Papers

Offline multi-objective optimization requires not only moving the objective vectors of candidate designs toward the Pareto front, but also distributing them effectively along it. Generative methods have recently emerged as a natural approach because they learn a distribution over feasible designs while allowing generation to be steered toward promising designs. Existing methods, however, largely retain classical sample-wise guidance strategies, leaving the distribution-level modeling capability of generative methods underused. We propose ParetoTransport, a training-free guidance method for pre-trained flow-matching models that explicitly specifies and refines a population-level distribution in objective space. ParetoTransport guides a flow-matching sampler to iteratively transport the empirical offline distribution toward the Pareto front, with Wasserstein matching to intermediate proxy distributions. This directly controls distributional displacement and mass allocation along the front. We establish a convergence result and demonstrate state-of-the-art performance on standard offline MOO benchmarks, extending recent evaluations beyond hypervolume to generational distance, inverted generational distance, and Wasserstein distance.

This study uses ethnographic immersion and observation as contextual inquiry to understand situated action at an Emergency Medical Dispatch critical care hub. The work is a response to the urgent need to recruit context-specific knowledge and participation into design that helps to narrow the AI Chasm - the gap between the promise of Artificial Intelligence (AI) systems and what they deliver for clinicians and their patients. The work of a pre-hospital critical care team's dispatch process is described and analysed to reveal both the structure of the workflow and the different cognitive demands it makes on staff tasked with dispatch decision-making. Elements of attention, communication and focus between the humans, as they carry out this work, are drawn out in order to understand the work-as-done and identify the many dependencies in the process. The motivation is to establish where AI support might be useful and to discover what challenges there could be in designing appropriate algorithmic assistance. We ask whether, where and how the design and implementation of an AI system might be considered. The ultimate objective is to improve the decision process itself to the benefit of clinicians and patients. The study identifies three key steps in the situated workflow and details how decision-makers negotiate each one as emergency calls follow complex routes between them. We find compelling evidence that the first of these decision steps constitutes the most promising candidate for unobtrusive assistance that could be safe and effective in improving both clinician workload and clinical outcomes.

Modern mission-critical coordination demands seamless communication across multiple domains. Traditionally, Voice Communication Systems (VCS) have relied on physically separated Red/Black architectures to ensure voice and data segregation. While these hardware-based methods provide strong security assurances and remain foundational in high-security contexts, they can introduce significant complexity and scalability challenges as mission parameters expand into highly dynamic, multi-domain integrations. As defence, emergency response, and critical infrastructure operations increasingly require interoperable, flexible, and cost-efficient communication environments, there is a growing need to understand whether software-based approaches can provide comparable assurance while supporting modern operational requirements. This SoK characterises a transition to software-based Multi-Domain Data Segregation (MDDS) by integrating systems security, networking, and cryptography. Its goal is to consolidate existing research, identify shared architectural patterns, and address the security challenges facing next-generation high-assurance software-based VCS architectures. By assessing software-defined and virtualized approaches, this SoK supports the development of scalable, high-assurance VCS architectures that facilitate secure real-time coordination across diverse operational domains. Specifically, this SoK examines the security implications of Software-Defined Networking, Network Slicing, Separation Kernels, and Cross-Domain Solutions while addressing challenges posed by quantum computing through Post-Quantum Cryptography (PQC). This SoK provides a comprehensive analysis of the shift from hardware isolation to software segregation, serving as a crucial foundation for researchers and industry stakeholders aiming to enhance secure and adaptable VCS infrastructures for mission-critical operations.

Curating Web corpora for regional language variants like European Portuguese (PT-PT) is heavily bottlenecked by dialectal overlap (mainly with PT-BR) and data processing scale. This paper presents an efficient pipeline to curate a production-ready PT-PT corpus from the Portuguese Web, spanning 411 TB of raw data from Arquivo.pt. We introduce a novel post-scraping block that removes boilerplate and line duplicates prior to filtering. This early-stage intervention increases final document yield by 19.04% by rescuing valid text that standard heuristic filters prematurely discard. Integrated with rigorous language identification, weighted fuzzy deduplication, and neural quality classification, our pipeline offers a scalable framework and a clean, representative corpus optimized for LLM pre-training.

