{"id":"9206b87d-d48d-4d12-acc6-2faabe32d97a","arxiv_id":"2603.01210","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A pre-shot workflow coupling RAPTOR transport and FBT equilibrium simulations predicts TCV discharge profiles and improves feedforward coil-current preparation for plasma shape control.","lead":"This paper describes a new workflow that predicts the plasma state of an entire TCV tokamak discharge before it is run, by combining two existing simulation codes. It shows that using these predicted profiles to prepare the magnetic coil currents improves the accuracy of the plasma shape during the experiment.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The KEP's headline benefit is not robustly demonstrated: the 211-shot benchmark uses post-shot n_e,l, and no sensitivity analysis shows how H98/n_e,l uncertainty propagates into l_i, β_N, and PF coil-current corrections.","rationale":"Reader's weakest_assumption is the same one I would stress: the workflow's value depends on inputs that are not themselves predicted reliably. The paper is transparent about this (Sec. 2.1 ad-hoc density rules; Sec. 3 post-shot n_e,l), so this is not an accusation of hidden error. But the missing analysis is a propagation/sensitivity study. The claimed improvement is a correction to feedforward PF currents; to know whether the correction is meaningful, one must show it is larger than the variation induced by realistic pre-shot input errors. The current evidence — one-shot Fig. 7 remark, 211-shot benchmark with experimental density, two experimental examples — does not establish that. The contradictory caption/shot-number issues in Sec. 4 are also worth fixing, but the conditional-input problem is the more load-bearing scientific gap. The reader's CONDITIONAL verdict is appropriate; my check would either confirm it or sharpen it into a REJECT if the sensitivity is large. I agree with the reader's weakest_assumption.","tokens_in":28288,"tokens_out":13276,"duration_ms":130372,"concrete_test":"Select a representative subset (e.g., all 51 PT H-mode transition shots plus ~20 NT shots). For each, run the RAPTOR-FBT coupling three ways: (i) nominal H98/n_e,l as in the paper; (ii) n_e,l perturbed by ±20% and H98 by ±0.1/±0.2 in each phase; (iii) the same with the ad-hoc predicted n_e,l. Record Δl_i, Δβ_N, and Δ|I_PF| per coil relative to (i). Compare these perturbation-induced spreads to the KEP-vs-initial-FBT correction distribution in Fig. 12 and to LIUQE-KER error bars. If the input-uncertainty spread is comparable to or larger than the KEP correction, the claim that KEP improves coil-current/shape prediction is unsupported without reliable H98/n_e,l estimates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is conditional on pre-shot estimates of H98(y,2) and n_e,l, and the paper provides no quantitative test of how errors in these inputs propagate to the quantities that are supposed to improve (l_i, β_N, PF coil currents). The 211-shot benchmark in Sec. 3 is run with post-shot experimental n_e,l; the ad-hoc density model is admitted to be insufficient for large-database validation, and the fully predictive workflow is demonstrated on only two discharges in Sec. 4. Fig. 7's single-shot statement that density quality has 'reasonably low impact' on p' and TT' is not a sensitivity study, and the coil-current corrections in Fig. 12 are often tens of amps and up to hundreds of amps — the same order as plausible changes from a ±20–30% n_e,l error or an H98 error of 0.1–0.2. Additionally, the NT transport parameters in Table 1 were selected using 41 of the 211 benchmark shots, making part of the 'wide range' validation in-sample. If the input sensitivities are as large as the claimed correction, the operational benefit of KEP over standard FBT is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a Kinetic-Equilibrium Prediction (KEP) workflow for TCV, coupling the RAPTOR 1.5D transport solver with the FBT free-boundary inverse equilibrium solver. RAPTOR predicts full-discharge profiles of current, temperatures, and density from pulse-schedule information using a gradient-based transport model whose pedestal gradients are PI-controlled to match prescribed values of H98 and line-averaged density. The resulting p' and TT' profiles are fed to FBT, and the two codes are iterated to self-consistency. The