LLM Evaluation

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

194 papers in the last four weeks, up 87% on the four weeks before. 1.9% of all new papers.

Jul 13Week of Sep 28

Latest papers 1,643

Aug 2, 2026cs.AI

Judging Is Not Enumerating: Silent Omissions in LLM-Authored Acceptable Sets

Language models are increasingly promoted from examinees to examiners: they write the test suites, answer keys, rubrics, and reward functions that define correctness for other systems. We measure the capability that role assumes and find it lacking under the protocol the role is usually deployed with, one-shot greedy authoring with no test-time reasoning. Across four reference constructions - two with complete finite truth, one with a hardened executable reference (HumanEval+/MBPP+), one with an explicitly incomplete lexical reference (WordNet) - models judge whether a candidate belongs far better than they author the set itself. On the incompleteness-proof algorithmic construction the gap is +0.34 to +0.29 F1 over a 24x parameter range and does not close; on executable code, models judging at F1 0.74-0.90 author suites admitting only 19-42% of oracle-correct solutions. A control locates the deficit: asked to emit the predicate rather than its extension, the same models reach F1 about 0.99. The failure is not missing knowledge or an inability to specify, but an inability to materialise the region a specification induces. The dominant error is omission, which resists audit: an over-inclusion is a token a reviewer can challenge, a missing member an absence whose discovery is the authoring problem itself. Models detect planted over-inclusions 6-7x more often than planted omissions, and a production deployment of 43,227 items fails omission-first at 10:1. Wired into RLVR, an authored key costs 1.9 points of accuracy against an exact oracle and 18.5 WordNet-relative (six paired seeds, p=0.031). Gating authored verifiers on a known-correct probe cuts false rejection from 58-92% to at most 5%, but keeps only 5-39% of suites. Repairing them instead, by rewriting each wrong expected value to what a reference execution returns, raises yield 3.3-10.6x across four author families.

Wenhui Chen, Jianlin Chen, Ziyao Lin +2

University of Macau · South China University of Technology · Independent Researcher

LLM Evaluation LLM-as-a-Judge LLM Auditing Automated Test Generation

Aug 2, 2026cs.AI

Passing Coarse Marginal Checks Can Be Cheap: Persona Mixtures and Imprecise Treatment-Response Estimates in an LLM Persona Panel

Large language models are increasingly used as synthetic research participants and are often validated by whether their marginal responses resemble human data. We study a fixed panel of sixteen lightweight persona-conditioned GPT-4.1 configurations in repeated strategic games. The panel met preregistered broad-reference condition-mean criteria in three of four repeated-game cells; the sole miss was 0.011 below the lower reference bound. Variation was strongly prompt-indexed, but its share depended on uncertainty assumptions: fixed-panel symmetric-Dirichlet sensitivities produced median between-prompt shares of 63%-71% under Jeffreys alpha=0.5 and 47%-53% under alpha=1, while finite-opportunity plug-in estimates were 85%-96%. Aggregate continuation-probability contrasts were +0.083 and +0.078, with conservative simultaneous 95% intervals [-0.171, +0.330] and [-0.181, +0.330]. The treatment jointly changed the continuation process and its textual representation. A separate wording-and-position operation shifted cooperation from 0/40 to 37/40 in the bare configuration, and a label conflict also revealed representation control. The original persona-level p13 result was not prospectively family-controlled, while a post-adjudication exact gate was structurally underpowered; p13 is therefore a replication target rather than a finding. External review exposed family-error, dependence, construct, and boundary-uncertainty defects, and zero-call reanalysis changed the interpretation without rewriting the historical record. The registered marginal criteria could be passed without precisely estimating the treatment-response object. A public capsule verifies 4,916 confirmatory Phase 3-5 runs with no live model calls. The results concern one fixed model-prompt panel and do not establish human substitutability.

Yohei Nakajima

Untapped Capital

Prompt Sensitivity LLM Evaluation Causal Inference

Aug 1, 2026cs.AI

When Does LLM Orchestration Pay Off? A Controlled Evaluation of Accuracy, Cost, and Task Difficulty

LLM orchestration is often assumed to improve reasoning by allocating additional inference-time computation, yet its gains may not justify its cost. Existing comparisons also frequently overlook differences in optimization effort, making it difficult to isolate the value of orchestration itself. We conduct a controlled evaluation of Self-Refine, Best-of-

N

, and Debate against task-only and chain-of-thought (CoT) single-call baselines across five LLM backbones and three domains: competitive programming, chess puzzles, and mathematics. For comparability, we optimize each method with GEPA under the same optimization budget and evaluate all methods on the same difficulty-stratified benchmark items. Orchestration yields moderate but benchmark-dependent gains: averaged across backbones within each benchmark, the largest improvement is 4.6 percentage points over optimized CoT inference and 4.5 points over task-only inference, while requiring approximately 2 to 4 times the mean total tokens of task-only inference. Human-derived difficulty is associated with lower absolute accuracy in all three benchmarks, but within-benchmark analyses do not indicate that orchestration effects increase with task difficulty. By contrast, exploratory mixed-effects analyses reveal strong interactions between orchestration method and backbone model across all three benchmarks, showing that orchestration effectiveness depends substantially on the underlying model. Our results suggest that orchestration decisions should be model-specific and account for whether moderate accuracy gains justify the additional inference cost. More broadly, evaluations of LLM orchestrations should control optimization effort and report model-specific accuracy--cost trade-offs rather than treating additional inference-time structure as uniformly beneficial.