Using Knowledge Graphs (KGs) to describe Cultural Heritage Objects (CHOs) supports semantic richness and interoperability. However, standard KGs fail to capture the context-dependent validity of statements. This limitation is critical for cultural heritage metadata, which must often accommodate evolving or conflicting viewpoints, such as colonial versus post-colonial perspectives or shifting scientific consensus. While current knowledge representation methods address basic contextualization via provenance, qualifiers or reification, they lack a unified framework to simultaneously model and query data across multiple social, cultural, and political dimensions. To bridge this gap, we introduce the conceptual foundations of Multi-dimensional Knowledge Graphs (MKGs) and discuss how they preserve complex, multi-layered contexts in CHO metadata.

We demonstrate X-DigCheck, a domain-independent environment for building and maintaining application profiles as they co-evolve with the data they describe. Profiles developed against a fixed ontology quickly drift from the schema they were meant to capture. X-DigCheck treats profile construction as a continuous ontology-data co-evolution loop: data are lifted into RDF against the profile, checked through competency questions and SHACL, and the resulting reports jointly drive revisions of the ontology, mappings, constraints, and graph. The loop is agnostic to the domain and to the pipeline that produces the graph. We validate and demonstrate the tool in the cultural heritage domain, on the construction of RupeMagna-RTI, the first Reflectance Transformation Imaging (RTI) specialisation of the Cultural Heritage Survey ODP (CHS-ODP), aligned with CIDOC-CRM/CRMdig, ArCo, CHAD-KG, and Getty AAT, with semRTI as the lifting pipeline of this use case. The demonstration lets visitors run one full turn of the loop -on the shipped Rupe Magna (Grosio, Italy) RTI survey, or on a profile and graph of their own -executing the competency-question and SHACL checks live and reading the bidirectional coverage report that flags modelling gaps and stale assumptions. The result is a portable co-evolution environment for profile engineering, together with a reusable RTI application profile produced through it. A screencast of the demonstration is available at https://zenodo.org/records/22210609.

Hilbert-space embeddings enable inference for populations of persistence diagrams, but separation between individual diagrams need not survive population averaging. We develop a framework for inference on population mean embeddings, with particular attention to the additive landmark representations PLACE and PALACE. Treating each diagram as one independent observation, we apply Hilbert-space limit theory to obtain covariance estimators, two-sample tests, and confidence balls under suitable moment conditions, without requiring a lower-distortion bound. For additive embeddings, we identify the population mean as an embedding of the mean counting measure and show that geometric separation of these measures alone cannot guarantee uniform testing power. We then introduce a model with latent template diagrams, missing features, and location perturbations. Under common or feature-specific prevalence conditions, a diagram-level lower-distortion certificate yields explicit lower bounds on population mean separation. These margins provide finite-sample uniform power guarantees, and an additional information-divergence comparison gives matching sample-complexity bounds over restricted scale ranges. Confidence sets yield lower bounds on transport separation of population mean measures and exclusion guarantees for specified structured alternatives. We also quantify how orthogonal truncation changes the certified signal and the approximation allowance needed for confidence sets targeting the full embedding, relating sample size, retained coordinates, and template separation. Simulations examine calibration, power, and coverage, and an analysis of resting-state connectivity from the Autism Brain Imaging Data Exchange illustrates the procedures.

We study online control of a scalar storage system with nonnegative adversarial resource arrivals and state-dependent action constraints. The adversarial, time-varying cost depends on both state and action. At each time, the controller selects a feasible action before the current resource arrival and cost are revealed. To handle the coupling between feasibility and online learning, we introduce Simplex Disturbance-Action Control (SDAC), whose policies have $H$ nonnegative parameters summing to at most one. Every fixed SDAC policy is feasible by construction. Our online SDAC controller updates these parameters using entropic online mirror descent while preserving feasibility along the time-varying trajectory. We prove a regret bound of $O\!\left(\sqrt{T\log(H+1)}\right)$ relative to the best fixed SDAC policy and a minimax lower bound showing that the dependence on $T$ is tight. We also introduce infinite-memory SDAC policies, which include every feasible fixed-fraction policy. SDAC policies approximate this class with an error that decreases geometrically with $H$. For a suitable choice $H=\Theta(\log T)$, the resulting regret against infinite-memory SDAC policies is $O\!\left(\sqrt{T\log\log T}\right)$. The framework applies to energy-harvesting batteries and other storage-constrained resource systems.