paper reports convergence in a few iterations, a benchmark on 211 TCV discharges (using experimental line-averaged density for the large database), and two fully predictive experimental demonstrations (one PT-LSN H-mode, one upper-NT snowflake), claiming improved X-point alignment and stationarity when the KEP-computed feedforward coil currents are used.","tokens_in":28741,"tokens_out":7283,"duration_ms":75656,"significance":"If fully validated, KEP would be a practically valuable tool: it is fast (minutes per discharge), uses existing TCV infrastructure, and directly addresses a known weakness of standard FBT preparation—the use of simple polynomial p' and TT' profiles with operator-chosen beta_pol and l_i. The paper's strengths include a clear coupling scheme, a convergence demonstration (Fig. 5), a large-database transport benchmark, and two experimental tests showing improved X-point targeting. However, the significance of the benchmark is limited by the use of post-shot density in Sec. 3, and the headline operational benefit is not yet supported by a sensitivity analysis connecting input uncertainties (H98, n_e,l) to the claimed coil-current improvements. These are addressable with existing data and additional analysis, so the core idea is defensible but the current evidence is incomplete.","major_comments":[{"comment":"The 211-shot benchmark does not validate the fully predictive pre-shot workflow. Sec. 3 states that all simulations in that section use post-shot experimental n_e,l because the ad-hoc density model 'does not yet allow for validation over a large shot database'; the fully predictive workflow (including the n_e,l model) is demonstrated only for the two discharges of Sec. 4. The abstract and conclusion nonetheless present the 211-shot result as evidence for pre-shot simulation 'across a wide range of plasma shapes and scenarios'. Since n_e,l is one of the two user-supplied inputs, and Fig. 7 shows a ~20% n_e,l error at one time in one shot, the operational benefit of KEP is not yet established. Please (i) run the ad-hoc density model over the 211-shot database and report its error statistics, and (ii) add a propagation study of realistic H98 and n_e,l uncertainties (e.g., ±20% in n_e,l, H98","section":"Sec. 3 and Sec. 4"},{"comment":"The energy-content validation is largely prescribed by the input H98. The PI controller adjusts mu_Te to match H98 = tau_E/tau_scal_E (Sec. 2.1), with default H98 values in Table 1, so W_e is forced to track H98*P_loss*tau_scal. The within-20% agreement in Figs. 10, A.3 and A.4 is therefore mostly a test of the assumed H98 values and the ITER scaling, not an independent test of predictive transport. The equilibrium-relevant quantities—l_i, beta_pol/beta_N, and the profile shapes of p' and TT'—are not compared against reconstructions over the database; only single-shot examples are given (Figs. 13, 15). Please add a database-level comparison of predicted beta_pol and l_i (or beta_N) to LIUQE-KER/MER values, and report profile-shape errors separately from energy-content errors.","section":"Sec. 2.1, Table 1, Figs. 10/A.3"},{"comment":"The NT transport parameters are selected in-sample. The text states that the NT parameters in Table 1 'were selected to match the set of discharges simulated in this study (41 NT shots, among the 211 shots presented in Section 3)'. Consequently, the good NT agreement in the 211-shot benchmark is not out-of-sample evidence for the 'wide range of plasma shapes and scenarios' claim. Please provide a cross-validation (e.g., train on a subset of NT shots and validate on the rest) or otherwise quantify the sensitivity of results to lambda_Te, lambda_Ti, lambda_ne and H98_NT. This is important because NT is one of the two headline scenarios in Sec. 4.","section":"Sec. 2.1, Table 1"},{"comment":"The L–H transition model relies on ad-hoc prohibitive thresholds: alpha_l is 'set to a high prohibitive value' and NT access is inhibited by construction. The intermediate s-region of f_div is constrained by essentially one discharge (#82274, Fig. 11), and the paper itself notes that DN behavior 'remains uncertain'. Since transition timing sets the confinement regime (and hence the effective H98 used by the controller) in the fully predictive workflow, the sensitivity of KEP outputs to alpha_f, alpha_u, and the s-criterion should