{"id":"52c4e58f-5de6-47f6-bc2a-d9009983146b","arxiv_id":"2411.14552","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A Bayesian fit to LHC jet data claims a universal jet energy-loss distribution and super-Casimir color dependence, but the color result depends on a theory-informed prior.","lead":"This paper applies Bayesian inference to extract the energy-loss distribution of quark and gluon jets in heavy-ion collisions using three flexible functional forms. The authors report a universal energy-loss distribution across jet measurements and a color ratio of about 3.5, which is higher than the standard Casimir expectation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The super-Casimir CR≈3.5 claim rests on a Sec. VC prior whose gamma range (alpha in [1.11,2]) contradicts the paper's own n_i≤2–3 estimate; with flat priors the three parametrizations disagree.","rationale":"The reader's weakest-assumption analysis points to the Sec. VC prior restriction, and I agree that this is the load-bearing element. The concrete new observation is that the prior is internally inconsistent with the paper's own microphysics: the gamma shape parameter alpha is identified with 1+n in Sec. IIB, and Sec. IIA quotes realistic calculations with n_i≈2–3, which would place alpha in 3–4 and the mean-mode ratio in 1.33–1.5, outside the alpha_s<0.5 window used to set Table III. So the prior does not just encode a generic weakly-coupled prejudice; it cuts out the specific parameter region the authors' theoretical discussion says is realistic. The stark parametrization dependence of CR under flat priors (Fig. 11) confirms that the data alone do not determine the color ratio. The proposed re-run with alpha in [2.5,4.5] directly tests whether CR≈3.5 survives when the prior is moved to the internally stated physical region. The paper's honest reporting, closure tests, and leave-one-out validation are genuine strengths, and the universality evidence from the inclusive-data prediction of the per-photon yields is moderately supportive. However, the headline super-Casimir claim should not be accepted without this prior-sensitivity test, so the CONDITIONAL verdict remains appropriate.","tokens_in":33134,"tokens_out":7963,"duration_ms":73041,"concrete_test":"Re-run the Sec. VC global analysis for the gamma parametrization with flat priors on alpha_q and alpha_g in [2.5,4.5] (the range implied by the n≈1.5–3.5 resolved-emitter estimate cited in Sec. IIA), keeping all other priors and the likelihood identical, and report the posterior for CR. If CR moves away from ~3.5 (e.g., back toward the flat-prior value ~2.25) or the three parametrizations no longer agree, the super-Casimir conclusion is an artifact of the prior range. As a complementary check, scan the alpha_s bounds in Table III over (0.05,0.5), (0.1,0.4), and (0.2,0.6) and verify whether CR≈3.5 persists for all three parametrizations.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline claim is the super-Casimir color ratio CR≈3.5 reported in Sec. VC and Eq. (38). This result is not stable without the prior restriction introduced in Sec. VC: with flat priors, the normal, log-normal, and gamma parametrizations give CR≈3.5, ≈1, and ≈2.25 respectively (Fig. 11). The prior restricts the mean-to-mode ratio <epsilon>/epsilon_max to the range 2–10 implied by Eq. (6) for alpha_s in (0.1,0.5), via the ranges in Table III. The problem is that this prior is not merely an external theoretical input; it is inconsistent with the paper's own estimate of the number of resolved emitters. Section IIB identifies the gamma shape parameter as alpha = 1 + n, where n is the number of resolved gluons, and Sec. IIA cites realistic calculations giving n_i ≲ 2–3 (near Eq. 8). That implies alpha ≈ 3–4 and <epsilon>/epsilon_max = alpha/(alpha−1) ≈ 1.33–1.5, corresponding to 1/alpha_s ≈ 1.33–1.5, i.e. alpha_s ≈ 0.67–0.75, which is outside the imposed 0.1–0.5 window. Table III instead restricts gamma alpha to [1.11,2], i.e. n in [0.11,1]. Thus the prior excludes the physically preferred region of parameter space before the data are considered, and the convergence of all three parametrizations to CR≈3.5 in Fig. 14 may reflect this shared prior boundary rather than information from the data. The universality claim is also tempered by the ~50% over-prediction of