In electric delivery fleets, mid-shift charging is non-trivial: each vehicle must decide when, where and how much to charge to finish on time with battery above a safety floor. The choices are coupled: queues build where too many vehicles pick the same station. Prior work resolves this coupling with central dispatching, precomputed schedules or reservations, machinery that charging infrastructure rarely supports. Instead, we use a family of learning agents under purely local control: every vehicle runs the same policy, deciding alone from its time budgets and broadcast station occupancies, leading to emergent coordination without central control or messaging. We validate this paradigm in simulation on real OpenStreetMap networks of twenty cities, each with a frozen scenario calibrated by an omniscient Oracle (99.5% of shifts completed on time), whereas a naive greedy rule (nearest station on low battery) completes just 73%. Agents trained with neuroevolution (NEAT) and policy gradients (PPO) on four cities and deployed zero-shot across all twenty, sixteen never seen in training, complete 96.8% and 98.6% of shifts, with the policy-gradient controllers proving more robust when demand or vehicle characteristics drift beyond the trained regime. In contrast, tuned threshold heuristics that read vehicle urgency alone fall short in contended cities (~80%). Through training, these learning agents rediscover partial charging and short opportunistic sessions, and route around busy stations, cutting per-session queue waits from about 45 minutes to under 2. In summary, this coordination paradigm balances local urgency against public occupancy, reaching near-Oracle performance at minimal implementation cost.

Live accompaniment models generate music for an incoming audio stream, committing to each output frame before hearing what comes next. In this strictly causal setting the model must infer tempo, meter, and metrical phase from its own imperfect past, whereby compounding errors quickly become audible as rhythmic drift. Put simply, the model has ears but no temporal reference, so when the ears hear imperfect, ambiguous music, the model will produce a flawed output. We propose Silent Metronome (SiMe), which gives it the temporal reference, encoding the phase within the beat and within the bar as periodic functions, pairing them with tempo and time signature, and supplying the result as a separate conditioning channel. Because this reference is independent of the generated audio, it cannot drift. Complementary auxiliary heads shape the latent representation, including a novel head that predicts the model's own future tokens. With the metrical signal taken from ground-truth annotations, beat alignment improves by a factor of 3.2 over the strictly causal baseline and surpasses a non-causal reference granted a full second of look-ahead. Coherence between input and accompaniment stays within a single point of that reference. These results suggest that streaming accompaniment systems should treat rhythm as a signal to be shared, as human ensembles do, rather than inferred.

Should multilingual LLMs answer medical questions consistently across input languages, or adapt responses to cultural cues? Existing multilingual medical benchmarks usually assume that medically correct answers should remain consistent across languages and treat cross-lingual variation as model error. In contrast, cultural adaptation research argues that appropriate medical answers may legitimately differ across contexts. We review the multilingual medical NLP literature through these two perspectives, we identify three gaps: limited stakeholder perspectives (e.g., of medical professionals), a lack of empirical evidence on which approach better serves users, and no benchmarks capable of distinguishing universally correct from culture-specific cases. To address the first gap, we survey 356 participants across three stakeholder groups (medical, NLP, and anthropology professionals) in three countries (Germany, Spain, and the United States). Anthropologists consistently favor adaptation, while medical and NLP respondents remain divided, with notable divergence between U.S. and European medical professionals. LLMs prompted with profession and country personas fail to reproduce this variation, overestimating cross-lingual consistency preference among NLP and medical personas. We conclude that neither consistency nor adaptation can currently be considered clearly preferable, highlighting the need for empirical evidence on which approach better serves users across cultural contexts.