Nicolas Leins, Nico Pelleriti, Jana Gonnermann-Müller +1

Zuse Institute Berlin & TU Berlin, Berlin, Germany · Zuse Institute Berlin & Weizenbaum Institute Berlin, Berlin, Germany

LLM Evaluation LLM Self-Refinement Inference-Time Optimization LLM Agent Orchestration Cost-Aware Inference

Aug 1, 2026cs.AI

Multi-Dimensional Assessment for AI Cognition (MAAC): A Theoretical Framework for Process-Oriented Cognitive Evaluation of Text-Based AI Systems

Evaluating artificial intelligence systems has historically relied on outcome-based benchmarks that measure task accuracy, robustness, or fairness. While indispensable, these benchmarks provide limited diagnostic insight into the underlying cognitive processes that generate performance-leaving critical questions unanswered about how AI systems reason, integrate memory, manage complexity, or avoid generating false information. This paper introduces the Multi-Dimensional Assessment for AI Cognition (MAAC), a theoretically grounded framework for shifting evaluation from what text-based AI systems produce to how they think. MAAC defines nine cognitively motivated dimensions: Cognitive Load, Tool Execution, Content Quality, Memory Integration, Complexity Handling, Hallucination Control, Knowledge Transfer, Processing Efficiency, and Process-Outcome Alignment. Each dimension is grounded in established cognitive science theory-drawing on Marr's tri-level hypothesis, Baddeley's working memory model, Sweller's cognitive load theory, and unified theories of cognition. Five theoretical analyses provide initial support for the framework's coherence and empirical testability: dimension-to-theory mapping; a coverage matrix assessing breadth and non-redundancy; a formal gap analysis relative to current evaluation practice; a worked diagnostic illustration; and a set of a priori interdependency predictions for future empirical testing. MAAC provides a theoretical and operational framework for principled process-level cognitive assessment of text-based AI systems, complementing existing outcome-based benchmarks with cognitively grounded, multi-dimensional evaluation.

Abdalla Doleh, Ratna Babu Chinnam

Department of Industrial & Systems Engineering, Wayne State University, Detroit, MI 48202, USA

Aug 1, 2026cs.AI

TrAC: Trace-Conditioned Answer Consistency for Efficient Uncertainty Quantification in LLMs

Large language models (LLMs) can generate fluent reasoning traces that nevertheless lead to incorrect answers, making response-level uncertainty estimation important for abstention, human review, and adaptive compute allocation. Existing approaches generally fall into three categories: passive single-trace methods use token-level confidence signals, sampling-based methods compare multiple complete traces at higher generation cost, and active prefix-based methods probe partial traces to study answer stabilization or preference transitions. However, none actively re-elicits an answer from a completed reasoning trace to measure its consistency with and support for the original answer. To address this gap, we introduce Trace-Conditioned Answer Consistency (TrAC), a correctness-supervised uncertainty quantification framework that combines active and passive signals anchored to one completed reasoning trace. Its active component, Prefix-Conditioned Elicitation (PCE), re-elicits a short answer conditioned on the completed trace and represents both its consistency with the original answer and its token-level probabilistic support. Its passive component, Trace Uncertainty Profile (TUP), summarizes how token-level uncertainty evolves throughout the original generation without additional decoding. A lightweight head then integrates the two representations into a response-correctness score. Across five mathematical reasoning benchmarks and three LLM families, TrAC improves macro AUROC by 1.8% and reduces AURC by 3.4% relative to eight-sample self-consistency, while using one complete reasoning trace and a short cached answer probe. When eight samples are already available, augmenting sample consensus with re-elicitation further improves macro AUROC by 4.3% and reduces AURC by 8.3%, without additional full-trace generation.

Dahai Yu, Lin Jiang, Rongchao Xu +1

Florida State University Tallahassee, Florida, USA

LLM Evaluation Reasoning Consistency in Language Models LLM Uncertainty Estimation Reasoning Trace Analysis

Jul 31, 2026cs.CL

CurveShift: Is Agent Progress Scalar? Separating Level from Shape

Progress in large language models is often summarized using a single scalar measure, such as a time horizon, a latent ability estimate, or an aggregate benchmark score. These summaries capture the overall performance, but they do not test whether progress is distributed differently across task difficulty. We find that most of the apparent shift in gains toward harder tasks does not reflect a change in the shape of the difficulty-response curve. On METR time-horizon data, a single Rasch model with rising ability reproduces this pattern, so it is largely explained by ceiling effects rather than a qualitative change in capability. This echoes how the choice of metric can make claimed emergent abilities look like a property of the models themselves. We then identify a smaller hard-task effect that survives this control. Isolating it is difficult on agentic benchmarks, because newer models are usually run with newer agentic harnesses, so a gain on hard tasks cannot be assigned to the model or its scaffold. We break the confound with LiveCodeBench, a public competitive programming benchmark that runs no agentic scaffold while pairing dated models with an exogenous difficulty ordering. After accounting for the rise in overall ability, models released after September 2024 still gain on the hardest problems beyond what their easy and medium performance predicts, by about +0.40 logits under our most conservative assumption, raising the hard-problem solve rate from roughly 18% to 25%. The effect is led by the strongest reasoning models and holds for hard tasks that need only short reasoning, not autonomy over long horizons. We present this as a result specific to competitive programming, since our clean identification rests on a single coding benchmark. We release the LiveCodeBench Difficulty Panel (66 dated models x 1,055 problems) and our analysis code.