the photon-tagged jet RAA discussed in Sec. VD, but the prior dependence of CR is the more load-bearing issue for the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper performs a Bayesian inference of the event-averaged jet energy-loss distribution (quenching weight) from ATLAS data on inclusive jet suppression, rapidity-dependent suppression ratios, and photon-tagged jet yields in 0–10% central Pb-Pb collisions at √s = 5.02 TeV. The quenching weights of quark- and gluon-initiated jets are parametrized independently by three flexible two-parameter distributions (normal, log-normal, gamma), with flavor fractions obtained from PYTHIA8, and the model is evaluated via a Gaussian-process emulator. The authors report two principal findings: (i) evidence for universality of the quark/gluon quenching weights across inclusive and photon-tagged observables, based on out-of-sample cross-predictions of Analyses A and B (Sec. VA); and (ii) an extracted color ratio CR = ⟨ε_g⟩/⟨ε_q⟩ ≈ 3.5, corresponding to super-Casimir scaling CR = (N_c/C_F)^{1+γ} with γ ≈ 0.5 (Eq. 38, Sec. VC). The color-ratio result is obtained after imposing a theory-informed prior that restricts the mean-to-mode ratio of the energy-loss distributions (Sec. VC); with flat priors the three parametrizations give inconsistent values CR ≈ 3.5, ≈ 1, and ≈ 2.25 (Fig. 11).","tokens_in":33555,"tokens_out":17608,"duration_ms":138007,"significance":"The methodology is the paper's main strength: the closure tests (Sec. IVC), the leave-one-out analysis (Sec. VB1), and the out-of-sample cross-predictions in Analyses A and B are carefully executed, and the authors are commendably transparent about the flat-prior disagreement among parametrizations (Fig. 11) and about the tension with the photon-tagged jet RAA (Sec. VD). I confirm the stress-test concern: the Sec. VC prior is load-bearing for the super-Casimir claim, and its window conflicts with the paper's own estimate of the number of resolved emitters, with the posterior sitting at the prior boundary. If the claim survived a reconciled prior choice, it would point to genuinely interesting multi-parton quenching effects beyond Casimir scaling; as it stands, the claim is conditional on the prior rather than established by the data, while the universality result is better supported but requires qualification given the photon-tagged RAA over-prediction.","major_comments":[{"comment":"The prior restriction that drives the central CR ≈ 3.5 result is internally inconsistent with the paper's own estimate of the number of resolved emitters. Section IIA quotes realistic calculations giving n_i ≲ 2–3, and Sec. IIB identifies the gamma shape parameter as α = 1 + n; together these imply α ≈ 3–4 and ⟨ε⟩/ε_max = α/(α−1) ≈ 1.33–1.5, which through Eq. (6) corresponds to α_s ≈ 0.67–0.75. Table III instead restricts the gamma distribution to α ∈ [1.11, 2], i.e. n ∈ [0.11, 1] and 1/α_s ≈ 2–10, so the region preferred by the paper's own physics discussion is excluded before the data are considered. Moreover, the posterior for the gamma parametrization prefers α_s near the lower boundary of the imposed window (Sec. VC), so the convergence of all three parametrizations to CR ≈ 3.5 in Fig. 14 may reflect the shared prior boundary rather than information from the data. The authors should either reconcile the α = 1 + n interpretation with the imposed α window, or demonstrate that CR is stable when the window is extended into the n ≲ 3 region; a closure test with truth CR = 2.25 under the restricted prior would be a useful check that the machinery does not bias the recovered ratio toward the boundary.","section":"Sec. VC; Eq. (6); Table III; Secs. IIA/IIB"},{"comment":"With flat priors the three parametrizations give mutually inconsistent color ratios: CR ≈ 3.5 (normal), ≈ 1 (log-normal), and ≈ 2.25 (gamma) (Fig. 11, right column), a disagreement the paper explicitly acknowledges in Sec. VC. The abstract nevertheless states that \"we extract that the color dependence of energy loss is slightly bigger than what expected from Casimir scaling,\" and Sec. VI concludes that \"we have established evidence of color charge dependence of energy loss,\" without the caveat that these statements hold only under the restricted prior of Sec. VC. Given the paper's stated philosophy of data-driven extraction