{"id":"2ab3e8e0-aad5-43a6-9bd2-c2d4c23c4bac","arxiv_id":"2606.12857","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"New framework for robust GP calibration that leverages intermediate variables, S-GaSP discretization, and space-filling designs for joint emulator-discrepancy modeling, demonstrated on nuclear physics binding energies.","lead":"The paper proposes a Gaussian process calibration framework that uses intermediate variables from computer experiments to model discrepancies via a discretized S-GaSP and space-filling designs. This joint modeling approach is tested on nuclear binding energies and claimed to improve predictions and uncertainty quantification over standard methods.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Whether discretized S-GaSP plus space-filling points actually constrain discrepancy without new bias or identifiability issues is the least-secured step.","rationale":"The reader's weakest assumption directly matches the load-bearing step required for the joint-modeling claim to hold. Full text does not remove the need for an explicit check that the constraint mechanism itself, rather than other modeling choices, drives the reported gains.","tokens_in":1695,"tokens_out":294,"duration_ms":11817,"concrete_test":"Re-run the nuclear-physics calibration after replacing the proposed constraint points with an equal number of randomly sampled points from the same domain; if the reported improvement in predictive RMSE or interval coverage disappears or reverses, the space-filling + S-GaSP construction is not the source of the claimed benefit.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim requires that the selected intermediate variables, when used to build the discretized S-GaSP constraint, produce a discrepancy posterior that is both tighter and less biased than the unconstrained case. If the intermediate variables are only weakly informative or share latent structure with the emulator, the joint model can still suffer from the original non-identifiability while appearing better calibrated on the nuclear binding-energy data. The paper provides no analytic bound on the approximation error introduced by discretization or on the coverage of the space-filling design relative to the support of the true discrepancy.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes a robust Gaussian process calibration framework that uses intermediate variables via a structured selection process, a discretized scaled Gaussian stochastic process (S-GaSP) to constrain the discrepancy term, and space-filling designs for selecting constraint points. This enables joint modeling of the emulator and discrepancy function, with claimed improvements in predictive performance, principled uncertainty quantification, and reduced identifiability risks. The approach is demonstrated on a nuclear physics application involving binding energies, where it is reported to outperform baseline approaches.","tokens_in":1819,"tokens_out":492,"duration_ms":9157,"significance":"If the central claims hold with rigorous validation, the framework could meaningfully advance discrepancy modeling in computer model calibration by systematically incorporating often-unused intermediate variables, potentially offering a practical route to better identifiability and uncertainty quantification in applications like nuclear physics where such variables are abundant.","major_comments":[{"comment":"The abstract asserts outperformance on the nuclear binding-energy application but supplies no quantitative results, baseline definitions, error metrics, or validation details. The results section must include these (e.g., specific RMSE, coverage probabilities, or identifiability diagnostics) with clear definitions of the baselines to substantiate the central claim of improved performance and alleviated identifiability risks.","section":"Abstract and Results"},{"comment":"The efficacy of the discretized S-GaSP constraint plus space-filling design for producing a tighter, less biased discrepancy posterior without introducing new identifiability problems is load-bearing for the framework. The manuscript should supply either analytic approximation-error bounds or targeted simulation studies (e.g., on synthetic data where the true discrepancy is known) demonstrating that the selected intermediate variables do not share latent structure with the emulator in a way that undermines the constraint.","section":"Method (discretized S-GaSP and space-filling design)"}],"minor_comments":[{"comment":"Notation for the scaled Gaussian stochastic process (S-GaSP) and the discretization scheme should be introduced with explicit equations early in the methods section to improve readability.","section":"Methods"},{"comment":"The intermediate variable selection process is described as 'structured' but lacks a clear algorithmic outline or pseudocode; adding this would clarify reproducibility.","section":"Methods"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their constructive comments, which help clarify the presentation of our framework. We address each major comment below and indicate the revisions we will make.","responses":[{"response":"We agree that the abstract would benefit from quantitative indicators. In the revision we will insert concise performance metrics (RMSE reduction and coverage improvement relative to baselines) into the abstract. The results section (Section 4) already reports RMSE, 95% predictive coverage, and identifiability diagnostics (posterior variance ratios) for the nuclear binding-energy example; we will add an explicit table defining the three baselines (standard GP calibration, separate emulator-discrepancy fitting, and S-GaSP without intermediate-variable selection) and ensure every metric is labeled with its exact formula and data split.","revision_made":"yes","referee_comment":"[Abstract and Results] The abstract asserts outperformance on the nuclear binding-energy application but supplies no quantitative results, baseline definitions, error metrics, or validation details. The results section must include these (e.g., specific RMSE, coverage probabilities, or identifiability diagnostics) with clear definitions of the baselines to substantiate the central claim of improved performance and alleviated identifiability risks."},{"response":"We acknowledge that analytic error bounds for the particular discretization are not supplied and would be difficult to obtain in closed form. The manuscript does contain synthetic experiments (Section 3.3 and supplementary material) that recover known discrepancy functions, but these do not explicitly test for latent-structure overlap between selected intermediate variables and the emulator. We will therefore add a dedicated simulation study that generates data with controlled latent correlation between intermediate variables and the computer model, applies the full selection-plus-discretized-S-GaSP pipeline, and reports posterior bias and identifiability metrics. This targeted study will be included in the revised manuscript.","revision_made":"yes","referee_comment":"[Method (discretized S-GaSP and space-filling design)] The efficacy of the discretized S-GaSP constraint plus space-filling design for producing a tighter, less biased discrepancy posterior without introducing new identifiability problems is load-bearing for the framework. The manuscript should supply either analytic approximation-error bounds or targeted simulation studies (e.g., on synthetic data where the true discrepancy is known) demonstrating that the selected intermediate variables do not share latent structure with the emulator in a way that undermines the constraint."