Hanwen Xing, Pengyun Wang, BingXu Meng +8

University of Southern California · University of Chicago · University of California, Berkeley +6

LLM Evaluation Agent Evaluation

Jul 31, 2026cs.AI

More Debate, Same Evidence: Structural Limits of Homogeneous Multi-Agent Groundedness

Large language model (LLM) judges are increasingly organized as multi-agent panels under the assumption that exchanging critiques improves judgment quality. We test this assumption for \emph{groundedness verification}, where a judge must determine whether a claim is supported by the supplied evidence. We evaluate a homogeneous three-agent panel on six public fact-verification and hallucination-detection benchmarks. Relative to a fixed single-agent reference, the panel's system-level accuracy difference ranges from

+8.5

to

-4.4

percentage points: two datasets show reliable gains, one shows a reliable loss, and three are statistically inconclusive. Because the reference and panel use different model variants, these differences characterize the complete systems rather than isolate a causal debate effect.

Yuelyu Ji

LLM Evaluation LLM Grounding LLM-as-a-Judge Multi-Agent Debate

Jul 31, 2026cs.AI

ModelEquivBench: Certifying Multi-Relational Evaluation of LLM-Generated Optimization Models

Large language models increasingly generate optimization models from natural language, but existing evaluation often reduces a generated model and its ground truth to a single equivalent/not-equivalent verdict or an execution-success rate--labels that are neither independently checkable nor faithful to the multiple distinct senses in which two formulations can agree. We present ModelEquivBench, a certifying, multi-relational evaluation system that reports a per-pair semantic profile E0--E6: model construction and exact ingestion (E0), verified representation alignment (E1), same-space and projected feasible-set relations (E2, E3), objective-order equivalence (E4), optimal-value equality (E5), and optimizer-set equivalence (E6). Each decided entry carries relation-appropriate, independently re-checkable evidence: replayable traces or explicit maps for E0--E1, exact-rational certificates for positive E2--E6 conclusions, and explicit witnesses for supported negatives. Incomplete mapping search, unsupported structure, and resource limits produce typed UNKNOWN or N/A outcomes rather than guesses, while unmet prerequisites are reported as ABSENT. Using ModelEquivBench to evaluate three model snapshots--GPT-5.4, Claude Sonnet 4.6, and Qwen3.5-397B-A17B--on the same frozen cohort of 173 base problems (346 cells per model) under a no-repair protocol, the resulting profiles expose distinctions that coarse baselines do not represent: 49, 35, and 25 cells contain executable candidates that are nevertheless certified negative on at least one supported relation, and 25, 8, and 18 structural rejections occur on pairs for which E2 certifies mapped feasible-set equality under a verified map. The three model snapshots fail at different stages of the profile and therefore cannot be meaningfully reduced to a single accuracy score.

Penglin Zhu, Jungang Xu

School of Computer Science and Technology, University of the Chinese Academy of Sciences

LLM Evaluation Pairwise Comparison Language Model Generation Evaluation Benchmark Design Natural Language to Optimization Modeling

Jul 31, 2026cs.SE

Instruction Stacking Collapse: A Benchmark and the Capability-Dependent Value of Prompt Compilation

Production prompts rarely carry a single instruction. One system message may require valid JSON, a word limit, three citations, and a fixed tone at the same time. We study how instruction-following degrades as such constraints accumulate. We introduce a benchmark that stacks 24 verifier-checked instructions, one to twenty at a time, and evaluate three production-tier LLMs (Claude Sonnet 4.6, GPT-5-mini, Gemini 2.5 Flash). Instruction-following degrades non-linearly: the follow rate falls from ~96% to as low as 20%, driven by a structured and reproducible set of pairwise conflicts. A single "output JSON" constraint, for example, is jointly unsatisfiable with nine others. We then evaluate a training-free remedy: an instruction compiler that rewrites the stacked prompt in a single LLM call and is reused across queries. Its benefit is capability-graded. It recovers up to +11 points of follow rate for weaker models, which are also the models most often deployed at scale, while leaving stronger models, which already internalise the same structure, essentially unchanged. Cluster-robust tests, same-baseline controls, and a within-family scaling ladder attribute the gain to the rewrite itself rather than to additional tokens, reordering, or measurement headroom. We release the benchmark, verifiers, and cached runs for full reproduction.