with \"minimal assumptions from theory\" (Sec. I), the central claim should be presented explicitly as conditional on the Sec. VC prior, or the flat-prior disagreement should be resolved by an argument that rules out the log-normal and gamma parametrizations.","section":"Abstract; Sec. VB (Fig. 11); Sec. VI"},{"comment":"The universality claim in Sec. VA is qualified by the behavior of the photon-tagged jet RAA, an observable that is excluded from the main analysis. When predicted from the global analysis with nPDFs, this observable is over-predicted by about 50% at pT ≈ 100 GeV (textured bands, upper panel of Fig. 16), and the paper concludes that \"there is a certain tension between the inclusive jet and photon-tagged jet data\" (Sec. VD). The abstract's assertion that the analysis establishes \"the consistency between different data-sets\" therefore goes beyond what Sec. VD reports; the qualification given in the body text should also appear in the abstract and conclusions.","section":"Sec. VD; Fig. 16"}],"minor_comments":[{"comment":"The manuscript contains numerous typos and duplicated words, including \"to loose energy\" (Sec. IIA), \"gluon-initiated jets loosing more energy\" (Sec. VB), \"can take place occur\" (Sec. II), \"For for Pb-Pb central collisions\" (Sec. III), \"the the local conditions\" (Sec. I), and \"mean energy energy loss\" (Sec. IIA); these should be corrected.","section":"Secs. I–III"},{"comment":"The caption begins with \"Left panel:\" and then \"Left panels:\" for the two photon-tagged panels, which is confusing; the three panels should be referenced unambiguously.","section":"Fig. 2 caption"},{"comment":"The reduced chi-square values for the global fit are 2.28 (normal), 1.41 (log-normal), and 1.70 (gamma), i.e. notably above unity for two of the three parametrizations; the paper does not comment on whether these values indicate a formally poor fit and what that implies for the reliability of the corresponding extraction.","section":"Table II"},{"comment":"The closure tests validate the inference machinery within each parametrization family, but the Fig. 4 caption notes that the mock data are chosen \"with a specific offset with respect to the experimental data\"; reporting the actual truth values used, in particular the truth CR, would help the reader judge whether the closure tests cover the CR range relevant to the super-Casimir claim.","section":"Sec. IVC, Fig. 4"},{"comment":"Only the EPS09 nPDF set is used in the baseline; given the demonstrated sensitivity of the high-pT inclusive RAA and the photon-tagged jet RAA to nPDF effects (Fig. 3 and App. B), the authors should comment on the potential impact of more recent nPDF sets (e.g., EPPS21 or nNNPDF3.0) on the extracted CR.","section":"Sec. IIIA; App. B"},{"comment":"The citation \"JHEP21, 206 (2020)\" appears to have a malformed volume field; the format should be checked against the journal's style.","section":"Ref. [8]"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is honest and methodologically careful, and the universality analysis has real value once the photon-tagged RAA tension is properly qualified. The obstacle to acceptance in its current form is the super-Casimir claim: CR ≈ 3.5 is dominated by the Sec. VC prior, and the prior window conflicts with the paper's own n_i ≲ 2–3 estimate (Secs. IIA/IIB), with the posterior sitting at the prior boundary. If the authors cannot reconcile these numbers in revision, I would expect the claim to be downgraded to a conditional statement rather than presented as evidence for super-Casimir scaling. I would also encourage the editor to ensure the flat-prior disagreement in Fig. 11 retains due prominence, as it is the clearest evidence that the data alone do not determine CR."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Things you should know about arXiv:2411.14552. The genuinely new piece is the cross-observable universality test of the quenching weight: they fit inclusive jet RAA plus rapidity ratios, then predict per-photon jet yields (and vice versa) without re-fitting. That is a real out-of-sample test, and it passes. The extraction of a quark/gluon color ratio is secondary and softer.