Papers

Electric arc furnaces inject stochastic reactive-power swings that cause voltage flicker at the point of common coupling. This paper compares two STATCOM inner controllers for an 11~kV, 50~Hz furnace feeder: a split-frequency DQ-PI design and a Lyapunov-based nonlinear design whose integral state is retained in the storage function, giving global asymptotic stability of the nominal unsaturated current-error subsystem. Gains for both controllers were tuned by a frozen metaheuristic campaign and then evaluated over 600~s with an IEC~61000-4-15 digital flickermeter. A post-campaign audit showed that the stochastic arc realization used for the original comparison was periodic, so the two selected gain sets were frozen without retuning or reselection and evaluated on ten independent stochastic realizations that had been specified and archived before the STO-ENS validation campaign, together with a paired uncompensated case for each realization. Both controllers reduce the mean short-term flicker severity by about 62% relative to the STO-ENS uncompensated baseline, and no statistically detectable difference in mean flicker severity is found between them. They separate instead in realization sensitivity: the Lyapunov standard deviation across arcs is roughly one third that of DQ-PI. Controller ordering inferred from a single stochastic trajectory is therefore not a reliable measure of stochastic robustness, and ensemble validation is necessary.

We study the asymptotic response time tail in the M/G/n queue with heavy-tailed (regularly varying) job sizes. For this setting, strongly tail optimal policies are known only at lower load, where n-1 servers suffice for stability. At higher load, the tail optimal scheduling has been open. This paper presents DynaSPLIT, the first strongly tail optimal policy at high load. Together with prior work, this result shows that strong tail optimality is achievable across the entire stability region of the M/G/n with heavy-tailed job sizes.

Thermalization in closed quantum systems and its breakdown are central to understanding the emergence of statistical behavior in isolated many-body systems. Kinetic constraints can hinder thermalization by restricting the allowed dynamics and giving rise to non-ergodic behavior. While various aspects of ergodicity breaking in kinetically constrained models have been extensively studied in Hermitian systems, their behavior in non-Hermitian kinetically constrained models remains less explored. In particular, the inequivalence of left and right eigenstate ensembles in non-Hermitian systems raises the question of whether they exhibit qualitatively different dynamical and eigenstate properties. In this work, we study non-Hermitian generalizations of the paradigmatic PXP and PPXPP models, focusing on their Krylov complexity, eigenstate structure, and dynamics. By exploiting the Arnoldi's algorithm, we find distinct behavior of Krylov complexity in the left and right eigenstate ensembles as non-Hermiticity is increased: in the right ensemble, it exhibits persistent oscillations with an increasing period, while in the left ensemble, it gradually saturates. The characteristic arch-shaped profile of the Arnoldi coefficients is also modified differently in the two ensembles, with a Hermitian-like arch re-emerging in the left ensemble at large non Hermiticity. Additionally, in the non-Hermitian PXP model, the right ensemble retains the characteristic scar structure, while the left ensemble develops additional non-scar eigenstates with anomalously low entanglement and large overlap with the vacuum state. These findings reveal that non-Hermiticity can affect Krylov dynamics and eigenstate structure differently in the left and right ensembles, demonstrating the need to treat them separately when characterizing non-Hermitian kinetically constrained systems.

Role-playing language models speak as characters in games and agent simulations, where the system records what each character can hear, see and understand. We test whether characters respect these boundaries in 480 scenes, each with a counterfactual twin that changes only the hidden fact, so mentions of the twin's fact measure guessing. Across 11 open-weight models, characters use content from speech they cannot understand or hear at 67 and 72 percentage points (pp) above this baseline, although most models correctly state what the character perceives. This is a curse of knowledge. The leak extends to hidden actions and private messages and changes where characters search. Five of six proprietary models rarely leak, but compliance in strong models depends on input format. Changing only the scene layout or character description raises the rate in two otherwise compliant open-weight models from under 10 to over 60 pp, and combined changes raise two proprietary models from near zero to 13 and 21~pp. Writing the line as the scene's author leaks nearly as much as role-play, and thinking rarely helps. Giving models only what the character observes stops the leak without hiding what it may report.