be quantified. At minimum, report the distribution of P_sep/P_LH and s for the benchmark and identify which shots lie in the sensitive region |s| <= 1.","section":"Eq. (2.8) and Sec. 2.1"},{"comment":"The experimental demonstration of improved coil-current programming is based on very few discharges without quantitative error analysis. For #81882, one FBT-RAPTOR-prepared shot is compared with one standard shot; for NT-SF, two vs three shots are compared. Fig. 17 shows time traces but no uncertainties, no X-point-error metric, and no control for shot-to-shot variability. Since the Delta|I_a| corrections in Fig. 12 are tens to hundreds of amperes—the same order as the likely effect of input uncertainties—the claim that KEP 'improves the evaluation of coil currents' needs a quantitative metric, e.g., time-integrated X-point gap error for KEP vs standard preparations, and a comparison of the KEP correction amplitude to the propagated H98/n_e,l uncertainty.","section":"Sec. 4, Figs. 14 and 17"}],"minor_comments":[{"comment":"The section is titled 'Benchmark of the pre-shot prediction of 207 shots' but the text consistently says 211 shots; the mismatch should be corrected.","section":"Sec. 3 title"},{"comment":"The caption refers to shot #83740 as the FBT-RAPTOR-prepared shot, while Sec. 4.1 text says #83940. Please reconcile.","section":"Fig. 14 caption"},{"comment":"The caption of Fig. 16 and the text of Sec. 4.2 appear to swap the shot numbers for the initial FBT and FBT-RAPTOR groups relative to Fig. 17; please clarify which shots used which preparation.","section":"Figs. 16 and 17"},{"comment":"The two H98 columns are labeled 'H 98(y,2) e' and 'H 98(y,2)' with no explicit explanation in the caption; a sentence defining electron vs total confinement factor would help.","section":"Table 1"},{"comment":"The ad-hoc density rules (125 ms decay lifetime, +30% for NBI, +30% for H-mode) are TCV-specific; please state more explicitly that these are not intended as a general model and cite any prior use, to avoid overgeneralization.","section":"Sec. 2.1, line-averaged density model"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest about many of its limitations, but the abstract and conclusion overstate the validation. The missing sensitivity analysis and database-level comparison of beta_pol/l_i are feasible with existing data and would substantially strengthen the paper. The NT in-sample parameter selection and the small-n experimental demonstration are additional load-bearing issues that need to be addressed or clearly reframed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: a solid engineering integration paper that is honest about its conditions, with a real 211-shot benchmark — but the operational payoff, better coil-current prediction, is only partially supported because the benchmark uses post-shot density and there is no sensitivity analysis.\n\nWhat is actually new: the loose RAPTOR–FBT coupling applied to full TCV discharges before the shot, the extended Martin scaling with the f_div(s) geometry factor, the NT branch of the gradient-based model, and the first wide-database validation of the ion heat equation in RAPTOR. The energy-content benchmark within ±20% over 211 shots is legitimate, and the two experimental demonstrations show better X-point alignment and longer shape control with the KEP feedforward currents. The paper is also explicit that the gradient-based model requires a starting hypothesis about H98 and that the density predictor is a set of ad-hoc rules.\n\nSoft spots, in proportion. The fully predictive claim rests on two discharges; the 211-shot benchmark runs with experimental post-shot n_e,l because the density model, in the paper's own words, does not yet allow validation over a large database. The stress-test concern about input sensitivity is real: Fig. 7 is a single shot's comparison, not a propagation study, and the coil-current corrections in Fig. 12 are tens to hundreds of amps — same order as plausible H98 or n_e,l errors. So \"KEP improves the coil-current estimate\" is plausible but not quantitatively demonstrated. The NT transport parameters were selected using 41 of the 211 benchmark shots, making that subset of the wide validation in-sample. The Sec 4 figure captions contradict the text on which shot was prepared with which initial condition; that is fixable but currently muddles the experimental evidence. No code or data is deposited yet.