In real-world applications, pedestrian trajectory prediction models rely on inputs from detection and tracking systems. Prior trajectory prediction benchmarks either contain relatively sparse pedestrian interactions, assume perfect tracking inputs, or rely on overhead viewpoints that minimize occlusion and perspective distortion, limiting evaluation in realistic dense-crowd scenarios. We present CrowdTraj, a benchmark for pedestrian trajectory prediction in natural dense crowd scenes. Unlike previous datasets, CrowdTraj supports end-to-end evaluation from detection through tracking to trajectory prediction under severe occlusion in CCTV views. It also captures diverse, natural pedestrian behaviours, including abrupt directional changes rarely observed in existing benchmarks. CrowdTraj includes five diverse scenes, with an average of 1,146 unique pedestrians per scene, maximum frame-level densities ranging from 114 to 372 pedestrians, and over 3.2 million annotated head bounding boxes. CrowdTraj provides pixel and real-world coordinates via per-scene homography matrices for physically meaningful analysis. Our experimental results show that tracking accuracy (IDF1) drops to 0.68 to 0.70 in the densest scenes, compared with approximately 0.90 in less crowded scenes. Trajectory prediction training also becomes substantially more computationally expensive in dense scenes, with training times increasing by up to 8 times. These findings show that CrowdTraj exposes limitations in current trajectory prediction pipelines that remain hidden on existing sparse-crowd benchmarks, particularly in robustness to tracking noise and computational scalability.

Associative Recall (AR) is the cognitive ability to learn and retrieve links between items in memory. In NLP, AR is used as a benchmark for evaluating the in-context memory capacity of architectures such as Mamba, and has been found to strongly correlate with language modeling performance. This paper explores AR from the perspective of mechanistic interpretability, aiming to reverse-engineer the exact internal algorithm used by Mamba to perform recall. Our key insight is that Mamba performs recall by implicitly learning linear hash functions, and we identify the low-level circuit that enables this behavior. Building on these findings and inspired by theoretical tools in similarity-preserving hashing, such as the Johnson-Lindenstrauss lemma, we develop a theoretical framework for analyzing AR, which we term Recall Scaling Laws. Given the vocabulary size and the number of facts in context, this framework allows us to (1) predict the embedding and state dimensions required for Mamba to achieve perfect recall, (2) predict recall success probability given the model dimensions, and (3) analyze multi-layer models and multi-head SSM patterns. Empirical results show that our theoretical findings are accurate and predictive, offering insights into how AR capacity scales with vocabulary, state, embedding size, and architecture.

Multi-agent LLM pipelines increasingly span organisational boundaries; when a fault surfaces, someone must determine where it entered. The artifact available is rarely a full execution trace: it is the reports each agent filed, and a filed report can state a conclusion alongside its observations. Using a pre-registered, institutionally partitioned pipeline of six agents with process-level information boundaries, balanced defect injection and matched clean twins (345,600 requests per chain model, two models), we first report that our pre-registered hypothesis – that collective responsibility framing degrades escalation with chain length – is not supported. The layer nevertheless fails asymmetrically. It originates almost nothing: zero allegations across 7,996 clean episodes where every agent stayed silent. It filters upstream error poorly, naming an innocent party in 34.4% and 62.6% of clean episodes where an agent raised a false alarm. Conditional on no agent proposing the true origin (59.5% of episodes on one chain model), an auditor reading the reports recovers it in 4.1% of cases – below a uniform guess (20%) and the best fixed-link accuser (31.0%) – while reaching 60.3% from the raw documentation of the same episodes. Deleting one clause, the field carrying the agents' own conclusion, isolates the cause at constant observations: accuracy rises to 45.2% (+41.2 pp, 95% CI +35.3 to +46.9) and adherence collapses from 94.4% to 3.4%; where the suggestion was correct the same deletion instead costs accuracy, 70.5% to 55.7%. The harm replicates on two frontier auditors in four conditions out of four (+8.5 to +39.0 pp) and in a second domain (+47.7 and +61.1 pp), where the cost disappears. The net effect is governed by upstream reliability together with both conditional magnitudes. An accountability layer needs evidence sufficiently independent of the conclusions it verifies.

Extrusion-based 3D concrete printing builds structural elements layer upon layer without formwork, so freshly deposited material must carry the growing weight of the structure while it is still gaining strength and stiffness. The largest number of layers that can be printed before failure, the buildability, is therefore central to process design. Its prediction rests above all on how accurately the constitutive model represents fresh concrete. Existing finite element frameworks pair gradual pre-failure hardening only with smooth yield surfaces, which leaves the Mohr-Coulomb surface that best describes fresh concrete restricted to perfect plasticity. This work combines a smoothed Mohr-Coulomb yield surface with nonlinear isotropic hardening of the cohesion, formulated in an updated Lagrangian setting with an activation field for layer deposition. Failure is identified from a tangent-stiffness eigenvalue criterion that matches the analytical self-weight buckling load of a slender wall and is insensitive to geometric imperfection. The framework reproduces the measured collapse of a printed straight wall within one layer and predicts the collapse of a hollow cylinder closer to experiment than existing frameworks. A study over six diameters and six hardening rates shows buildability peaking at intermediate diameters and rising with the hardening rate. Recording how much of its strength the material has used at failure, slender cylinders fail with much still in reserve while intermediate ones use nearly all of it. The two failure modes are therefore identified as the limits of one process rather than separate failure mechanisms.