Household robots must successfully navigate im-perfect circumstances, both in their internal models and the external world. We introduce a benchmark of seven long-horizon household tasks in Habitat that vary initial robot memory and environment configurations. We evaluate adaptations of ReAct, VLM-TAMP, and Tru-POMDP within a shared execution framework across 90 tasks-environment-memory combinations, totaling 1,350 episodes. Our evaluation measures task success, partial completion, and action efficiency, along with rule-based analysis of execution traces. ReAct achieves the highest overall task success (45.3%) and completion (71.7%), while having accurate initial memory does not consistently yield the highest performance for every planner. We identify recurring placement errors, precondition errors, and redundant actions through trace analysis. These findings highlight differences in how the evaluated planners use initial knowledge and respond to uncertainty.

We show how the methods of delta geometry can be used to bound the degree of points in the Zilber–Pink locus of a subvariety of an abelian variety, if the varieties are defined over a number field and we restrict to points with bounded ramification at a fixed prime.

AI-generated outputs are frequently perceived as highly creative, but also tend to be more similar to each other (homogenized) than human ideas. Does the human heterogeneity that extends beyond AI reflect valuable thinking that AI misses? Alternatively, is greater human heterogeneity driven by lower-quality ideas and less effective thinking that AI avoids? We developed Expansion of Divergent Generative Expression (EDGE) to map AI's semantic "blind spots" by assessing how strongly/weakly individual human ideas were represented in a task-matched AI semantic comparison set. Specifically, we compared human-generated vs. AI-generated ideas across three open-ended tasks: college admissions essays, alternative uses for everyday objects, and interpersonal advice. Across all three tasks, human responses extended into semantic regions sparsely covered by AI outputs. These distinctively human responses showed no meaningful deficits in creativity, quality, or effectiveness, were associated with equal-or-better academic performance, and with credibly superior ratings and achievement for some outcomes. These ideas were thus not merely poor-performing variation; they performed as well as or better than ideas that overlapped more strongly with AI ideas. More distinctive human ideas tended to receive higher creativity ratings, but this relationship was weak or reversed for AI ideas. Taken together, these results suggest that differentiation among human-generated ideas does not reflect poor thinking that AI can safely omit, and instead may afford both individual and population-level advantages.

Face recognition systems are deployed widely enough that a photograph shared online can be matched without its subject's consent. Generator-based protection methods rely on heavy pretrained backbones, multi-stage inference or visible semantic edits. We introduce SSM-Attack, a black-box face privacy framework built on Selective State Space Models. Rather than synthesizing the perturbation, the state-space model predicts in one linear-time pass a spatial allocation of the perturbation budget, marking which facial locations carry identity evidence; a short projected refinement then sets the values within this per-pixel budget. Against published benchmark results, SSM-Attack attains the highest average black-box attack success rate on all three evaluation sets: 87.10 percent on CelebA-HQ, 86.90 on LADN and 81.15 on FFHQ, the last from a network that never saw that dataset. The allocation network holds 5.73 million parameters, or 16.13 million with the frozen networks used by its quality terms. Under the same cap of 0.10 on the maximum per-pixel change, allocation improves structural similarity by 0.044, peak signal-to-noise ratio by 1.62 decibels and Frechet Inception Distance by 11.2 points over a uniform map while spending 37 percent of its capacity. Strong facial privacy protection therefore does not require a large generative prior; it requires knowing where to spend a small one.

Recent efforts to detect non-Markovian memory have become increasingly important, motivated by the broad relevance of memory effects in quantum information processing, metrology, and thermodynamic engines. Here, we show that interferometric methods, particularly the phase shift and visibility, provide an experimentally accessible framework for probing environmental effects and witnessing non-Markovian dynamics. Specifically, suitably constructed witness operators enable the detection of non-Markovianity through measurable phase shifts in the interference pattern. Furthermore, we prove that the interferometric visibility is monotonically non-increasing under incoherent CP-divisible (Markovian) dynamics. Consequently, any revival of the optimized visibility constitutes a sufficient signature of information backflow and hence non-Markovianity. The effectiveness of the proposed phase-based and visibility approaches is illustrated for representative examples of non-Markovian dynamics. Our results offer a simple and experimentally feasible framework for probing memory effects, with potential applications in quantum sensing and noisy intermediate-scale quantum devices.

