muon, and cluster removal is additionally gated on compatibility with the expected muon energy loss, so the seed jet becomes a clean tauhad signature. The validation in $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ events shows a data-to-prediction ratio of $0.97 \\pm 0.12$ in the signal region and roughly three times more signal events than the standard reconstruction selects, confirming that the procedure works on real data.","pith_inferences":["The same removal logic may extend to other overlapping-object cases, such as a hadronic tau overlapping with an electron jet or with tracks from pile-up, though the electron case is harder because electromagnetic showers spread more than a muon's minimum-ionising deposits.","The roughly 90% muon-removal efficiency in the detector region $|\\eta|<0.1$, limited by a gap in the muon spectrometer, implies a residual efficiency loss for boosted tau pairs pointing there; a future detector with more hermetic muon coverage would remove even that loss.","At higher luminosity or with Run 3 data, the same $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ control region could be used to measure the tauhad energy scale for boosted objects, since the muon provides a clean tag of the true tau direction."],"forward_implications":["Signal efficiency for boosted $\\tau_\\mu\\tau_{\\mathrm{had}}$ pairs is restored to the isolated-tau level at all TauID working points, for both one- and three-prong tauhad decays.","The $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ validation selects about three times more signal events than the standard reconstruction, improving the statistical power of searches for boosted tau pairs.","The improved pseudo-rapidity and transverse-momentum resolution of the visible tauhad system (pseudo-rapidity core resolution improved by a factor of 15) makes the collinear mass reconstruction show a clear $Z$-boson peak.","Background rejection against semileptonic heavy-flavour jets, measured on $t\\bar{t}$ events, does not degrade; the ROC curves show an order-of-magnitude gain at fixed signal efficiency.","The method is ready to be used in beyond-the-Standard-Model searches for high-mass resonances decaying to tau pairs, such as $G\\to HH\\to 4\\tau$."],"supporting_citations":[{"why":"Provides the Medium muon working point and establishes that muon reconstruction is independent of isolation, enabling removal of a muon that sits inside the tau seed jet.","marker":"[11]"},{"why":"Defines the TauID recurrent neural network and the working points used to evaluate the identification efficiency before and after muon removal.","marker":"[4]"},{"why":"Supplies the standard tauhad reconstruction and calibration scheme that is re-run after muon removal on the cleaned seed jet.","marker":"[5]"},{"why":"Gives the expected muon energy loss in the calorimeters, used as the gate for removing clusters associated with the muon.","marker":"[43]"},{"why":"Introduces the collinear approximation used to reconstruct the Z-boson mass in the validation analysis.","marker":"[46]"},{"why":"Provides the Randall-Sundrum graviton signal model ($G\\to HH\\to 4\\tau$) used to develop and optimise the muon-removal method in simulation.","marker":"[12]"},{"why":"Generates the $Z(\\to\\tau\\tau)+$jets simulated sample that is the dominant signal contribution in the $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ validation.","marker":"[36]"},{"why":"Supplies the isolated-tauhad reconstruction results against which the improved residuals after muon removal are compared.","marker":"[44]"}],"fun_headline_variants":["Muon removal boosts tau tagging for boosted pairs","ATLAS recovers isolated-tau efficiency via muon removal","Muon removal lifts tauhad ID for boosted tau pairs","Tau reconstruction improved by muon removal in ATLAS","Muon removal revives tau tagging in boosted decays"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the muon's track and calorimeter deposits inside the seed jet can be identified and removed without also removing genuine tau-hadron energy, so that the standard tau-id neural net, trained on isolated taus, works unchanged on the cleaned jet.","fun_headline_variants_meta":{"raw":{"variants":["Muon removal boosts tau tagging for boosted pairs","ATLAS recovers isolated-tau efficiency via muon removal","Muon removal lifts tauhad ID for boosted tau pairs","Tau reconstruction improved by muon removal in ATLAS","Muon removal revives tau tagging in boosted decays"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000729,"raw_usage":{"total_tokens":3292,"prompt_tokens":997,"completion_tokens":2295,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":2216}},"tokens_in":613,"tokens_out":2295,"duration_ms":9602,"temperature":1.0,"reasoning_tokens":2216,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:47:38.949516+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the $Z\\to\\tau_\\mu\\tau_{\\mathrm{had}}$ yield ratio in a larger data set (for example the full LHC Run 3) with a total uncertainty below 5%: if the observed-to-predicted ratio moves away from unity, or if the TauID score distribution of muon-removed jets in data disagrees with isolated-tau jets from $Z\\to\\tau\\tau$ events with the same reconstructed transverse momentum, the assumption that the cleaned jet behaves as an isolated tau would be falsified.","supporting_citations":[{"cited_title":"Ellis, I","cited_arxiv_id":null,"evidence_quote":"Introduces the collinear approximation used to reconstruct the Z-boson mass in the validation analysis."},{"cited_title":"Randall and R","cited_arxiv_id":null,"evidence_quote":"Provides the Randall-Sundrum graviton signal model ($G\\to HH\\to 4\\tau$) used to develop and optimise the muon-removal method in simulation."}],"review_version":1}