Atul Anand, Sourav Chattaraj

LLM Evaluation LLM Prompting Instruction Following

Jul 31, 2026cs.IR

Language Models Agree With Each Other, Not With Readers

Claims that language models homogenise are usually measured against human judgements collected for the study, which makes the human side an artifact of the design: a crowdworker given the model's instruction is running the model's prompt. We measure convergence against a human reference nobody built for the purpose -- 2,523 reader mark sets across 120 web documents, produced by people highlighting for their own reasons on a platform where the overlay of others' marks is off by default. Agreement is the overlap between two size-matched sentence sets minus the overlap expected when each is resampled within its own depth-and-length bands. The null's calibration is demonstrated, not asserted: every pair involving a random baseline lands within 0.006 of zero. On the median document each party names 14 sentences of 70; two readers share 4.1 and two models 8.7. Across 18 model arms spanning 11 vendors, 3 countries and both weight regimes, the median of 153 model pairs is +0.093 against a human yardstick of +0.040, and 99 sit entirely above the human interval. Two frontier models from rival labs reach +0.203, twice what GPT-4o agrees with itself on a second call. The effect is not determinism, prompt wording, procedure, vendor or routing, and it is graded: the smallest models agree at the human level. No model agrees with readers detectably more than a reader does, and at equal depth and length no surface feature separates their choices. The multiples are procedure-dependent and the ordering is not: models are cut to their sharpest set while a reader's is a random draw from what they marked, and blunting the models alike halves the gap without closing it. Tested out of sample on four models released after this analysis, against predictions fixed beforehand, none clears the human interval. A population simulated from several models is not several populations.

Kazuki Nakayashiki, Keisuke Watanabe

Glasp Inc.

LLM Evaluation Human-AI Agreement

Jul 31, 2026cs.CL

CalibratedRubric: Task-Adaptive Rubric Banks for Open-Ended LLM Evaluation

Reliable evaluation of open-ended LLM outputs requires fine-grained rubrics, yet expert curation is costly and difficult to scale. Existing automated pipelines rely on strict judge unanimity and binary variance filters, which cannot distinguish measurable rubrics from informative ones. We introduce CalibratedRubric, a task-adaptive framework that combines type-specific scoring, Bayesian rubric-measurability filtering, and item response theory (IRT)-based bank assembly. CalibratedRubric estimates each rubric's measurability with a Beta--Bernoulli agreement posterior and uses a submodular information-coverage objective to construct compact rubric banks over the observed capability range. Across financial, healthcare, general, and legal benchmarks, measurability filtering improves human-gold agreement on JudgmentBench from

κ=0.604

to

0.743

. IRT-based greedy selection improves cross-fitted rank fidelity over random selection across all six evaluated response blocks and requires only 49 rather than 131 rubrics to reach the target correlation on FinResearchBench decision-support tasks. Task-label perturbations further reduce system separation, confirming the practical relevance of task-adaptive scoring. These results support CalibratedRubric as an efficient, uncertainty-aware approach to open-ended LLM evaluation, with calibration gains depending on sufficient judge redundancy.

Mengting Chen, Yanshu Sun, Wanting Liang +5

FinStep · StepFun

Open-Ended Generation LLM Evaluation Language Model Generation Evaluation Rubric-Based Evaluation Automated Evaluation

Jul 31, 2026cs.CL

Hy-MultiTurn: A Six-Dimensional Benchmark for Deep Multi-Turn Dialogue Understanding

Long-running multi-turn interactions with chatbots and agents are now common, and a correct response often depends on remembering earlier details, tracking later revisions, identifying intended objects or referents, and withholding action when required conditions are unmet. Existing multi-turn benchmarks typically cover short exchanges and do not fully evaluate these capabilities in long multi-turn interactions, particularly in Chinese, while offering limited insight into how and why models fail. To address these limitations, we analyze real chatbot failures to identify six recurring mechanisms and use them to define six controlled evaluation modes in Hy-MultiTurn, a Chinese benchmark for deep multi-turn dialogue understanding. The six modes evaluate constraint memory, precise execution, constraint synthesis, object localization, action suppression, and reference resolution. Across the six modes, we construct 209 controlled tasks spanning 12-76 turns, with dialogue length, irrelevant-topic distraction, and colloquial phrasing adding further difficulty. Evaluation of 22 frontier model configurations shows that Hy-MultiTurn is broadly challenging, as even GPT-5.5, the strongest overall configuration, satisfies all requirements in only 41.1 percent of responses and no model performs best in all six modes.

Eileen Ye, Jiawen Tao, Yaoming Li +7

Hunyuan Team, Tencent · Peking University

Conversational Memory LLM Evaluation Multi-Turn Dialogue Evaluation

Jul 31, 2026cs.CL

Semantics of Subterfuge: Benchmarking Legal Deception Detection Against General-domain State-of-the-Art

Deception detection has critical implications for legal proceedings, law enforcement, and online security. Although human judgment is limited in accuracy and scalability, Natural Language Processing (NLP) offers a data-driven alternative. We present a survey and comparative analysis of NLP-based Automatic Deception Detection (ADD) focusing on the legal domain, reviewing the evolution from feature-based machine learning to Large Language Model (LLM) approaches. We conduct a unified empirical evaluation across seven datasets (two legal, five general-domain), comparing six fine-tuned transformer models and seven LLMs under four prompting strategies. The results show strong domain sensitivity, with fine-tuned models excelling in data-rich general domains and few-shot LLMs remaining competitive in low-resource legal settings. Chain-of-Thought prompting often underperforms direct classification. These findings highlight the need for domain adaptation and interpretable systems in high-stakes legal contexts.