As demand for LLM inference grows and consumer GPUs grow more capable, edge deployment is becoming increasingly attractive. WebGPU is compelling in this setting because it offers cross-platform portability and browser accessibility. However, these same strengths also make high performance difficult to achieve, as an implementation that performs well in one context may not perform well in another. To address this, we present GPU Goldmines, a framework for specifying, executing, and tuning WebGPU kernels. We instantiate GPU Goldmines with kernels for key LLM building blocks: matrix multiplication, matrix-vector multiplication, and flash attention. We characterize performance across diverse devices, shapes, and browsers, and use the resulting database to accelerate subsequent autotuning and configuration selection. On shapes drawn from Llama and Qwen models, our matrix multiplication kernels outperform ONNX Runtime and llama.cpp WebGPU kernels by 7.0x and 1.41x geometric mean, respectively, and deliver end-to-end improvements of 1.48x in the prefill stage of llama.cpp for chat workloads.

Stochastic optimization models sequential decision-making under uncertainty, typically assuming known input distributions. When these distributions are unknown, \citet{agarwal2026semibandit} give an online-to-offline reduction for monotone stochastic optimization under semi-bandit feedback, where only probed variables are observed. Its expected regret relative to the offline guarantee is $\widetilde O(n\sqrt T)$, where $n$ is the number of variables and $T$ the horizon. For optimal stopping and Pandora's box, \citet{liu2025improved} achieve $\widetilde O(\sqrt{nT})$ using problem-specific algorithms. Can a general online-to-offline reduction achieve this improved rate? We give such a reduction for strongly monotone stochastic optimization, answering the corresponding open question. Strong monotonicity is defined relative to the chosen offline algorithm. Our oracle-efficient reduction achieves $\widetilde O(\sqrt{nT})$ expected regret while preserving approximation guarantees for oracles satisfying the same condition. It combines Bernstein-type optimistic distribution estimates with a Hellinger stability bound for adaptive policy transcripts, aggregating squared marginal errors weighted by the executed policy's probing probabilities. The rate is optimal up to logarithmic factors for $T\ge n$, even under full feedback. Under monotonicity and uniformly bounded coordinate-wise variation, the same algorithm achieves $\widetilde O(n^{3/4}\sqrt T)$ expected approximation regret, improving the previous linear dependence on $n$. The exponent $3/4$ is sharp for its ordinary regret over instances and exact oracle choices. Our framework applies to prophet inequality, Pandora's box, best-of-$K$ selection, stochastic knapsack, stochastic orienteering, and stochastic submodular maximization.

We develop a Sobolev theory for the kinetic Fokker-Planck equation subject to a singular Morrey-form $L^\infty+M^q_K$ force field which can produce local blow-ups due to strong attraction to a submanifold. Our approach is based on the fractional representation for the Duhamel series of the equation and on establishing kinetic Adams-type estimates. Accordingly, we construct weak solutions of Brownian-driven kinetic SDEs in a critical Morrey set-up, obtain kinetic Krylov-type estimates and a strongly continuous Feller propagator, and establish uniqueness criteria for a large class of weak solutions.

Sampling multiple reports creates an inference problem: how should sampled evidence be converted into a final decision, and does a better structured decision yield a better report? We study these questions through a framework that separates the decision signal, scoring objective, and feasible decision set, followed by text realization. On MIMIC-CXR and IU-Xray, controlled comparisons across 13 generator configurations per benchmark under grouped development/test evaluation reveal three findings. First, changing the objective under fixed evidence and rule capacity changes predictions and their held-out metric trade-offs. Second, relaxing candidate constraints has objective-dependent effects under shared signals and frozen scores. Diagnostics connect the IU macro-oriented assertion increase to near-zero label thresholds and show that limited candidate coverage coexists with selection headroom. Third, a controlled realization study with SFT-Qwen3-VL-8B shows that decision-level gains can be attenuated, amplified, or reversed. Change decomposition reveals that realization can remove incorrect assertions while losing correct findings, so target fidelity and reference accuracy need not improve together. Finding gains also do not consistently improve reference-based report quality. These results motivate separate evaluation of decision rules, candidate constraints, and realized reports.