Transaction-local controls answer whether one financial request may proceed, but market behavior can be distributed across messages, agents, assets, and time. We study this interpretation gap in a virtual exchange populated by ten role-conditioned language-model agents. The agents communicate, trade reference assets and futures, launch tokens, and manage concentrated-liquidity pools under prescriptive adversarial roles. We analyze eight 72-cycle trajectories across two time-blinded hourly replay paths, with a runner-side wallet policy enabled or disabled. The retained artifacts connect generated outgoing messages, policy events, balances, positions, and cycle-end market state. A focal reconstruction shows a launch–promotion–exit scenario realized across private coordination, public claims, follower positioning, repeatedly withheld exits, and a later non-blocking request aligned with a token balance change. Across policy-enabled runs, the gate withholds direct requests selectively; most policy-categorized candidates are flagged rather than blocked, while the surrounding interaction can continue. Repeated runs also show that category-level and within-trajectory relations can recur even when normalized score-change rankings do not. These findings motivate agent-behavior evaluation that links communication, authorization, and evolving state instead of treating individual transaction verdicts as complete safety judgments.

Citation-based measures of scientific influence typically treat citations as uniform signals, ignoring the different roles that cited works play in a paper's contribution. We introduce contribution-based credit scoring for research articles: a structured citation analysis that decomposes a paper's credit between its own original contribution and the prior work it builds on. Motivated by a cooperative-game view of scientific credit, we propose the contribution tree, a hierarchical framework that conserves importance across the document structure and separates original from citation-derived contribution. To make this framework scalable, we use LLMs as noisy comparative estimators of local importance. We further extend the model to article collections by propagating contributions through weighted citation graphs, yielding corpus-level contributions and normalized influence scores. Our experiments suggest that our framework captures contribution signals beyond surface-level heuristics. Our code is available at https://github.com/sanaebrahimi/Importance_Scoring/

Production content-generation systems must integrate a user's immediate task, long-term brand identity, historical evidence, and revision feedback. We present Aegix Pulse, a production-oriented three-stage architecture that separates current-task clarification and Task Persona finalization, long-term Account Profile (Brand DNA) assembly, and controlled generation and revision while preserving provenance across content versions. We evaluate four preregistered claims using 96 synthetic social-media generation tasks. Four initial-generation conditions progressively introduced a Task Persona, Account Profile, and successful-history style evidence, while two revision conditions compared plain and context-preserving revision. The experiment produced 480 completed generation records and 1,440 blinded LLM-Judge evaluations, supplemented by human review. Adding the Account Profile increased mean brand-consistency scores by 0.1562 points on a five-point scale compared with Task Persona alone (Holm-adjusted p=.1224). Preserving task and brand context during revision increased mean task-preservation scores by 0.2917 points compared with plain revision (Holm-adjusted p=.2432). Neither improvement was statistically conclusive after multiple-comparison correction. Task Persona alone showed a small observed effect, while successful-history evidence provided no additional improvement in brand consistency under the current setting. Human validation did not consistently reproduce the LLM-Judge effect directions and showed low inter-reviewer agreement. These findings provide preliminary evidence for persistent brand context and context-preserving revision while identifying priorities for stronger evidence processing and evaluation.

