{"id":"49bc7296-d7e7-48d5-8f98-a8dbe2152dc5","arxiv_id":"1908.08461","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":6,"one_line_summary":"ATLAS finds no evidence for tqγ flavour-changing neutral currents and sets the most stringent limits to date on the associated production cross sections and branching ratios.","lead":"ATLAS searched LHC proton collisions for flavour-changing neutral currents involving a top quark and a photon. No signal was found, and the new upper limits on these rare processes are the strongest published so far.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Dominant e→γ fake background is normalized by a single SF with a flat 10% uncertainty; a pT/η-dependent fake rate could bias the NN-based limit, though the exposed size is likely within systematics.","rationale":"The reader correctly identifies the data-driven fake-photon estimates as the weakest assumption. I agree with that general area, but I would place the emphasis on the electron-fake background rather than the hadron-fake correlation factor: e→γ fakes are about 47% of the SR background, whereas hadron fakes are only 260 of 9500 events. The paper does assign a 10% uncertainty to fe→γ after a validation-region check and quotes a 9% uncertainty on the e→γ fake yield in Table 1, so the concern is partially covered. The remaining gap is shape: a single normalisation uncertainty does not protect against a pT- or η-dependent fake rate that distorts the NN output distribution. The photon-pT reweighting is also flagged by the reader, but the paper assigns an explicit conservative uncertainty to it by comparing with and without the correction, and the hadron-fake correlation factor is varied by ±50%, so those are less likely to move the final limits. Given the strong internal consistency checks, the null result, and the fact that observed limits are close to expected limits, this concern does not by itself invalidate the central claim. A dedicated pT-binned cross-check would settle whether the flat-SF approximation is numerically important, but the current verdict of ACCEPT with high confidence remains reasonable.","tokens_in":13,"tokens_out":9681,"duration_ms":175456,"concrete_test":"Recompute Table 2 with the e→γ fake scale factor binned in photon pT and |η|, varying each bin within the 10% normalisation envelope and adding a smooth shape variation consistent with the validation-region discrepancy; if any 95% CL limit on σ(pp→tγ) or B(t→qγ) moves by more than its quoted total uncertainty, the flat-SF assumption is load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central output is a set of one-sided exclusion limits, so the predicted background composition in the SR matters most, especially in the high-NN-output bins where signal would appear. The dominant SR background is electrons misidentified as photons (4500±400 of 9500 predicted events in Table 1). Section 6 normalises this background with a single scale factor (0.978±0.040) measured in Z→ee-like events and applied to simulation; after a validation-region check, the relative uncertainty on fe→γ is inflated to 10%, but this is a flat normalisation uncertainty. The NN discriminant for this dominant background is then taken from simulation, and no dedicated shape uncertainty for a possible pT or η dependence of fe→γ is described in Section 9. If the true fake rate rises with photon pT or varies in the SR phase space differently than in the Z→ee control sample, events would migrate between NN bins, changing the extracted signal limit by an amount a single normalisation nuisance parameter cannot cover. The hadron-fake correlation factor (0.85±0.14, varied ±50%) and the photon-pT reweighting are guarded by explicit conservative systematics and contribute much less to the SR, so the e→γ fake shape is the most load-bearing weakness.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a search for flavour-changing neutral current (FCNC) tqγ couplings using 81 fb−1 of proton-proton collision data at √s = 13 TeV collected with the ATLAS detector. Events with exactly one photon, one electron or muon, one b-tagged jet, and missing transverse momentum define the signal region. A neural network based on ten kinematic variables is used to discriminate signal from background, and a profile likelihood fit to the NN output distributions in the signal region and to the photon pT distributions in two control regions extracts the signal and background normalisations. The data are found to be consistent with the background-only hypothesis, and 95% CL upper limits are set on the effective tqγ couplings, on the FCNC tγ production cross sections, and on the branching ratios B(t→γu) and B(t→γc) for left-handed