}],"tokens_in":1313,"tokens_out":516,"duration_ms":16371,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing to know is that this work combines structured selection of intermediate outputs, a discretized scaled Gaussian stochastic process to constrain discrepancy, and space-filling designs to enable joint modeling of emulator and discrepancy. That combination is presented as new and is applied to a nuclear physics calibration task.\n\nIt does a solid job recognizing that computer experiments often generate many intermediate variables that standard calibration ignores, and it tries to use them to tighten the discrepancy term without the usual separate fitting step. The nuclear binding energies case is a reasonable choice because those models produce exactly this kind of intermediate data.\n\nThe soft spot is the lack of clear evidence that the constraints actually reduce identifiability problems rather than mask them. The stress-test concern lands: without analytic bounds on discretization error or checks that the space-filling points cover the relevant discrepancy range, it is possible the joint posterior looks better on this dataset while still carrying the original non-identifiability or adding new bias from weakly informative intermediates. If the results section only shows outperformance on one application without multiple strong baselines, sensitivity to the selection step, or coverage diagnostics, that part stays under-supported.\n\nThis is aimed at people doing surrogate modeling and Bayesian calibration on physics or engineering codes that output many intermediates. A reader already working on similar nuclear or complex simulation problems could extract the selection procedure and constraint strategy to test on their own data.\n\nI would send it to peer review. The problem is real and the proposed integration is concrete enough to be worth referee scrutiny, even if the current validation needs tightening.","headline":"The paper's integration of intermediate-variable selection with discretized S-GaSP and space-filling designs for joint GP calibration is a practical step forward, but the nuclear binding-energy results do not clearly establish that it avoids new bias or identifiability issues.","tokens_in":2342,"tokens_out":408,"would_cite":false,"duration_ms":15189,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A framework uses selected intermediate variables and a discretized scaled Gaussian stochastic process to constrain the discrepancy term during joint emulator calibration.","keywords":["Gaussian process calibration","discrepancy modeling","intermediate variables","scaled Gaussian stochastic process","nuclear binding energies","model emulation","uncertainty quantification","identifiability"],"falsifier":"On the nuclear binding-energy data, the framework produces predictions or uncertainty intervals no better than those from separate emulator and discrepancy fitting.","tokens_in":2608,"feed_emoji":"","tokens_out":636,"duration_ms":11742,"temperature":0.7,"pith_summary":"The paper proposes a calibration method for imperfect computer models that incorporates intermediate simulation outputs, which standard approaches ignore. It combines a structured selection of those variables, a discretized scaled Gaussian stochastic process to limit the discrepancy, and a space-filling design to pick constraint locations. This setup supports fitting the emulator and discrepancy together rather than separately. The approach is shown to improve predictions and uncertainty estimates on a nuclear binding-energy example while reducing risks of non-identifiability between model components.","feed_headline":"Intermediate variables constrain discrepancy in GP calibration","feed_subtitle":"Joint emulator-discrepancy modeling via discretized S-GaSP and space-filling points improves predictions on nuclear binding energies over se","key_machinery":"The discretized scaled Gaussian stochastic process (S-GaSP) that uses selected intermediate variables to constrain the discrepancy term.","core_discovery":"By integrating a structured intermediate variable selection process, a discretized scaled Gaussian stochastic process to constrain the discrepancy term, and a space-filling design strategy for selecting constraint points, the framework enables joint modeling of the emulator and discrepancy in Gaussian process calibration, which improves predictive performance, provides principled uncertainty quantification, and alleviates identifiability risks, as demonstrated on a nuclear physics application involving binding energies where it outperforms baseline approaches.","pith_inferences":["The same selection and constraint steps could be tested on simulation outputs from other domains that produce many intermediate quantities, such as fluid dynamics or materials science.","If the space-filling constraint design is replaced by an adaptive choice based on current posterior uncertainty, the method might require fewer constraint points while maintaining the same calibration quality.","Checking whether the selected intermediate variables remain informative across different parameter regimes would provide a practical diagnostic for when the framework can be applied."],"forward_implications":["Joint modeling of emulator and discrepancy becomes feasible when intermediate variables are available.","Predictive performance on held-out observations improves relative to separate-fitting baselines.","Uncertainty quantification follows from the joint posterior rather than from post-hoc adjustments.","Identifiability risks between the computer model and the discrepancy decrease because the intermediate variables supply additional constraints."],"fun_headline_variants":["Intermediates aid GP discrepancy modeling","Discretized S-GaSP constrains discrepancy in GPs","Joint emulator-discrepancy modeling via S-GaSP","Space-filling strategy enhances GP calibration"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The intermediate variables carry information that can constrain the discrepancy without creating new identifiability problems or biases, and the discretized S-GaSP with space-filling points will produce effective constraints in practice.","fun_headline_variants_meta":{"raw":{"variants":["Intermediates aid GP discrepancy modeling","Discretized S-GaSP constrains discrepancy in GPs","Joint emulator-discrepancy modeling via S-GaSP","Space-filling strategy enhances GP calibration"]},"model":"grok-4.3","cost_usd":0.005701,"raw_usage":{"total_tokens":2685,"prompt_tokens":594,"num_sources_used":0,"completion_tokens":56,"cost_in_usd_ticks":57012000,"prompt_tokens_details":{"text_tokens":594,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2035,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":594,"tokens_out":56,"duration_ms":11701,"temperature":1.0,"reasoning_tokens":2035,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T06:24:54.034682+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"On the nuclear binding-energy data, the framework produces predictions or uncertainty intervals no better than those from separate emulator and discrepancy fitting.","supporting_citations":[],"review_version":1}