\n\nNone of this sinks the paper. The authors do not oversell: the abstract conditions the method on H98 and n_e,l estimates, and the conclusion acknowledges the density model as a limitation. A referee should push for a sensitivity analysis of H98/n_e,l uncertainties into l_i, beta_N, and coil currents, plus the caption fix and a public dataset. With those, this becomes a useful reference for TCV operations and for similar pulse-preparation workflows elsewhere.\n\nRecommendation: yes, send to a serious referee. It deserves referee time, and the missing sensitivity analysis is exactly what a good referee should ask for.","headline":"Solid, honest engineering integration paper with a real 211-shot benchmark, but the operational payoff (better coil-current prediction) is under-proven because the benchmark uses post-shot density and there is no sensitivity analysis.","tokens_in":29248,"tokens_out":3789,"would_cite":true,"duration_ms":39935,"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":"By coupling the fast transport code RAPTOR with the inverse equilibrium solver FBT, a tokamak discharge can be simulated from its pulse schedule before it runs, and the resulting kinetic profiles yield improved coil-current predictions.","keywords":["kinetic equilibrium prediction","RAPTOR","FBT","tokamak transport","Grad-Shafranov equation","coil currents","H-mode transition","negative triangularity"],"falsifier":"Run the fully predictive workflow, with no post-shot data, on a set of TCV discharges outside the 211-shot training set, using the ad-hoc density model and a literal reading of the H98 scaling; then compare the predicted beta_N, li3, and PF coil currents against LIUQE kinetic reconstructions and measured coil currents. The central claim would be falsified if the signed errors in beta_N or li3 are comparable to the correction the coupling is supposed to provide, or if shots prepared with the FBT-RAPTOR traces do not show measurably smaller X-point misalignment than shots prepared with the stand","tokens_in":1417,"feed_emoji":"🧲","tokens_out":5094,"duration_ms":92033,"temperature":0.7,"pith_summary":"This paper builds a pre-shot Kinetic Equilibrium Prediction workflow by coupling RAPTOR, a fast 1.5D transport solver, to FBT, an inverse free-boundary Grad-Shafranov solver. It aims to show that a full TCV discharge, from current ramp-up to ramp-down across L-mode, H-mode, and a variety of shapes, can be simulated before the shot using only the pulse schedule, given an estimate of the confinement quality factor H98(y,2) and the line-averaged density. The two codes relax to a self-consistent answer in a few iterations, and feeding the predicted pressure and current-function profiles into FBT changes the computed poloidal-field coil currents by tens to hundreds of amperes, most importantly improving estimates of internal inductance and normalized beta. If correct, this gives tokamak operators a physics-based equilibrium for shot preparation instead of hand-tuned polynomial profiles, and the paper backs it with statistical comparisons over 211 discharges and two full experimental validations.","feed_headline":"Sharpen tokamak coil planning with pre-shot plasma forecasts","feed_subtitle":"RAPTOR-FBT coupling feeds kinetic profiles into equilibrium solvers, improving X-point alignment and beta_N estimates before each TCV pulse.","key_machinery":"The central object is the pair of free functions p'(psi) and TT'(psi) that enter the Grad-Shafranov equation. RAPTOR generates these profiles from the pulse schedule using a stiff logarithmic-gradient transport model whose pedestal gradients are controlled by a PI controller tied to H98(y,2) and line-averaged density; FBT solves the inverse free-boundary equilibrium with these profiles. The coupling loop iterates between the codes, seeding each FBT run with the previous plasma current distribution, until the FBT and RAPTOR profiles agree (alpha approximately 1), with residual systematic differences in poloidal beta below a few percent after two iterations.","core_discovery":"The paper's central claim is that the usual separation between pre-shot equilibrium preparation and transport prediction can be removed: RAPTOR's predicted p' and TT' profiles are used directly