Let $G$ be a split connected semi-simple group over a local non-archimedean field $F$. In this paper we study the action of Hecke operators on the Schwartz space of half-densities on the moduli stack $\bpo$ of $G$-bundles on ${\mathbb P}^1$ endowed with a trivialization at $0$ and $\infty$. This space is a representation of $G(F)\times G(F)$ which contains the regular representation of $G(F)$. We show that this inclusion is an isomorphism on the cuspidal part. In addition we show that every irreducible square-integrable representation $\pi$ of $G(F)$ defines a Hecke eigenvalue. As a byproduct we show that every irreducible square-integrable representation of $G(F)$ with an Iwahori fixed vector defines a ${\mathbb C}$-point on the Calogero-Moser space associated to $G$. We also show that for $G=PGL(N)$ the eigenvalues of certain "simplest non-trivial" Hecke operators are essentially equal to the values of the character of $\pi$ on regular elliptic elements. These results are parallel to some of the results of our previous publication where some analogous statements are proved in the case when $F$ is a finite field.

Deep unrolling methods for inverse problems augment traditional reconstruction algorithms with learned priors to improve their performance. These methods have been developed under the implicit assumption that the geometry of the measurement operator does not change between training and deployment; when this is no longer the case, the quality of obtained reconstructions can degrade significantly. In this paper, we argue that this is not an inherent limitation of the deep unrolling paradigm and can be mitigated by a careful choice of the underlying iterative algorithm. We propose a deep unrolling framework for the phase retrieval problem that augments the alternating projections method with a learned denoising step. Our method obtains high-quality reconstructions without the need for retraining in settings where conventional model-based alternatives struggle, such as when the dynamic range of the measurement operator changes significantly at test time.

Over the last decade, the emerging field of two-dimensional van der Waals magnetism has focused mainly on long-range-ordered magnetic phases. Disordered magnetic phases, however, remain largely unexplored. Here, we report a highly anisotropic Griffiths phase driven by self-intercalation in the van der Waals magnet Fe5GeTe2. We find evidence of short-range-ordered ferromagnetic clusters embedded in a paramagnetic matrix, which dominate the magnetic response between the Curie temperature and a higher characteristic Griffiths temperature. An anomalous power-law divergence in the magnetic susceptibility, universal Lee Yang scaling, and distinct entropy anomalies provide further evidence for this phase. Strikingly, the non-analytic Griffiths response emerges exclusively under in-plane magnetic fields, whereas out-of-plane fields abruptly alter this behavior. This directional dependence indicates that the local magnetic interactions within nanoscale clusters are closely coupled to the layered lattice geometry. Our findings establish a regime of dimensionally constrained, disorder-driven magnetism and identify self-intercalated van der Waals magnets as a platform for investigating the interplay between magnetic disorder, anisotropy, and reduced dimensionality.

Deep learning has become the dominant form of machine learning due in part, to its ability to automatically engineer complex meta features from data. However, even deep learning features can perform erratically because of unobserved latent factors. incorporating latent factor information explicitly, can improve performance by letting the model alter its behavior based on such inferred factors. Because neural networks often capture such complexity inherently, external clustering of the original feature space offers limited benefit. clustering observations by feature importance profiles, rather than raw features, offers a principled way to uncover latent structure, since observations with similar importance patterns may share predictive dynamics. In a prior work, Feature Oriented Representation with Clustering and Explanation (FORCE) we showed that this approach measurably improved neural network performance. In this article, we present JEDI-net(Joint Embedding with Dynamic Integration Network) a unified architecture that periodically clusters SHAP values during optimization to identify latent subgroups with similar feature importance patterns, embed their identities directly into the network, allowing predictions to adapt to each observations predictive structure.

Aligned large language models (LLMs) remain vulnerable to jailbreak prompts that circumvent their safety guardrails, making the identification of these vulnerabilities important for understanding and mitigating model weaknesses. Existing automated jailbreak methods explore prompt candidates or reuse attack experience, but generally do not learn over a persistent hierarchy that connects related jailbreak strategies, such as role-play, encoding, and obfuscation. This leaves the relationships among reusable strategies underused in guiding adaptive attack selection. We present Tree-Guided Reinforcement-Augmented Probing (TRAP), a hierarchical framework for automated jailbreak strategy discovery that requires only black-box access to the target LLM. TRAP organizes established jailbreak methods and successful attack templates into a semantic strategy tree, connecting broad attack approaches to concrete, reusable strategies. A reinforcement-learning policy based on a tree-structured gradient bandit explores this tree and uses evaluator feedback to update strategy preferences along each selected path. The policy retains these preferences across attack objectives, allowing experience from earlier attempts to guide subsequent strategy selection. Retrieval-augmented prompt construction translates the selected strategy into an attack for the current objective using relevant successful examples. Experiments on AdvBench across seven aligned LLMs show that TRAP outperforms its counterparts on six models under the reported method settings and the common primary evaluator, achieving attack success rates of 45.6% to 97.9% on these models. These findings support hierarchical strategy learning as an effective approach to discovering jailbreak vulnerabilities in aligned LLMs.