Theekshana Samaradiwakara, Nisansa de Silva, George C. Lobb

Department of Computer Science & Engineering University of Moratuwa, Sri Lanka · The Law Office of George C. Lobb Austin, Texas, USA

Legal NLP LLM Evaluation

Jul 31, 2026cs.LG

Benchmarking Frontier Large Language Models Against Official Crash Database Coding Using Police Crash Narratives

Police crash narratives contain information that may supplement structured crash databases, but manual review is labor-intensive and it remains unclear how well large language models (LLMs) reproduce official crash coding. This study benchmarked six frontier LLMs by comparing narrative-derived crash attribute codes with corresponding fields in the Arkansas fatal-crash database. The analysis linked 5,587 fatal-crash narratives with 5,889 structured crash records from Arkansas (2015-2025), yielding 4,194 matched crashes. Six LLMs were evaluated using an identical zero-shot prompt to code crash manner, non-motorist relation, intersection type, work-zone relation, roadway surface condition, and light condition. Performance was evaluated using agreement, macro-averaged F1 score, Cohen's kappa, coverage, selective agreement, and comparisons with always-majority, always-Unknown, and keyword-rule baselines. Repeated-measures analyses and a generalized estimating equations model assessed differences among models and attributes. GPT-5.5 High achieved the highest agreement among the evaluated LLMs, but the always-majority baseline produced higher raw agreement and the keyword-rule baseline achieved macro-averaged F1 score and Cohen's kappa comparable to the best-performing LLM. Agreement was highest for non-motorist relation and crash manner and lowest for light condition, roadway surface condition, and work-zone relation. Differences across crash attributes exceeded differences across models. These results provide a benchmark for evaluating LLM-based crash coding and show that deployment should be evaluated on an attribute-specific basis using transparent baselines and human review.

Sudhir Bharati, Rajendra K C Khatri, Sudip Bharati

Planning and Research Division, Arkansas Department of Transportation, Little Rock, Arkansas, USA · Department of Mathematics, Philander Smith University, Little Rock, Arkansas, USA · Itahari International College, Itahari, Nepal

Jul 30, 2026cs.CL

Benchmarks Are Not Validation: A System-Level View of Financial LLM Applications

Large language models are increasingly deployed in financial applications that combine retrieval, proprietary data, tool use, orchestration logic, monitoring, and human escalation. Yet evaluation often remains model-centric: benchmark scores, task accuracy, or one-off qualitative reviews are treated as evidence of readiness. In financial settings, this is insufficient. We take the position that financial LLM systems should not be approved for production based on benchmark performance alone. They require system-level validation evidence across the application stack: data, model design, retrieval and generation performance, agent behavior, governance, and implementation. Drawing on industry experience validating GenAI applications in financial institutions, we outline a multi-layer validation view and explain why hybrid evaluation is necessary. We discuss where LLM-as-a-judge methods are useful and why they require controls such as multiple judges, rubrics, agreement, and auditability checks. We also highlight failure modes poorly captured by static benchmarks, including retrieval failures, unfaithful generation, tool misuse, escalation errors, and operational instability. Our position is that financial LLM validation should be an ongoing system discipline rather than a one-time model scoring exercise. Validation should produce decision-ready evidence, not only scores. We conclude with a research agenda for system-aware benchmarks, agent trace validation, judge alignment protocols, and lifecycle validation standards.

Burak Payzun, İrem Demirtaş, Simona Scala +2

Prometeia S.p.A.

Financial Services LLM Evaluation Finance

Jul 30, 2026cs.CL

Benchmarks Are Not Monolithic: Sample-Level Auditing and Orchestration for LLM Evaluation

Benchmark datasets are central to evaluating Large Language Models (LLMs), yet they are typically conceived as monolithic tasks, obscuring substantial variation in the demands of individual samples. We introduce a dataset-centric meta-evaluation framework that audits benchmark datasets at the sample level along five latent dimensions: 1. Cognitive and Knowledge Demands, 2. Language and Content Quality, 3. Task Properties, 4. Context, and 5. Ethics, Safety, and Fairness. Applying this framework, we annotate five influential benchmarks -- MMLU, ARC, WinoGrande, HellaSwag, and TruthfulQA -- revealing pronounced internal heterogeneity that is not captured by aggregate accuracy scores. We show how these annotations enable criterion-driven orchestration of composite benchmark subsets across datasets, supporting targeted evaluation of model capabilities such as Reasoning Depth or Ethical Sensitivity. This approach reframes benchmark evaluation as dataset introspection, providing a principled methodology for analyzing and re-composing existing benchmarks to better reflect diverse evaluation needs.

Philipp D. Siedler, Jordan Sassoon

Aleph Alpha Research, Heidelberg, Germany.

LLM Evaluation Benchmark Auditing Benchmark Construction LLM Auditing

Jul 30, 2026cs.AI

When Derived Measurements Mislead: Quantifying and Mitigating LLM Over-Trust with Privileged-Modality Reliability Evidence

Derived measurements increasingly enter large language model (LLM) pipelines as direct facts despite their instance-dependent validity. We define derived-feature over-trust (DFOT) as the failure in which a downstream LLM assigns such a measurement the epistemic status of a direct fact or uses it outside its valid scope. Using physiological sensing as a case study, D1 tests acceptance of a PPG-derived rhythm contradicted by offline ECG, whereas D2 tests rejection of an offline-confirmed reliable PPG rhythm under misleading severe history. ECG supplies training supervision and offline reference construction but is never shown to the LLM. Five estimands quantify this chain: conflict over-trust rate (COTR) and context-induced error rate (CIR) characterize D1/D2; correct repair rate (CRR) measures frozen-error repair; evidence-specific repair margin (ESRM) contrasts matched and patient-disjoint shuffled evidence; and utility harm rate (UHR) measures unnecessary verification among HIGH-reliability cases used without verification at baseline. The framework does not depend on a particular reliability generator. We demonstrate it on 50,000 paired PPG-ECG records using ECG-to-PPG privileged distillation as an illustrative baseline and PPG-only inference. On a protocol-locked 187-patient test, the baseline improves four repair and specificity endpoints by 1.82-6.69 percentage points, with all paired confidence intervals excluding zero; UHR increases by 0.67 percentage points (95% CI: -0.4 to +1.7). DFOT provides a common evaluation target for stronger mitigation methods. The code is available at https://github.com/Zongheng-Guo/When-Derived-Measurements-Mislead.