The growing realism and accessibility of manipulated and generated faces threaten the trustworthiness of digital media. To detect such forgeries, deepfake detectors based on vision foundation models have shown promising performance, but they typically rely on a single pretrained representation and are prone to overfitting to particular training distributions. To improve generalization to unseen forgeries, we propose UCF-Net, an uncertainty-aware cascaded fusion network that harnesses CLIP's language-aligned semantic priors and DINO's self-supervised visual-structure priors. UCF-Net extracts hierarchical features across Transformer depths, uses layer-wise expert aggregation to adaptively combine each encoder's multi-level cues, and performs weighted fusion of the resulting representations based on entropy-derived uncertainty. We further consolidate public deepfake datasets into a unified benchmark of approximately 4M images and construct a separate cross-generator evaluation set with over 8K face images from eight recent generators. On the unified benchmark, UCF-Net achieves the best mean AUC among the evaluated methods in both in-domain and cross-domain evaluations. On the cross-generator set, it adapts effectively with limited target-domain data, although zero-shot transfer remains challenging.

We present a concrete implementation in Haskell of a monadic framework that includes both a small-step interpreter and a type-and-effect checker for the corresponding language. Our approach separates the language syntax from the semantics of its effects. This design allows the interpreter to remain parametric over the underlying monad, while the static checker approximates effects independently of their concrete implementation. The theoretical foundation of this framework-applied to a call-by-value lambda calculus with generic effects represented by operations that produce monadic values and are managed through handlers-was introduced in previous work, where the focus was on proving the soundness of the approach. In contrast, the present work leverages Haskell's support for modular programming and monads to demonstrate that the framework is practically implementable and usable. We illustrate the approach with examples using the monad of exceptions and the one of nondeterminism and expressions both with and without handlers.

Full-parameter fine-tuning of large language models has substantial memory costs because backpropagation stores activations and gradients. Zeroth-order optimization avoids this by estimating update directions from loss evaluations, but existing methods require tuning a sensitive learning rate for each model and task. We propose the momentum $p$-dimensional subspace trust-region method (MpSub). At each iteration, MpSub searches within a $p$-dimensional subspace: one direction preserves historical momentum from the most recent accepted step, while the remaining directions explore via fresh random sampling. The subspace gradient is estimated by central differences, a trial step is computed from a linear trust-region model, and the trust-region radius adapts according to the agreement between predicted and observed loss reduction, eliminating the learning rate. For LLM fine-tuning, evaluations within an iteration share a minibatch, and directions are regenerated in place from seeds, using forward passes alone. For smooth deterministic objectives under unorthogonalized Gaussian directions, we bound the finite-difference error, quantify gradient energy captured by the subspace, and prove that $\lim_{k\to\infty} \|\nabla f(x_k)\|_2 = 0$ almost surely under a safeguarded radius update. Under a matched budget of 8,400 training-objective forward passes, we fine-tune OPT-125M and OPT-350M on CommitmentBank. With the same preset parameters at both model sizes, MpSub attains mean test accuracies of 0.673 and 0.690 over three seeds, matching tuned MeZO (0.685) without any learning-rate search.

Deployment of Large Language Models (LLMs) on memory-constrained edge devices relies heavily on aggressive post-training quantization. However, evaluating these models is largely based on zero-shot task accuracy, which depends solely on argmax predictions and is insensitive to changes in the underlying predictive distribution. Consequently, accuracy can exhibit unstable, non-monotonic behavior under progressive quantization, masking substantial fidelity loss relative to the BFloat16 (BF16) uncompressed base model and providing misleading deployment signals. We introduce a distribution-sensitive evaluation framework quantifying information loss in quantized LLMs as the divergence between full-vocabulary predictive distributions at the token decision boundary. We compute statistical distances, including Jensen-Shannon Divergence and Total Variation Distance, between outputs of full-precision and quantized models, enabling a fine-grained analysis of distributional shift. Using this framework, we quantify probability mass displacement and distributional drift relative to the BF16 reference, capturing predictive distribution changes not reflected in top-1 accuracy. We conduct a 120-run experimental matrix across five foundation architectures and four reasoning benchmarks under progressive quantization regimes, from uncompressed BF16 to Q2_K, providing a systematic fidelity analysis. Our results show divergence metrics generally increase under stronger quantization, complementing task accuracy with a fidelity signal. Across tested llama-cpp schemes, mixed-precision Q4_K generally yields lower divergence than uniform Q4_0 at similar memory footprints. These findings motivate distribution-aware evaluation as a practical diagnostic complement to task accuracy; they do not directly establish correctness, calibration, safety, or user-perceived quality.

