{"id":"f44d1f39-a3ef-498e-a285-c29267f0f3b7","arxiv_id":"2606.09778","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Intervention-aware training of variational quantum circuit policies in building control reduces pre-filter violations and safety-layer reliance compared to classical policies while maintaining energy performance.","lead":"The paper introduces a training method for quantum policies that penalizes reliance on downstream safety filters to ensure the policy itself learns safety. This allows measuring whether safety comes from the learned controller or its protective layers in building control tasks.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Primal-dual budget may fail to isolate policy safety if differentiable CBF projection leaks corrective gradients into VQC training","rationale":"The reader's weakest_assumption already flags exactly this isolation risk between the differentiable CBF, the intervention budget, and the attribution protocol. No other assumption in the abstract is more load-bearing for the 'who earns the safety' claim; the full text would need explicit checks (e.g., gradient-flow ablation) to close it. Hence the UNVERDICTED status is unaffected.","tokens_in":1801,"tokens_out":308,"duration_ms":16441,"concrete_test":"Re-train the VQC policy with the CBF projection replaced by a non-differentiable hard clip (while preserving the exact primal-dual intervention penalty and all other hyperparameters); then repeat the guard-off BOPTEST evaluation on the same 5 seeds. If the pre-filter violation reduction loses statistical significance, the original attribution does not isolate policy-level safety.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that intervention-aware training produces intrinsically safer policies rather than policies that merely exploit the differentiable CBF projection's repair during gradient updates. Because the projection remains inside the differentiable loop while the primal-dual penalty only counts interventions, policy gradients can still route through the projection's correction term. The post-hoc safety-attribution protocol decomposes executed trajectories but cannot retroactively remove training-time gradient bias. This directly threatens the assertion that guard-off evaluation confirms policy-level gains.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces Intervention-Aware Variational Quantum Differentiable Predictive Control (IA-VQC-DPC), which trains a compact VQC policy for building control under a primal-dual intervention budget that penalizes reliance on a differentiable Control-Barrier-Function (CBF) projection. It further proposes a safety-attribution protocol that decomposes trajectory corrections into CBF and runtime-guard terms and uses guard-off evaluation. On BOPTEST emulators (5 seeds, 60 episodes), the authors claim intervention-aware training yields significantly lower raw pre-filter violations and safety-layer reliance (both p < 10^{-4}) with no energy regression, that the quantum policy outperforms a matched classical policy at ~400 parameters, and that guard-off evaluation confirms the gains are policy-intrinsic rather than filter-dependent.","tokens_in":1902,"tokens_out":768,"duration_ms":19020,"significance":"If the attribution protocol and isolation of policy-level safety hold, the work supplies a concrete, measurable criterion for determining whether safety in filtered learned controllers originates in the policy itself. This is relevant for reliable deployment of ML controllers in constrained systems. The fixed-parameter quantum-vs-classical comparison and the negative result on the learned energy head provide useful benchmarks. The protocol is presented as domain-general. The use of closed-loop high-fidelity emulators and reported statistical tests are positive features, though the absence of full statistical details limits immediate impact.","major_comments":[{"comment":"Methods, IA-VQC-DPC training procedure: the primal-dual intervention budget penalizes post-projection interventions, yet the differentiable CBF projection remains inside the gradient loop. The manuscript provides no explicit statement, computational graph, or detachment operation showing that policy gradients are prevented from routing through the projection's corrective term. Because the post-hoc attribution protocol operates only on executed trajectories, it cannot retroactively correct for training-time exploitation of the projection; this directly threatens the central claim that guard-off evaluation isolates policy-level safety improvements.","section":"Methods (IA-VQC-DPC training and primal-dual budget)"},{"comment":"Abstract and Results (empirical claims): the reported p < 10^{-4} values for violation reduction and safety-layer reliance rest on 5 seeds and 60 episodes but supply neither error bars, confidence intervals, exact statistical test description, nor verification that the attribution decomposition was applied consistently across guard-on/guard-off conditions. Without these, the quantitative support for \"significantly safer\" and \"policy-level\" cannot be assessed.","section":"Abstract and Results section"},{"comment":"Evaluation protocol: the claim that guard-off evaluation confirms intrinsic policy safety assumes the runtime guard is the only external corrective mechanism, yet the differentiable CBF projection used at training time may have already shaped the policy parameters. A concrete test (e.g., ablation removing the projection entirely from the training graph) is needed to substantiate that the observed guard-off improvement is not an artifact of gradient leakage.","section":"Evaluation and guard-off protocol"}],"minor_comments":[{"comment":"Notation for the safety-attribution decomposition should be introduced with an explicit equation rather than prose description to allow readers to verify the CBF versus guard term split.","section":"Safety-attribution protocol"},{"comment":"The manuscript should state the precise form of the primal-dual Lagrangian and the update rules for the dual variable to make the intervention budget reproducible.","section":"Methods"}],"recommendation":"major_revision","confidential_remarks":"The quantum aspect appears secondary to the safety-attribution contribution; the journal may wish to confirm fit with its scope if the primary novelty is the attribution protocol rather than any quantum-specific advantage."