Zongheng Guo, Tao Chen, Tianli Li +6

Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milan, Italy · State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou, China · Department of Integrated Traditional Chinese and Western Medicine Cardiology, China-Japan Friendship Hospital, Beijing, China +2

LLM Evaluation LLM Reliability Photoplethysmography Trust in AI

Jul 30, 2026cs.AI

When Specifications Conflict: A Symmetry-Based Framework for Measuring LLM Preferences

Large language models (LLMs) are increasingly required to integrate multiple sources of information that may be inconsistent or conflicting. However, there is still a lack of controllable and attributable methods for analyzing how models resolve conflicts between competing specifications. We propose a controlled experimental framework for studying model preferences under conflicting specifications. By constructing specifications with explicit conflicts, the framework enables model choices between competing specifications to be directly observed and analyzed. A symmetry-based design further reduces confounding factors, allowing preferences across representation types to be compared systematically. We evaluate the framework on an executable mathematical benchmark with 550 conflict instances spanning 11 function families, comparing four representation types: pure natural language, formal language, naturalized formal language, and input--output examples. Results show systematic preference patterns rather than random behavior, with a consistent ordering:

\text{Formal} \approx \text{Naturalized Formal} > \text{Pure Natural Language} > \text{Input--Output Examples}

. Example effects further depend on model capability and function family. We extend the framework to heterogeneous specification conflicts in Boolean algebra, code generation, and the clinical domain, demonstrating its applicability across diverse tasks and specification forms. The framework provides a unified approach for measuring how LLMs resolve conflicts between competing sources of information.

Tairan Wang, Liang Zhou, Zikang Zhan +1

Department of Computer Science University College London London, United Kingdom

LLM Evaluation Knowledge Conflicts in Language Models

Jul 30, 2026cs.CL

(Towards) Scalable Reliable Automated Evaluation with Large Language Models

Evaluating the quality and relevance of textual outputs from Large Language Models (LLMs) remains challenging and resource-intensive. Existing automated metrics often fail to capture the complexity and variability inherent in LLM-generated outputs. Moreover, these metrics typically rely on explicit reference standards, limiting their use mostly to domains with objective benchmarks. This work introduces a novel evaluation framework designed to approximate expert-level assessments of LLM-generated content. The proposed method employs pairwise comparisons of outputs by multiple LLMs, reducing biases from individual models. An Elo rating system is used to generate stable and interpretable rankings. Adjustable agreement thresholds, from full unanimity to majority voting, allow flexible control over evaluation confidence and coverage. The method's effectiveness is demonstrated through evaluating competency profiles extracted from scientific abstracts. Preliminary results show that automatically derived rankings correlate well with expert judgments, significantly reducing the need for extensive human intervention. By offering a scalable, consistent, and domain-agnostic evaluation layer, the framework supports more efficient and reliable quality assessments of LLM outputs across diverse applications.

Bertil Braun, Martin Forell

KIT

LLM Evaluation Pairwise Comparison LLM-as-a-Judge LLM Reliability Automated Evaluation

Jul 30, 2026cs.CL

MORFES: A Benchmark for Productive Inflectional Competence in Modern Greek

Modern Greek is a richly inflected language, yet the language models built for it are evaluated mainly on factual knowledge, and no benchmark is dedicated to their inflectional competence. We introduce MORFES (Morphological Open-class Recognition-and-Formation Evaluation Suite), a benchmark of 500 expert-verified items that tests the recognition and production of Greek inflected forms, favoring lower-frequency lemmas so that a correct answer reflects the rule rather than a memorized form. We make it publicly available at https://huggingface.co/datasets/KIEFERSA/MORFES. We evaluate a range of open language models on MORFES, situating them within the rapidly scaling open-weight ecosystem from LLaMA to Qwen3, DeepSeek-R1, Magistral, and Kimi K2, where multilingual coverage grows but grammatical competence in morphologically rich languages remains under-measured. Among them, Sophea-Genesis-1, a model we developed and release as open weights at https://huggingface.co/KIEFERSA/Sophea-Genesis-1, leads on inflectional morphology while matching similarly sized models in general capability.