This paper proposes a method for constructing a neural network that achieves perfect classification of all samples in a given training set, without taking regularization into account. The network is constructed on-demand, with its layer sizes and neuron parameters determined through a geometrical analysis of the input data, in which separating hyperplanes are created and mapped to neurons with step activation functions. The proposed architecture consists of two hidden layers with distinct roles: inter-class separation and transformation of the data into a linearly separable representation. Inter-class separation requires every bi-chromatic pair of samples belonging to different classes to be separated by at least one hyperplane. We show that when this separation is achieved through a hierarchical partitioning of the data, the leaf nodes of the oblique decision tree, obtained as a byproduct, directly define the neurons of the linear separability layer, whose size is only one neuron greater than that of the inter-class separation layer. Experiments on the MNIST dataset demonstrate that the proposed method achieves perfect training-set classification, and compact models. The direct correspondence between the decision tree and network neurons enhances interpretability and explainability, while the construction procedure offers opportunities for parallel implementation and improved scalability.

We identify the $(\infty,1)$-category presented by the dagger Bergner, dagger Joyal, and dagger Rezk models of Parts I and II with that of univalent flagged dagger $(\infty,1)$-categories. We first construct a realization–nerve adjunction between reversal simplicial anima and Hermitian flags. We prove that its nerve restricts to an equivalence between flagged dagger $(\infty,1)$-categories and dagger Segal anima, under which univalence corresponds to unitary completeness. We then construct the adjunctions for univalentization and Hermitian completion and identify univalentization with unitary completion.

Language-model agents are expected to adapt to others whose behavior changes through interaction. Yet adaptation may rely on learned behavioral patterns rather than theory of mind (ToM): the ability to infer others' mental states, including beliefs and intentions. We introduce a benchmark and diagnostic framework for repeated hide-and-seek against algorithmic opponents with explicitly defined recursive belief updates. Their ground-truth beliefs and action probabilities let us distinguish adaptive play, action prediction, and belief tracking through live interaction, fixed-history evaluation, and controlled information interventions. With structured elicitation and behavioral supervised fine-tuning (SFT), we find improved play in several tested settings, but diagnostics reveal limitations in opponent prediction and belief tracking. We test whether supervising opponent action probabilities can address this predictive deficit in Qwen2.5-7B-Instruct. Predictions improve when the model is told the opponent's type and parameter settings, but become less accurate when it must infer the opponent from history without this information. Training on examples that also omit these details reduces the gap, although improvements vary across opponents. Is this limitation specific to smaller open-weight models, or does it also appear in a frontier model? GPT-5.4 also benefits from correct opponent beliefs: supplying them improves its predictions and choices, whereas deeper explicit belief reports and self-generated memory do not consistently improve forecasting from interaction history. This contrast suggests that using useful belief information is easier than inferring it autonomously, even for a frontier model. Together, our findings distinguish improved play and action prediction from reliable belief tracking, providing a framework for evaluating mentalizing and testing targeted improvements.

We introduce StartLux-Decision, decision models from 0.8B to 27B parameters that combine efficient single-pass prediction with adaptive, confidence-guided reasoning. The Instant models use bidirectional attention in full-attention layers to improve decision accuracy while keeping single-pass inference. The AutoThink models invoke explicit reasoning for uncertain decisions and answer confident cases directly. A three-stage training pipeline establishes shared decision capabilities, specializes the two variants, and equips AutoThink with reasoning through reasoning supervision and reinforcement learning. We introduce Reinforcement Learning for Calibrated Deliberative Decisions (RLCD$^2$), which uses a frozen model and proper scoring rules to reward improvements in decision probability quality and applies a soft penalty to overlong reasoning traces. RLCD$^2$ raises the share of reasoning traces that terminate within the generation limit by up to 48 percentage points and cuts average reasoning length by up to 65%, while largely preserving decision accuracy. On the public subset of Decision Index 0.3, the Instant models outperform the strongest evaluated comparable-scale open-weight decision models by 7.4-16.9 points, and the 27B AutoThink model reaches 73.4. We also introduce DecisionPivot, a programmatically verifiable benchmark of computational decisions built from contrast pairs that differ in a single answer-changing detail. On DecisionPivot, AutoThink outperforms its Instant counterparts by 11.5-18.4 points in pair accuracy and leads among evaluated systems. The models run on vLLM, llama.cpp and MLX, with quantized variants for lower memory use.