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments that help clarify the training procedure, statistical reporting, and evaluation protocol. We address each major comment below and commit to revisions that strengthen the manuscript without altering its core claims.","responses":[{"response":"We agree that an explicit description of the computational graph is required. In the revised Methods, we will add a figure and text specifying that the CBF projection's corrective term is detached via stop-gradient before the policy loss is computed; only the post-projection intervention magnitude enters the primal-dual budget. This prevents gradient flow through the correction while still penalizing reliance, ensuring the policy cannot exploit the projection during training. The guard-off protocol then evaluates the resulting policy parameters on trajectories without any runtime guard.","revision_made":"yes","referee_comment":"Methods (IA-VQC-DPC training and primal-dual budget): the primal-dual intervention budget penalizes post-projection interventions, yet the differentiable CBF projection remains inside the gradient loop. The manuscript provides no explicit statement, computational graph, or detachment operation showing that policy gradients are prevented from routing through the projection's corrective term. Because the post-hoc attribution protocol operates only on executed trajectories, it cannot retroactively correct for training-time exploitation of the projection; this directly threatens the central claim that guard-off evaluation isolates policy-level safety improvements."},{"response":"We accept this point. The revision will report mean ± standard deviation across the 5 seeds, 95% confidence intervals, the exact test (two-sided paired t-test on per-episode metrics), and confirmation that the attribution decomposition (CBF term vs. runtime-guard term) was applied identically in both guard-on and guard-off conditions. These details will be added to the Results section and supplementary material.","revision_made":"yes","referee_comment":"Abstract and Results (empirical claims): the reported p < 10^{-4} values for violation reduction and safety-layer reliance rest on 5 seeds and 60 episodes but supply neither error bars, confidence intervals, exact statistical test description, nor verification that the attribution decomposition was applied consistently across guard-on/guard-off conditions. Without these, the quantitative support for \"significantly safer\" and \"policy-level\" cannot be assessed."},{"response":"The referee correctly identifies that an ablation removing the CBF projection from the training graph would provide direct evidence against gradient leakage. We will add this ablation experiment in the revision: we retrain the VQC policy with the projection completely excised from the graph (i.e., no differentiable CBF at training time) while retaining the primal-dual budget on raw violations, then compare guard-off performance against the original IA-VQC-DPC. Results will be reported alongside the existing guard-off curves.","revision_made":"yes","referee_comment":"Evaluation protocol: the claim that guard-off evaluation confirms intrinsic policy safety assumes the runtime guard is the only external corrective mechanism, yet the differentiable CBF projection used at training time may have already shaped the policy parameters. A concrete test (e.g., ablation removing the projection entirely from the training graph) is needed to substantiate that the observed guard-off improvement is not an artifact of gradient leakage."}],"tokens_in":1675,"tokens_out":685,"duration_ms":19470,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that this work makes the question of who earns the safety measurable by training a VQC policy under a primal-dual penalty on CBF interventions and then decomposing trajectory corrections afterward.\n\nWhat is new is the intervention-aware objective paired with the attribution protocol and guard-off evaluation, applied specifically to variational quantum policies in BOPTEST building control. The experiments show lower pre-filter violations and less total safety-layer use after intervention-aware training, with the quantum policy also beating a matched classical one on safety and comfort at roughly 400 parameters and no energy regression.\n\nThe paper does a solid job setting up the closed-loop tests on the external emulator and reporting p-values below 10^-4 across seeds. The guard-off check is a direct way to test whether gains are policy-level rather than filter-level.\n\nThe soft spot is the stress-test concern about gradient flow. Because the CBF projection stays differentiable and inside the training loop, policy gradients can still route through its corrections even while the primal-dual term only counts interventions. The post-hoc attribution decomposes executed trajectories but cannot remove that training-time bias, so the claim that the policy itself became safer rests on whether the methods block this path. The abstract gives limited statistical detail beyond the p-values, which makes effect sizes harder to judge.