A wrong number is worse than no answer. Across factuality-critical domains – audience metrics, scheduling and rights in media; dosages and lab values in healthcare; figures and citations in finance and legal – a confident but fabricated value is more damaging than an honest admission of uncertainty. Yet this is the dominant failure mode we observe on small local language models: even when correct evidence is present in context, models fabricate plausible numbers and timestamps. Recent work characterizes a real limit of this regime: below 7B parameters, the bottleneck of retrieval-augmented generation (RAG) is not retrieval quality but context utilization. We present Noesis, the deterministic-first query plane of the Noesis architecture, which makes every deterministic judgment before generation. Its mechanisms follow from the ingestion architecture (subject of a separate patent application): (a) a producer-side fact layer rendering precomputed metric facts verbatim without ranking; (b) positional addressing with deterministic cross-source alignment, resolved ahead of query time at zero LLM cost; (c) provenance scoping as an attribution constraint with multi-tier named-reference routing; and (d) two-tier context with model-triggered verbatim hydration. Across four ablations, a 2B model reaches parity with a 35B model on factual integrity (exact values in all runs; zero confabulated numbers on absent-entity traps); structured retrieval beats flat RAG by +11.4 points at 2B; skeleton-only context preserves quantitative answers at 20-30% smaller prompts; and hydration recovers verbatim narrative in ~8s versus ~29s. Two properties matter for regulated domains: each query resolves in a single generation call, and every reported value is traceable to its exact source and position by construction.

Large language models are increasingly used as sources of advice and information, including in high-stakes settings, yet little is known about how they respond to user disagreement. We study how a model manages its epistemic authority, referring here to its claim to knowledge, competence, or the right to advise, once a user challenges its answer. Building on Conversation Analysis, we introduce a taxonomy of six challenge types and a four-layer framework for analysing each response: whether the original claim is maintained or changed, where authority is located, how the disagreement is socially managed, and what kind of evidential support is offered. We construct a new dataset of 2,310 controlled challenge scenarios and 32,340 corresponding responses from 14 models, and analyse them using our framework with an LLM-as-judge pipeline, providing a vocabulary which future evaluation and benchmark design can build on. We find that models show conflicting behaviour: they validate users in 85% of responses but maintain their original claim in 65%. They explicitly apologise in 33% of responses, yet 59% of those apologies accompany maintenance of the original claim. They transfer authority most often in advice tasks, doing so in 28% of responses and reaching 57% in health advice and 49% in legal advice, compared with 6% in fact and 3% in explanation tasks. Abandonment of the original claim ranges from 0.8% for GPT-5.2 to 40% for DeepSeek 7B, while complete replacement of the original claim is rare overall at 1.5%.

The concept of Markov chain Monte Carlo (MCMC) cycles, an analogy to cyclic processes in heat engines, is presented in order to examine Bayesian inference problems. In this effort, we develop adaptive ensemble schedulers that allow the tuning of external parameters of a Bayesian canonical ensemble during an MCMC run, realising the MCMC cycles in practice. We run these cycles on different statistical models. As a fundamental insight, we find (both theoretically and in practice) that such systems can produce a non-zero net work output if and only if the considered model is non-Gaussian. As such, they may serve as a measure of non-Gaussianity in Bayesian inference, which we test on an example from supernova cosmology.

The paper is concerned with the numerical approximation of some two-dimensional asymptotic models for the propagation of internal waves in a two-layer system in the form of pde's for the interfacial deviation and the velocity. After the study of well-posedness of the periodic initial-value problem, the convergence of a semidiscrete approximation based on a spectral Fourier-Galerkin method is analyzed. The resulting semidiscrete system is then integrated numerically in time with the implicit midpoint rule. Some properties of the fully discrete scheme are described and its performance is illustrated with some numerical experiments.