This work studies certified reach-avoid control of stochastic hybrid systems in which a stochastic differential equation governs continuous flow and a controlled stochastic kernel governs guard-triggered resets. The input appears only in the reset kernel, where it determines the post-jump distribution. Certificates for such systems have so far assumed spontaneous or controller-chosen jump times, a finite horizon, or discrete-time dynamics; we certify, over the infinite horizon, that the target is reached before the unsafe set with at least a prescribed probability. First, we introduce reach-avoid contact certificates, whose flow and jump conditions are imposed only on a sublevel corridor, and prove that any such certificate with a compatible reset policy meets the specification; a pointwise safety filter extends this guarantee from verification to control by constraining arbitrary nominal controllers. Next, we prove that executions are non-Zeno and bound the expected settling time and reset count. We frame certificate search as a linear program over any twice-differentiable dictionary followed by independent verification. We illustrate our approach via comparisons where the certified region is a subset covering 93.6% of the ground truth and where our certificate-based safety filter reaches the goal while reducing collisions from 68 to zero across 200 episodes of a dribbling ball problem.

We study the rough differential equation $$ \frac{du_t}{dt}= F(u_t) \zeta_t, \ \ u(0) = u_0 \in \mathbb{R}^d, $$ where $F : \mathbb{R}^d \to L(\mathbb{R}^d, \mathbb{R}^d)$ is a sufficiently regular vector field, and the noise $\zeta$ is a distribution in $\mathscr{C}^{\alpha-1}(\mathbb{R})$ for $\alpha \in (\frac{1}{4}, \frac{1}{3})$. We construct solutions to this equation at lower regularity than previous work of Gubinelli-Imkeller-Perkowski and Prömel-Trabs using ideas from the high-order paracontrolled viewpoint of Bailleul-Bernicot. The key tool in our work is the development of a harmonic-analytic framework for proving a set of iterated commutator estimates which are necessary for the solution theory. We also show the procedure of enhancement when the noise $\zeta = dZ$, where $Z$ is a realization of fractional Brownian motion with Hurst parameter $H \in (\frac{1}{4}, \frac{1}{3})$.

Explanations of LLM outputs are considered an important way to make model behavior legible, helping human users make good decisions on downstream tasks. However, default training approaches for improving explanations target model task performance or human preference rather than human decision quality. To bridge this gap, we introduce HD-RL, a reinforcement learning framework for optimizing LLM explanations for human decision making. Our approach calibrates a human decision simulator on observed human decisions, applying a proxy reward model to optimize the LLM to generate explanations that help people to make good decisions. We apply this framework to AI-generated text detection. In our initial data collection phase, we observed that while LLMs possess strong autonomous detection capabilities, the explanations they generate for users yield near-random human accuracy (51.3%). By applying the HD-RL framework to the open-source Qwen 3.6 model, we generate optimized explanations that improve human performance, achieving a GLMM-estimated human accuracy of 63.7% in a randomized between-subjects experiment with repeated measures and outperforming RLVR, RLAIF, and other baselines by over 8%.

Large Language Models (LLMs) have recently improved Text-to-SQL generation, but their computational cost can hinder deployment in resource-constrained settings. This paper evaluates the Text-to-SQL capabilities of small and medium open-source LLMs (1B-8B parameters) under a controlled inference budget, focusing on how model specialization and prompt formulation affect performance. Using the Spider benchmark, we compare multiple prompt representations and assess the accuracy of generated queries. Our results show that SQL-specialized models are consistently more reliable than general-purpose and chat-oriented models at small scales, and that performance can vary substantially depending on the prompt used. To reduce this variability without resorting to larger models, we introduce three lightweight inference workflows that automatically retry with alternative prompts and/or models when failures occur, including an optional syntax-correction step. These workflows significantly improve end-to-end success rates, with a trade-off in additional inference calls. These results demonstrate a practical path to stronger Text-to-SQL performance with small LLMs, by trading modest additional inference cost for higher reliability.