\n\nThis is aimed at researchers working on safe learned control, especially those using quantum circuits or building-energy applications. It has a clear framing, a reproducible testbed, and empirical results that are worth checking, so it deserves a serious referee to examine the implementation and the isolation of the effect.\n\nI would send it to peer review.","headline":"The paper gives a workable protocol for checking whether safety comes from the policy or the filter in quantum building control, but the gradient-leakage worry needs direct verification in the methods.","tokens_in":2359,"tokens_out":412,"would_cite":false,"duration_ms":20085,"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":"Training quantum controllers with an explicit penalty on filter interventions makes the policy itself responsible for safety rather than downstream guards.","keywords":["quantum control","safety filters","variational quantum circuits","control barrier functions","building energy management","intervention-aware training","policy attribution"],"falsifier":"A replication in which guard-off evaluation shows the same violation rates as guard-on evaluation after intervention-aware training, or in which classical and quantum policies exhibit statistically indistinguishable safety metrics at equal parameter count.","tokens_in":2692,"feed_emoji":"⚛️","tokens_out":656,"duration_ms":10220,"temperature":0.7,"pith_summary":"The paper identifies a measurement problem in safe control: hard safety filters can mask an incompetent policy so that success is credited to the filter, not the learner. It introduces an intervention-aware training loop that budgets and penalizes reliance on a differentiable control-barrier projection while a safety-attribution protocol decomposes executed corrections into policy versus guard contributions. On closed-loop building emulators the resulting quantum policy shows statistically lower pre-filter violations and lower total safety-layer use with unchanged energy cost; at matched parameter count it also outperforms a classical counterpart. Guard-off tests confirm the gains are policy-level. The attribution method itself is presented as domain-general.","feed_headline":"Quantum policies learn to earn safety instead of borrowing it from filters","feed_subtitle":"Intervention budget during training cuts pre-filter violations and safety-layer reliance on building emulators, with quantum outperforming c","key_machinery":"Intervention-aware training under a primal-dual intervention budget combined with a safety-attribution protocol that decomposes trajectory corrections into CBF term and runtime-guard term.","core_discovery":"Intervention-Aware Variational Quantum Differentiable Predictive Control trains a compact VQC policy under a primal-dual budget that penalizes CBF-projection interventions and is scored by a safety-attribution protocol that isolates policy versus runtime-guard corrections; on BOPTEST emulators this yields significantly lower raw pre-filter violations and safety-layer reliance (p < 10^-4) with no energy regression, and the quantum policy is safer and more comfortable than a matched classical policy at equal parameter budget.","pith_inferences":["The protocol may allow designers to trade off filter complexity against policy complexity during training.","Negative result on the energy head suggests that distribution-aware guards remain necessary even after intervention-aware training.","Similar attribution could be used to audit other safety-filtered learning pipelines beyond building control."],"forward_implications":["The policy can be deployed with reduced or removed runtime guards while still meeting constraints.","Safety credit can be assigned to the learned controller rather than the protective wrapper.","Quantum policies can be compared directly to classical ones on safety earned rather than safety filtered.","The same attribution protocol can be applied to non-quantum learned controllers in other domains."],"fun_headline_variants":["Intervention budget trains quantum policy to cut filter reliance","Safety attribution reveals policy-level gains on building emulators","Quantum policy lowers pre-filter violations versus classical match","Intervention-aware training reduces safety layer use with no energy loss"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The primal-dual budget and attribution protocol correctly measure policy-level safety gains without the differentiable projection or guard introducing hidden biases in the closed-loop tests.","fun_headline_variants_meta":{"raw":{"variants":["Intervention budget trains quantum policy to cut filter reliance","Safety attribution reveals policy-level gains on building emulators","Quantum policy lowers pre-filter violations versus classical match","Intervention-aware training reduces safety layer use with no energy loss"]},"model":"grok-4.3","cost_usd":0.003681,"raw_usage":{"total_tokens":1958,"prompt_tokens":756,"num_sources_used":0,"completion_tokens":60,"cost_in_usd_ticks":36812000,"prompt_tokens_details":{"text_tokens":756,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1142,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":756,"tokens_out":60,"duration_ms":7500,"temperature":1.0,"reasoning_tokens":1142,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T16:20:19.980099+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A replication in which guard-off evaluation shows the same violation rates as guard-on evaluation after intervention-aware training, or in which classical and quantum policies exhibit statistically indistinguishable safety metrics at equal parameter count.","supporting_citations":[],"review_version":1}