\n\nWhat they do well: the inference machinery is standard but carefully validated, with closure tests, leave-one-out checks, and explicit verification that the GP emulator does not distort results. The paper is unusually transparent when the three parametrizations disagree: Fig. 11 shows CR ≈ 3.5, ≈ 1, and ≈ 2.25 with flat priors. Most papers would bury that. They also flag the ~50% over-prediction of the photon-tagged RAA that was excluded from the main fit.\n\nThe soft spot is load-bearing for the headline. The super-Casimir CR≈3.5 only emerges after imposing the Eq. (6) prior that mean/mode ~ 1/alpha_s and restricting alpha_s to [0.1,0.5]. With that prior, all three parametrizations converge to CR≈3.5 (Fig. 14). But the prior window is internally inconsistent with their own estimate that n_i ≲ 2–3 resolved emitters, which gives a gamma shape parameter alpha ≈ 3–4 and mean/mode ≈ 1.33–1.5, i.e. alpha_s ≈ 0.67–0.75, outside the imposed range. So the restricted prior seems to exclude the physically preferred region before the data get a vote. The agreement across parametrizations in Fig. 14 may just reflect the shared prior boundary. That does not sink the universality result, but it means the color-ratio claim is a conditional statement, not a robust extraction. The authors acknowledge prior sensitivity in Sec. VC, but they do not confront the inconsistency with the resolved-emitter estimate.\n\nBottom line: this paper deserves a serious referee. The universality test is a genuine step forward for jet quenching phenomenology. The color-ratio extraction needs reworking, either by justifying the alpha_s window against the resolved-emitter estimate, testing sensitivity to prior boundaries, or including the photon-tagged RAA in the fit and publishing the code and data so others can check. I would send it to review with encouragement, expecting a major revision.","headline":"Careful, honest Bayesian analysis; the universality test is new and passes, but the super-Casimir color ratio rests on a prior that contradicts the paper's own resolved-emitter estimate.","tokens_in":34139,"tokens_out":2050,"would_cite":true,"duration_ms":17854,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["12.38.Mh","25.75.-q","24.85.+p"],"model":"deepseek-v4-flash","headline":"This paper claims that the same energy-loss distribution describes jet suppression in inclusive and photon-tagged heavy-ion data, and that gluon jets lose roughly 3.5 times more energy than quark jets — faster than simple color-charge…","keywords":["jet quenching","quark-gluon plasma","energy loss distribution","Bayesian inference","color factor","Casimir scaling","nuclear modification factor","photon-tagged jets"],"falsifier":"Measure an energy-loss-sensitive observable that bypasses the steeply falling spectrum bias — for example, the subleading-jet spectrum in $Z$-boson events or the low-$x_{J\\gamma}$ tail of photon-jet momentum imbalance, where large energy losses are not exponentially suppressed. If such data, analyzed with the same three parametrizations and no mean-to-mode prior, give a color ratio consistent with Casimir scaling ($\\mathrm{CR} \\approx 2.25$) for all three distribution shapes, the super-Casimir conclusion would be shown to rest on the prior rather than on the data.","tokens_in":32885,"feed_emoji":"⚛️","tokens_out":10196,"duration_ms":94664,"temperature":0.7,"pith_summary":"Jet suppression in heavy-ion collisions is usually modeled with Monte Carlo generators that mix many assumptions; this paper asks what the data themselves say about the energy-loss mechanism. It tests whether a single flavor-dependent probability distribution — the 'quenching weight' for the energy a jet loses to the quark-gluon plasma — can describe both inclusive jet suppression and photon-tagged jet yields from the LHC, and finds those data consistent with such a universal weight. It then extracts from the same data the color dependence of energy loss, i.e., how much more energy gluon-initiated jets lose than quark-initiated ones, a quantity that distinguishes single-parton from multi-parton quenching. The paper's headline result is a color ratio $\\langle \\varepsilon_g \\rangle / \\langle \\varepsilon_q \\rangle \\approx 3.5$, corresponding to super-Casimir scaling $(N_c/C_F)^{1+\\gamma}$ with $\\gamma \\approx 0.5$. This matters because it suggests jets lose energy as partially developed showers of resolved subjets, not as single color charges, and