and right-handed couplings. The paper claims these limits are the most stringent to date.","tokens_in":42941,"tokens_out":5546,"duration_ms":65440,"significance":"If the results are correct, they represent a substantial improvement over previous searches, constraining FCNC top-photon couplings to branching ratios of order 10^-5 and cross sections of order tens of femtobarns. The analysis has several clear strengths: the simultaneous profile likelihood fit treats many correlated systematic uncertainties, the electron and hadron fake backgrounds are estimated with data-driven methods in dedicated control regions, the neural network is validated with cross-validation, and the CLs procedure is used correctly. There is no circularity in the statistical procedure: signal and background normalisations are obtained from the fit, and the limits follow from the fitted signal contribution. The manuscript is concise and generally clear. The main reservation concerns the treatment of the dominant electron-fake background, whose kinematic shape uncertainty is not addressed in the systematic evaluation.","major_comments":[{"comment":"","section":"Section 6 and Section 9"}],"minor_comments":[{"comment":"The notation |C(13)*uW + C(13)*uB| is not defined in the text; please define the operator coefficients and explain the meaning of the parentheses and the star superscript.","section":"Table 2"},{"comment":"The validation region used to check the modelling of the electron-fake kinematics is restricted to EmissT < 30 GeV, while the SR requires EmissT > 30 GeV; a sentence explaining why this validates the SR phase space would strengthen the argument.","section":"Section 6"},{"comment":"The limited number of Monte Carlo events is listed among the largest systematic uncertainties, but the implementation of this uncertainty in the fit (for example, Barlow-Beeston or equivalent) is not described; please clarify.","section":"Section 9"},{"comment":"The statement that events must have 'no further jets' is important for the decay-mode acceptance, which is only 0.45–0.51%; please state explicitly whether the required b-tagged jet counts toward this jet multiplicity or whether 'no further jets' means no additional jets beyond the b-tagged one.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"My main concern is the missing shape systematic for the electron-fake background, which is the largest background in the signal region and is normalised with a single scale factor. This is a fixable issue: the authors could add a pT-dependent shape variation for fe→γ or provide a quantitative argument that the flat 10% uncertainty covers the NN-shape effect. If that is addressed, I would consider the paper acceptable. The statistical procedure and the rest of the analysis appear sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a competent, incremental but genuinely new result. It sets the best current limits on FCNC tqγ couplings using 81 fb⁻¹ of 13 TeV data, improving the CMS 8 TeV limits by roughly a factor of 2–5. The analysis is technically careful: profile likelihood with control regions, data-driven fake estimates, a neural network discriminator, and CLs limits. The central claim—data consistent with background-only—is credible, and the numbers in Tables 1 and 2 hang together.\n\nWhat is actually new: this is the first ATLAS search in this final state using a neural network, and the limits on B(t→γu), B(t→γc), and σ(pp→tγ) for both left-handed and right-handed couplings are numerically the most stringent to date. The EFT interpretation is standard. There is no circularity: the W/Z+γ normalizations come from control regions, fakes are data-driven, and the limits follow from the fitted signal strength in the signal region.\n\nSoft spots: the dominant background is electrons misidentified as photons, about half the signal-region yield. The fake rate is measured in a Z→ee-like phase space and applied as a flat scale factor, with the normalization uncertainty inflated to 10% after a validation-region check. The stress-test concern is fair: there is no explicit pT/η-dependent shape uncertainty on the e→γ fake rate, and the validation region checks kinematic variables but not the NN output itself. A shape error could reshuffle events between NN bins and move the limit by more than a single normalization nuisance would cover. I would not call it fatal: the background model is checked in control regions, the photon pT reweighting carries a dedicated with/without systematic, and the observed and expected limits are close for most couplings. But a referee would reasonably ask for a closure test or shape systematic on the e→γ fake.