Multiplicative background risk models (MBRMs) provide a tractable framework for describing dependent risks. We construct a class of MBRMs with Pareto margins by combining Marshall–Olkin exponential and additive gamma risk factor structures, each allowed to incorporate idiosyncratic and systemic background risk factors. Our main result establishes, under an admissibility condition on the risk factor systems, a finite sub-mixture representation of the joint survival function in terms of Pareto survival functions, whose weights are Dirichlet probabilities of orderings induced by the background risk factors and sum to at most one. This yields a new interpretation of such MBRMs, in which dependence is expressed through Pareto survival functions and their Dirichlet weights. The resulting family encompasses many known multivariate Pareto constructions, is closed under marginalization, and admits characterizations of independence and, under a non-degeneracy condition, of its singular and absolutely continuous parts. It is associated, its covariance decomposition separates the contributions of the two risk factor structures and their interaction, and its joint tails are regularly varying along rays. The presence of asymptotic tail dependence and the decay rates of joint exceedance probabilities are governed by the gamma risk factors alone; the exponential risk factors affect only the tail constants.

We prove that, for any two integers $2\le m<n$, a bounded operator $A$ on a complex Hilbert space satisfying \[ A^{*k}A^k=(A^*A)^k, \qquad k\in\{m,n\}, \] is quasinormal. More generally, we show that the inequalities \[ A^{*m}A^m\ge(A^*A)^m \quad \text{and} \quad A^{*n}A^n\le(A^*A)^n \] already imply quasinormality. Together with the known single-exponent counterexamples, this completely resolves the reduced Embry problem for bounded operators: the identities $A^{*k}A^k=(A^*A)^k$, $k\in S\subseteq\{1,2,\ldots\}$, characterize quasinormality if and only if $S$ contains at least two distinct integers greater than one. The proof combines Hansen's inequality and the transformer inequality for weighted operator geometric means with an extremal argument for optimal power bounds.

Creating haptic experiences from real-world environments remains difficult: visual scans capture appearance but not material feel, while manual haptic authoring requires specialized expertise. We present TouchTwin, a human-AI approach that turns phone scans of tabletop scenes into editable haptic scenes. TouchTwin uses AI to propose material regions that users can experience and revise through language and spatial edits. A model trained on measured tool-surface interactions generates the haptic responses, allowing users to work with material descriptions rather than low-level signal parameters. In a study with N=12 participants and physical reference objects present, automatic drafts scored 4.0/7 for realism and 5.4/7 as useful starting points. After a median of 6.6 minutes of editing, realism increased to 5.6/7, while scoring the saved material layers against ground truth showed material-family accuracy improving from 52% to 92%. Participants also used the same workflow to author material feels without a physical reference.

Language-model agents sometimes act beyond their authorization and then misreport what they did. On CAPE-Bench, matched agent tasks in which one fact decides whether an action is authorized, frontier open-weight agents do both, and one behavior marks the lie: retroactive permission laundering, acting and then asking permission as if the action were pending, precedes a false report in 70% of 1081 fresh informed violations, against 8% otherwise (preregistered odds ratio 7.0); frontier agents on tau^2-bench instead fabricate the authorization before violating. A linear readout of the agent's residual stream forecasts the violation before the action (AUROC 0.94) and the false report before the report, and beats chain-of-thought monitors when the agent is pressured to hide its reasoning. Gating every proposed action on this forecast cuts violations over whole episodes in four model families, and sending repeated flags to an approver keeps authorized work (Gemma-4: violations 0.70 -> 0.09, completion 1.00; a model approver that escalates unclear cases keeps this gain with a person seeing 25% of asks), at about 10-8 of the agent's compute and 109 times less than a chain-of-thought monitor. On AgentDojo, an external prompt-injection benchmark, the same readout calibrated on 29 tasks, with the steering unchanged, cuts attack success on unseen tasks in two families (Gemma-4 11.7% -> 5.7%, gpt-oss 28.5% -> 17.5%) while keeping benign utility (Gemma-4 0.637 vs 0.637; gpt-oss 0.667 vs 0.681). Preventing the violation is also the most reliable way to prevent the lie (misreported violations 0.10 -> 0.01 pooled over three families, preregistered), whereas lie-specific fixes after a violation are weak. All tests are preregistered and hash-locked; code, data and the full record, including tests that did not pass, are at https://github.com/AryaFayyazi/Honesty.