it shows how much of that conclusion rests on a physically motivated prior.","feed_headline":"Gluon jets lose 3.5 times more energy than quark jets","feed_subtitle":"Same quenching law fits inclusive and photon-tagged jets, but the color ratio exceeds Casimir scaling.","key_machinery":"The central object is the flavor-dependent quenching weight $D_i(\\varepsilon)$ entering the factorization $\\sigma_{\\rm med}(p_T) = \\sum_i \\int_0^\\infty d\\varepsilon\\, D_i(\\varepsilon)\\, \\sigma_{{\\rm vac},i}(p_T+\\varepsilon)$, which encodes all medium effects on the jet. The analysis wraps three two-parameter distributions (normal, log-normal, gamma), each with separate quark and gluon copies (four parameters per parametrization), in a Bayesian inference with a Gaussian-process emulator used for fast model evaluation, and checks predictions on held-out observables via posterior predictive distributions. The physical anchor that decides between otherwise degenerate fits is the mean-to-mode relation $\\langle \\varepsilon \\rangle / \\varepsilon_{\\rm max} \\approx 1/\\alpha_s$, which converts the requirement $\\alpha_s \\in (0.1,0.5)$ into a bound on the tail thickness of $D_i(\\varepsilon)$; when the authors restrict the prior to that bound, the three parametrizations agree on the color ratio $\\mathrm{CR}$.","core_discovery":"Under the usual factorization that heavy-ion jet spectra are a convolution of the vacuum spectrum with a flavor-dependent energy-loss distribution $D_i(\\varepsilon)$, the authors use Bayesian inference over three flexible parametrizations (normal, log-normal, gamma) to extract $D_q$ and $D_g$ from LHC measurements of the inclusive jet nuclear modification factor, its rapidity-dependent ratios, and photon-tagged jet yields. They find, first, that a single set of quenching weights describes both classes of observables, establishing what they call universality of the energy-loss distribution. Second, they find that the mean energy loss of gluon-initiated jets exceeds that of quark-initiated jets by a factor that, once a physical restriction on the mean-to-mode ratio ($0.1 < \\alpha_s < 0.5$) is imposed, converges to about 3.5, parametrized as $\\mathrm{CR} = (N_c/C_F)^{1+\\gamma}$ with $\\gamma \\approx 0.5$. They interpret this super-Casimir scaling as evidence that energy loss acts on the partially developed parton shower — multiple resolved subjets contribute — rather than on a single parent parton.","pith_inferences":["Editorial inference: If the super-Casimir color dependence is real, the ratio $\\langle \\varepsilon_g \\rangle / \\langle \\varepsilon_q \\rangle$ should grow with the number of resolved subjets, i.e., with jet energy at fixed radius; repeating the same analysis in higher-$p_T$ bins once less biased data are available would test this prediction.","Editorial inference: The near-flat, tail-heavy shapes extracted for $D(\\varepsilon)$ may encode the survival bias of steeply falling spectra as much as the underlying physics; a cleaner test would come from observables that amplify rare large energy losses, such as the far tail of photon-jet momentum imbalance distributions.","Editorial inference: Allowing $\\alpha_s$ to vary with flavor or with jet $p_T$, or adding a fourth shape parameter to the parametrizations, could reveal whether the converged value $\\mathrm{CR} \\approx 3.5$ survives without a hand-set prior on the mean-to-mode ratio.","Editorial inference: The mean-to-mode argument ties the Poisson-like radiative spectrum to the distribution shape; if elastic energy loss or medium response contribute substantially at these kinematics, the same prior logic would need revision, plausibly changing the extracted color ratio."],"forward_implications":["The same quenching weight describes inclusive and photon-tagged jet measurements, so future global fits can treat energy loss as medium-dominated and largely flavor-independent before the final color dependence is applied.","The rapidity-dependent inclusive jet data carry most of the constraining power on the gluon quenching weight; photon-tagged jet data alone cannot pin down the gluon energy-loss parameters.","Without the mean-to-mode prior, the inferred color ratio $\\langle \\varepsilon_g \\rangle / \\langle \\varepsilon_q \\rangle$ varies strongly across parametrizations (about 3.5, 1, and 2.25), so the prior is what currently permits a stable statement about color dependence from existing data.","The photon-tagged jet nuclear modification factor sits in tension with the global inclusive-jet fit unless nuclear-PDF effects are treated in a particular way, indicating an inconsistency between data sets that future measurements or recalibrations should resolve."],"supporting_citations":[{"why":"Provides the factorization of the heavy-ion jet spectrum into a vacuum spectrum convolved with an energy-loss distribution, the starting equation of the analysis.","marker":"[5]"},{"why":"Supplies the argument that steeply falling spectra bias observables toward small energy loss, motivating the prior restriction on the distribution tail.","marker":"[7]"},{"why":"The Bayesian inference framework with parallel tempering and a Gaussian-process emulator that this analysis follows for posterior sampling.","marker":"[17]"},{"why":"The factorized jet-spectrum formula with semi-inclusive jet functions used to build the quark- and gluon-initiated baseline spectra.","marker":"[56]"},{"why":"The multi-parton picture of in-medium jet evolution that underlies the lower bound on the color ratio and the super-Casimir interpretation.","marker":"[62]"},{"why":"The inclusive jet nuclear modification factor data, the most constraining observable for the gluon quenching weight.","marker":"[67]"},{"why":"The per-photon jet yield data that provides the quark-enriched photon-tagged observables used in the analysis.","marker":"[68]"},{"why":"The photon-tagged jet nuclear modification factor, used as a held-out prediction and in the tension study.","marker":"[69]"},{"why":"The nuclear PDF set used to generate the heavy-ion baseline spectra, whose inclusion changes the photon-tagged jet prediction.","marker":"[73]"}],"fun_headline_variants":["Gluon jets lose 3.5x more energy than quarks in heavy-ion collisions","Super-Casimir jet quenching: gluon loss 3.5 times quark's","Bayesian analysis reveals universal jet energy-loss law","Heavy-ion jets: gluon energy loss exceeds Casimir scaling","Gluon quenching beats Casimir prediction by factor 3.5"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The super-Casimir color ratio rests on the prior assumption that the mean energy loss is not too far from the most probable energy loss (specifically, $\\langle \\varepsilon \\rangle / \\varepsilon_{\\rm max}$ between 2 and 10, i.e., $0.1 < \\alpha_s < 0.5$); without that restriction, the three flexible distribution shapes used in the fit give color ratios of about 3.5, 1, and 2.25, so the data alone do not single out a value above Casimir scaling.","fun_headline_variants_meta":{"raw":{"variants":["Gluon jets lose 3.5x more energy than quarks in heavy-ion collisions","Super-Casimir jet quenching: gluon loss 3.5 times quark's","Bayesian analysis reveals universal jet energy-loss law","Heavy-ion jets: gluon energy loss exceeds Casimir scaling","Gluon quenching beats Casimir prediction by factor 3.5"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000224,"raw_usage":{"total_tokens":1454,"prompt_tokens":934,"completion_tokens":520,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":550,"completion_tokens_details":{"reasoning_tokens":422}},"tokens_in":550,"tokens_out":520,"duration_ms":4989,"temperature":1.0,"reasoning_tokens":422,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:10:30.478760+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure an energy-loss-sensitive observable that bypasses the steeply falling spectrum bias — for example, the subleading-jet spectrum in $Z$-boson events or the low-$x_{J\\gamma}$ tail of photon-jet momentum imbalance, where large energy losses are not exponentially suppressed. If such data, analyzed with the same three parametrizations and no mean-to-mode prior, give a color ratio consistent with Casimir scaling ($\\mathrm{CR} \\approx 2.25$) for all three distribution shapes, the super-Casimir conclusion would be shown to rest on the prior rather than on the data.","supporting_citations":[{"cited_title":"Quantifying the Jet Energy Loss in Pb+Pb collisions at LHC","cited_arxiv_id":"2410.07852","evidence_quote":"The photon-tagged jet nuclear modification factor, used as a held-out prediction and in the tension study."}],"review_version":1}