as the free functions in FBT's Grad-Shafranov solve, and iterating between the two codes produces self-consistent equilibria within a few minutes. With this coupling, the deliberately low poloidal beta and high q_a used in conventional FBT programming are corrected to realistic values, and the resulting feedforward PF coil currents differ from the standard preparation by tens to hundreds of amperes. In two experimentally repeated scenarios, the FBT-RAPTOR prepared shots kept the X-point closer to its","pith_inferences":["Inference: If the line-averaged density prediction is improved, the same workflow could run fully autonomously across a much larger scenario space, moving the demonstration from two validated cases to routine operations.","Inference: The same p'/TT' handoff could be reused in a tight-coupling mode or in real-time kinetic reconstruction, extending the benefit of better internal profiles from pre-shot planning to post-shot analysis.","Inference: The coil-current corrections driven by the Shafranov shift suggest a sensitivity test: how strongly do the prepared currents and beta_N estimates change per unit change in the assumed H98(y,2)? The paper does not quantify this sensitivity."],"forward_implications":["Tokamak operators can prepare feedforward PF coil traces from a physics-based equilibrium, reducing X-point and shape misalignment and lowering the risk of vertical displacement events.","The 211-shot benchmark indicates that a default parameter set predicts electron and ion stored energies within about 20% across a wide range of TCV scenarios, when the line-averaged density is known.","The extended H-mode threshold model lets confinement transitions be predicted automatically from the pulse schedule rather than assumed by the operator.","The coupled simulation runs in a few minutes per second of discharge, making pre-shot iteration feasible within the inter-shot latency at TCV.","More accurate internal inductance and normalized beta estimates give operators more realistic information about operational limits before a pulse begins."],"fun_headline_variants":["RAPTOR-FBT coupling pre-shots realistic equilibria and coil currents","Pre-shot kinetic equilibrium forecast improves TCV coil planning","Iterative transport-equilibrium loop corrects beta_N and l_i pre-shot","Self-consistent kinetic equilibria in minutes: RAPTOR meets FBT","RAPTOR-FBT pre-shot workflow sharpens X-point and coil current estimates"],"cache_read_input_tokens":30464,"weakest_assumption_plain":"The whole prediction rests on pre-shot guesses of the confinement quality factor H98(y,2) and line-averaged density; the paper's ad-hoc density model is admitted to be insufficient for large-database validation, so the large benchmark uses post-shot experimental density, and the full pre-shot workflow is demonstrated on only two cases.","fun_headline_variants_meta":{"raw":{"variants":["RAPTOR-FBT coupling pre-shots realistic equilibria and coil currents","Pre-shot kinetic equilibrium forecast improves TCV coil planning","Iterative transport-equilibrium loop corrects beta_N and l_i pre-shot","Self-consistent kinetic equilibria in minutes: RAPTOR meets FBT","RAPTOR-FBT pre-shot workflow sharpens X-point and coil current estimates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000277,"raw_usage":{"total_tokens":1486,"prompt_tokens":744,"completion_tokens":742,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":488,"completion_tokens_details":{"reasoning_tokens":645}},"tokens_in":488,"tokens_out":742,"duration_ms":7556,"temperature":1.0,"reasoning_tokens":645,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T19:41:01.536629+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the fully predictive workflow, with no post-shot data, on a set of TCV discharges outside the 211-shot training set, using the ad-hoc density model and a literal reading of the H98 scaling; then compare the predicted beta_N, li3, and PF coil currents against LIUQE kinetic reconstructions and measured coil currents. The central claim would be falsified if the signed errors in beta_N or li3 are comparable to the correction the coupling is supposed to provide, or if shots prepared with the FBT-RAPTOR traces do not show measurably smaller X-point misalignment than shots prepared with the stand","supporting_citations":[],"review_version":1}