Ioakeim Perros, Cleopatra Papadopoulou, Ayoub Kirouane +1

Sophea AI, KIEFER SA, Athens, Greece

Multilingual Language Model Evaluation LLM Evaluation

Jul 30, 2026cs.CL

Rethinking LLM-Judged Helpfulness as a Pedagogy Signal: A Pre-Registered Audit Across Tutor Models

LLM tutoring poses a measurement problem: can a general-purpose helpfulness rubric distinguish direct answer-giving from pedagogical guidance? We audit this signal in a pre-registered study. Within each of three tutor bases, we compare conversational and pedagogical policies instantiated with the same underlying model and paired with one fixed weak simulated student. Deterministic detectors measure answer leakage and next-turn independent work. Claude Opus 4.8 is the frozen, condition-blind primary judge. After the Opus scores were fixed, GPT-5.6 Sol was prospectively specified for a post hoc robustness audit of the same 1,179 confirmatory answer-phase tutor turns under the frozen helpfulness and pedagogy rubrics. On the primary base under Opus, the policies do not differ significantly in helpfulness but are perfectly rank-separated under the pedagogy rubric (Cliff's

|δ|{=}0.10

vs.

1.0

). Across the two judges, pedagogy contrasts retain their direction where detected, whereas the helpfulness ordering is judge-contingent, reversing between judges on two of three bases. In an Opus-only ablation, seven primary-base policies span

2.3

points in mean judged pedagogy within a

0.25

-point band of mean judged helpfulness. Separately, answer-revealing turns are followed by less independent student work on every base, a result that is judge-invariant by construction. In this controlled setting, general-purpose helpfulness is not a reliable pedagogy signal. Tutor evaluation should pair pedagogy-targeted rubrics with deterministic process measures.

Shuyi Fan, Boyuan Deng, Mengyu Xu +4

Columbia University, USA · Johns Hopkins University, USA · The University of Chicago, USA +4

LLM Evaluation LLM-as-a-Judge Intelligent Tutoring Systems Generative AI in Education

Jul 30, 2026cs.CL

Challenges in annotations by humans and LLMs: A case study of evaluative language

In this paper, we draw a comparison between linguists in training, a trained linguist, and annotations generated by large language models (LLMs) to find out if they struggle with complex linguistic phenomena in a similar way. For this purpose, we analyse evaluative language in spoken popular science discourse, with the example of a corpus of English TED talk transcripts. We focus on the Appraisal theory and its Attitude subsystem, including the categories (classes) of Affect, Judgement, and Appreciation. In this context, Appraisal theory is an example of a highly subjective annotation task, making it a suitable example for the study of complex annotation challenges. First, we assess human annotations on a sentence level in specific scientific domains. Then, we develop three prompts and compare them for model performance for the automatic classification of Appraisal classes. We assess the performance of three LLMs using the best-performing prompt and finetune the model, reaching an F1-score of 0.77. We find that models perform best compared to annotations conducted by the trained linguist, while linguists in training do not reach high agreement scores. We conclude that LLMs can aid in complex annotation task resolution, opening new pathways for the complex theories annotated and analyzed in digital humanities studies.

Mirela Imamovic, Aenne Cecilia Kristine Knierim, Khushi Pitroda +1

University of Hildesheim, Institute of Translation Studies and Specialised Communication, Germany · University of Paris 8 Vincennes-Saint Denis, TransCrit, France · University of Hildesheim, Institute of Information Science and Language Technology, Germany

LLM Evaluation LLM-Assisted Annotation Human-in-the-Loop Annotation

Jul 30, 2026cs.AI

An Instrument to Evaluate Governance Proposals: AI Policy Analysis at Scale

This paper introduces a policy analysis framework for systematic, transparent assessment of AI governance proposals in an evolving and contested regulatory landscape. AI policy debates often collapse into binary positions that obscure underlying tradeoffs and normative assumptions. The framework structures policy analysis around multiple policy attributes, allowing users to surface priorities and tensions without prescribing outcomes. We use a mixed-methods approach that integrates qualitative insights from subject matter experts with computational text analysis to inform the design of policy attribute rubrics. This quantifies the relative emphasis of different policy objectives and presents them through comparative visualizations that support interpretability and cross-policy comparison. The paper also examines the use of commercial LLMs for rubric-based policy analysis, benchmarking their outputs against a domain-trained rubric-calibrated model with explicitly defined analytical assumptions. Rather than assessing policy effectiveness or desirability, the framework focuses on relevance and alignment across attributes. By making analytical assumptions explicit, including attribute selection, rubric construction, and weighting schemes, the framework enables users to evaluate whether its embedded priorities align with the users' own normative commitments. The approach is jurisdiction-agnostic and intended to support policymakers, analysts, and researchers navigating complex AI governance environments. Contributions: (1) multidimensional policy assessment through empirically grounded rubrics that surface tradeoffs rather than resolving them; (2) a transparent hybrid methodology combining feedback from subject-matter experts with computational validation; and (3) use of domain-trained rubric-calibrated models as a benchmark for comparing different general-purpose large language models.

Paulo Carvao, Claudio Mayrink Verdun, Isabel Adler +1

Mossavar-Rahmani Center for Business and Government · Harvard Kennedy School · Harvard Kennedy School of Government +2

LLM Evaluation Policy Evaluation Rubric-Based Evaluation AI Governance

Jul 30, 2026cs.CL

RepBench: Compiling Benchmarks into Capability Representations for Large Language Models

Representation engineering reads and steers capability directions in large language models, yet methods are typically evaluated on paper-specific synthetic data. The resulting measurements are difficult to compare or reproduce and may reflect surface patterns rather than capabilities. We present RepBench, a benchmark-grounded data layer for capability-aligned representation probing. Crawling 13,427 benchmark papers yields a taxonomy of 182 capability clusters in 13 families; harvesting 353 public benchmark datasets yields 46,149 audited probe texts covering 94 capabilities, each supported by at least two independent benchmarks. This multi-benchmark design reduces dependence on any single source: raw per-text vectors exhibit no natural cluster granularity, whereas benchmark-pooled capability vectors show an interior clustering optimum at a small number of clusters on all 12 evaluated models, with low agreement to the human taxonomy. Under cross-benchmark transfer evaluation across twelve models completed by all four readouts, difference-in-means attains the highest model-level mean on ten models, while logistic regression wins the most capability-model cells. This disagreement shows that the readout method and aggregation criterion are meaningful evaluation dimensions. The pipeline, corpus, and evaluation code are released as a reusable closed-loop workflow.