\n\nThe hadron-fake scale factor (1.7 ± 1.0) is ugly but the contribution is small. The photon pT correction tuned to data in five bins is mildly uncomfortable but is assigned a conservative systematic. The citation pattern is clean, and the result is already published in Physics Letters B, so the question is whether the record deserves scrutiny rather than whether to publish.\n\nWho should read it: anyone doing top-quark EFT fits or LHC projections for FCNC top couplings. It is not a paradigm shift; it is a well-executed search that sets the reference numbers.\n\nRecommendation: yes, send it to peer review. If I were refereeing, I would accept after asking for one explicit e→γ fake shape-closure test, but the main claim is sound.","headline":"A solid, incrementally new ATLAS limit on tqγ FCNC couplings; the e→γ fake-rate shape is the one caveat worth probing, but the central result holds.","tokens_in":43484,"tokens_out":2924,"would_cite":true,"duration_ms":34299,"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 search finds no evidence for flavour-changing top-photon currents and sets the tightest limits yet on the tqγ coupling.","keywords":["flavour-changing neutral currents","top quark","photon","effective field theory","proton-proton collisions","branching ratio","neural network","single top production"],"falsifier":"A tag-and-probe measurement of the electron-to-photon misidentification rate inside the signal-region phase space that disagrees with $f_{e\\to\\gamma}=3.11\\% \\pm 0.01\\% \\pm 0.13\\%$ by more than the assigned systematic, or a future observation of a significant $pp\\to t\\gamma$ excess at high neural-network output, would contradict the background-only conclusion.","tokens_in":42434,"feed_emoji":"⚛️","tokens_out":12404,"duration_ms":114799,"temperature":0.7,"pith_summary":"This paper searches for flavour-changing neutral currents at the coupling of a top quark, a photon, and an up or charm quark, a process that the Standard Model suppresses to branching ratios around $10^{-14}$ but that several new-physics models can enhance to observable rates. Using $81~\\mathrm{fb}^{-1}$ of proton-proton collisions at 13 TeV, it selects events with a photon, a charged lepton, a $b$-tagged jet, and missing transverse momentum, and separates potential signal from background with a neural network. The data are consistent with the background-only hypothesis, so no new signal is claimed. Instead, the paper sets 95% confidence-level upper limits on the effective $tq\\gamma$ coupling, on the cross section for FCNC $t\\gamma$ production, and on the branching ratios $\\mathcal{B}(t\\to\\gamma u)$ and $\\mathcal{B}(t\\to\\gamma c)$. If correct, these are the most stringent limits to date, and they would compress the parameter space of models such as R-parity-violating supersymmetry and two-Higgs-doublet models that predict larger top-photon flavour-changing rates.","feed_headline":"No sign of top-photon flavour change in 81 fb^-1 of proton-proton data","feed_subtitle":"The new 95% CL limits on t→γu and t→γc branching ratios reach 10^-5, shrinking the room for new physics.","key_machinery":"The argument rests on an effective-field-theory parametrization of the $tq\\gamma$ vertex through the dimension-six operators $O_{uB}$ and $O_{uW}$, whose Wilson coefficients are the parameters being constrained. Signal samples are generated at next-to-leading order in QCD with this model, and the event selection is optimized for $pp\\to t\\gamma$ production. A fully connected feed-forward neural network, trained on ten kinematic variables, separates signal from background, and a profile-likelihood fit to the network output in the signal region plus the $W+\\gamma$ and $Z+\\gamma$ control regions extracts the signal contribution. The two dominant backgrounds, electrons and hadrons misidentified as photons, are normalised with data-driven estimates: the electron fake rate $f_{e\\to\\gamma}=3.11\\% \\pm 0.01\\% \\pm 0.13\\%$ and a hadron-fake estimate built from three hadron-fake regions via $N(\\mathrm{HFR}_{\\mathrm{pass}|\\mathrm{fail}}) \\times N(\\mathrm{HFR}_{\\mathrm{fail}|\\mathrm{pass}}) / N(\\mathrm{HFR}_{\\mathrm{fail}|\\mathrm{fail}})$, corrected by a simulation-derived correlation factor of $0.85\\pm0.14$.","core_discovery":"The central claim is that, in the selected final state, the observed event yield agrees with the Standard Model background-only prediction, and the 95% CL upper limits derived from this agreement are the tightest published bounds on the $tq\\gamma$ flavour-changing vertex. For a left-handed (right-handed) $tu\\gamma$ coupling the limits are 36 fb (78 fb) on the $\\sigma(pp\\to t\\gamma)$ production cross section and $2.8\\times10^{-5}$ ($6.1\\times10^{-5}$) on $\\mathcal{B}(t\\to\\gamma u)$; for $tc\\gamma$ they are 40 fb (33 fb) and $22\\times10^{-5}$ ($18\\times10^{-5}$). Interpreted through the dimension-six operators $O_{uB}$ and $O_{uW}$ at an effective scale $\\Lambda = 1$ TeV, the limits on the coupling combinations range from 0.19 to 0.52 depending on quark flavour and handedness. The search is driven by the single-top-plus-photon production mode, but it also includes the $t\\bar{t}$ decay mode, which contributes substantially to the charm-quark case because of the smaller charm-quark parton density.","pith_inferences":["At a coupling near the current limit, a future dataset several times larger should either reveal a $t\\gamma$ signal or push the branching-ratio bound below $10^{-5}$, a region where several new-physics models predict observable rates.","Because this analysis is optimized for the production mode, a dedicated selection for the decay mode that allows additional jets could improve the sensitivity to the $tc\\gamma$ coupling, where the decay mode contributes substantially.","The same neural-network-plus-profile-likelihood strategy could be adapted to searches for top-quark flavour-changing couplings to a $Z$ boson or a Higgs boson, whose signal topologies and background compositions are similar."],"forward_implications":["The $tq\\gamma$ vertex is constrained to effective-coupling combinations below 0.19–0.52 at $\\Lambda = 1$ TeV, excluding new-physics parametrizations that predict larger values.","The upper limits on $\\mathcal{B}(t\\to\\gamma u)$ and $\\mathcal{B}(t\\to\\gamma c)$ reach $10^{-5}$ to $10^{-4}$, roughly an order of magnitude beyond the previous best limits.","FCNC single-top-plus-photon production cross sections are bounded to 33–78 fb, so models predicting production above a few tens of femtobarns are ruled out.","The limits complement existing constraints on $tqZ$ flavour-changing couplings, jointly narrowing the allowed flavour structure of dimension-six operators."],"supporting_citations":[{"why":"Supplies the previous best limits on $\\mathcal{B}(t\\to\\gamma u)$ and $\\mathcal{B}(t\\to\\gamma c)$ that this analysis improves upon.","marker":"[8]"},{"why":"Provides the effective-field-theory model and NLO QCD simulation used to generate the FCNC signal samples.","marker":"[6,18]"},{"why":"Defines the dimension-six operators $O_{uB}$ and $O_{uW}$ in whose Wilson coefficients the limits are expressed.","marker":"[7]"},{"why":"Establishes the data-driven method and measured value of the electron-to-photon misidentification rate used for the dominant background.","marker":"[56]"},{"why":"Supplies the CLs statistical procedure used to derive the 95% CL upper limits.","marker":"[74]"},{"why":"Provides the complementary $tqZ$ coupling limits against which the $tq\\gamma$ constraints are compared.","marker":"[75]"}],"fun_headline_variants":["No FCNC in top-photon events: ATLAS sets 10^-5 limits","ATLAS tightens top-photon FCNC bounds to 10^-5","Top-quark photon flavour change ruled out at 10^-5","81 fb^-1 of ATLAS data: no top-photon flavour change","ATLAS constrains top-photon FCNC to 10^-5 branching"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The limits assume that the probabilities for electrons and hadrons to be misidentified as photons, measured in dedicated control regions, remain valid in the signal region after the applied data-driven corrections.","fun_headline_variants_meta":{"raw":{"variants":["No FCNC in top-photon events: ATLAS sets 10^-5 limits","ATLAS tightens top-photon FCNC bounds to 10^-5","Top-quark photon flavour change ruled out at 10^-5","81 fb^-1 of ATLAS data: no top-photon flavour change","ATLAS constrains top-photon FCNC to 10^-5 branching"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000747,"raw_usage":{"total_tokens":3412,"prompt_tokens":1114,"completion_tokens":2298,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":730,"completion_tokens_details":{"reasoning_tokens":2194}},"tokens_in":730,"tokens_out":2298,"duration_ms":15415,"temperature":1.0,"reasoning_tokens":2194,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:38:35.064625+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A tag-and-probe measurement of the electron-to-photon misidentification rate inside the signal-region phase space that disagrees with $f_{e\\to\\gamma}=3.11\\% \\pm 0.01\\% \\pm 0.13\\%$ by more than the assigned systematic, or a future observation of a significant $pp\\to t\\gamma$ excess at high neural-network output, would contradict the background-only conclusion.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the CLs statistical procedure used to derive the 95% CL upper limits."}],"review_version":1}