Closed-loop AI scientists can generate candidate designs at low marginal computational cost, whereas reliable feedback may require wet-lab synthesis, characterization, or high-fidelity computation. Addressing this imbalance through custom laboratory automation remains infrastructure-intensive and costly, while replacing new experiments with a fixed surrogate leaves persistent model errors that can be amplified by optimization. We propose online surrogate repair (OSR), a closed-loop algorithm that uses sparse high-fidelity evaluations to update the surrogate throughout a longer agent search conducted primarily with inexpensive surrogate feedback. An acquisition rule selects which designs from the agent's accumulated proposals receive high-fidelity evaluation, and the resulting labels update the surrogate used in subsequent episodes. Across controlled synthetic environments, we demonstrate that improving global surrogate fit does not necessarily reduce maximum regret, whereas Q90-UCB and expected improvement (EI) substantially reduce regret by directing evaluations toward regions that determine the optimizer's decisions. \osradded{Across LLM and Chemeleon workflows, Online EI reduces the amount of oracle queries required by 84.3–90.3% for matching discovery counts and by 77.9–81.6% for matching the quality of the best material found.} Online surrogate repair introduces a third feedback regime between fixed-surrogate operation and high-fidelity feedback after every episode, separating the frequency of high-fidelity evaluation from the duration of the agent's search.

GSMA Remote SIM Provisioning (RSP) enables over-the-air delivery of eSIM profiles, but it exposes long-lived identifiers during profile ordering and download. In particular, stable device identifiers (e.g., EID), profile identifiers, and long-lived certificate material enable mobile operators and profile-delivery infrastructure to link provisioning events to the same eUICC and, when combined with account records, to the same subscriber. This undermines subscriber anonymity and enables cross-session tracking. We present ZK-eSIM, a privacy-preserving redesign that achieves subscriber anonymity and provisioning-session unlinkability while retaining accountable traceability by exception. ZK-eSIM (i) replaces direct disclosure of device identifiers with a zero-knowledge proof of device validity and eligibility; (ii) enforces session unlinkability through short-lived, one-time pseudonymous credentials and per-session identifiers to prevent cross-session tracking; and (iii) provides privacy-preserving accountable traceability through a jointly authorised escrow mechanism, so that no single entity can unilaterally deanonymise a user. We formalise a multi-entity, honest-but-curious threat model and prove subscriber anonymity and the unlinkability of provisioning sessions under standard cryptographic assumptions. We implement a Java Card applet on a test eUICC to evaluate performance on commodity hardware with a modified LPA and SM-DP+ server. Our experiments quantify end-to-end cryptographic overhead relative to conventional RSP, confirming that ZK-eSIM adds only practical overhead, closing a critical privacy gap while preserving deployability within existing GSMA roles and interfaces.

As software engineering research submission counts surge, peer reviewers face severe time constraints, making systematic verification of citation-supported claims prohibitively expensive. Consequently, unsubstantiated claims and semantic drift can propagate undetected across scientific literature. Existing approaches such as fact-checking and retrieval-augmented generation tools operate on open-domain web data or evaluate claims in isolation without processing complete manuscripts. To address this gap, we present RefVerifier, a semi-automated, citation-bounded reference verification prototype designed to support in academic peer review. RefVerifier extracts citation-bearing sentences from manuscripts, checks bibliography metadata against scholarly databases, resolves references to full-text open-access PDFs, localizes relevant evidence passages, and generates verdicts with natural language explanations. Evaluating RefVerifier on public benchmarks shows claim detection at an F1 score of 0.990, open-access resolution of 57.6% of references, and evidence localization with a hit rate of 98% on abstracts and 68% on complete cited papers. In an end-to-end test with eight manuscripts, RefVerifier achieves a verdict accuracy of 71%. By automating document retrieval and evidence localization while preserving reviewer oversight, RefVerifier provides first indicators for the feasibility of semi-automated integrity checks in scholarly publishing.

This paper presents a small-scale quantitative experiment that links syntactic structure to stylistic functions in narrative prose. Starting from a dependency-parsed corpus of 3,300 sentences, we derive sentence-level stylistic labels across five categories — descriptive, introspective, causal, ideological, and neutral — using a transparent rule-based procedure that inspects lemmas, universal part-of-speech tags, and syntactic relations. For each sentence we construct a compact representation of its syntactic profile as a sequence of linearised triples combining lemma, POS tag, and dependency relation. These patterns serve as input to standard machine-learning classifiers trained to predict sentence-level style. The best-performing model achieves a macro-F1 of 0.948 under 10-fold cross-validation. The experiment is implemented entirely in Python using open-source tools. Our goal is not to propose a fully fledged stylistic theory, but to offer a reproducible and extensible workflow for exploring how grammatical structure contributes to narrative interpretation.