Yanshi Li, Xueru Bai, Shuman Liu +1

Shopee, Shanghai, China · Shopee, Beijing, China

LLM Evaluation LLM Interpretability Representation Geometry in Language Models Language Model Probing

Jul 30, 2026cs.AI

Share the Judge, Learn the Deferral: Where Specialization Helps LLM Evaluation

Agentic systems have widened the gap between producing candidate outputs and reviewing them. This paper asks a practical architectural question: should domain specialization be built into an evaluator's weights, or into the rule that decides when its judgment can be trusted? We study 99,952 public, rubric-conditioned examples. Supplying the correct rubric improves locked-test accuracy by 2.11 points over a response-only control; replacing it with an unrelated rubric costs 2.66 points. Dividing the same training corpus among eight criterion-family LoRA judges, however, loses 10.05 points and cuts audited coverage at a 5% risk target from 24.44% to 5.43%. Matching the bank's stored capacity with one rank-64 adapter does not reproduce this loss. Nor is the result explained by learning rate or optimizer steps. Initializing the family adapters from a shared, trained judge recovers test accuracy to 76.85%, 19.94 points above scratch training at the same learning rate (95% interval 18.88-21.02). The result changes when specialization governs deferral rather than judgment. On RewardBench 2, learned correctness heads route examples through a 0.6B-4B-8B cascade without changing any reward score. Across 20 locked repartitions, the cascade attains 89.40% accuracy, compared with 84.75% for 8B alone, at 0.415 normalized parameter compute. Every run passes an exact one-sided 95% risk audit; margin-based rules remain near 84.8% accuracy while using at least 0.94 compute. These results suggest a qualified design rule: share the learning of judgment until there is enough data to justify a split, and place domain-specific adaptation in an audited release boundary.

Weining Zhang

Cheung Kong Graduate School of Business

LLM Evaluation LLM-as-a-Judge Selective Prediction Rubric-Based Evaluation

Jul 30, 2026cs.AI

IFHierBench: Hierarchical Instruction Following for Large Language Models

Instruction-following ability is critical for deploying large language models in real-world applications, where downstream components depend on the output satisfying specific constraints. Modern deployments increasingly handle the full task in a single LLM call, with one prompt specifying a layered output whose overall artifact, structural sections, and nested fields must each satisfy concrete constraints. Existing instruction-following benchmarks treat the constraint set as a flat list applied uniformly to the response, so they cannot scope a check to a particular section of the output. We introduce IFHierBench, a hierarchical instruction-following benchmark of 600 prompts stratified across four constraint-tree depths and 35 distinct constraints, each prompt paired with a deterministic checker that verifies satisfaction at every scope. Evaluating seven leading proprietary and open-weight models, we find that even the strongest model only marginally exceeds 50% prompt-level accuracy and that accuracy degrades sharply as constraint depth grows. Reliably following nested constraints remains a substantial gap for current LLMs, motivating future training methods that consider constraint adherence at finer granularity to achieve better instruction-following ability.

Yuetian Mao, Chunyang Chen

Technical University of Munich

LLM Evaluation Language Model Generation Evaluation Instruction Following

Jul 30, 2026cs.CL

AutoSupervision: Closing the Feedback Loop in Scientific Workflows with Grounded Revision Verification

Recent advances in large language models (LLMs) have enabled AI systems to assist scientific research and peer review. However, an essential capability for reliable AI-assisted scientific workflows remains underexplored: verifying whether reviewer feedback leads to meaningful and evidence-supported manuscript improvements. We introduce AutoSupervision, which evaluates whether scientific manuscript revisions genuinely address reviewer concerns through grounded evidence. AutoSupervision leverages transparent peer-review records as a natural source of supervision, where reviewer comments specify scientific concerns, author responses describe claimed resolutions, and revised manuscripts provide evidence of changes. Given reviewer comments, author responses, and revised manuscripts, models must characterize reviewer concerns, determine whether concerns have been addressed, and identify supporting manuscript evidence. We construct AutoSupervision from 56,000 Nature Communications articles and corresponding review records. Then we conducted experiments on LLMs, the ablation study, and the case study. Our results show that while LLMs perform well in characterizing reviewer concerns, with GPT-5.5 achieving a score of 0.754, evidence-based verification remains the primary bottleneck, with the best-performing model reaching only 0.501.

Haobo Li, Eunseo Jung, Wenxiao Zhao +8

1Shanghai AI Laboratory · 3The Hong Kong University of Science and Technology · University of California, Los Angeles

LLM Evaluation Scholarly Peer Review Automated Peer Review

Jul 30, 2026cs.CL

Beyond Borrowed Histories: Person-Aligned User Simulation for Interactive Role-Playing Evaluation

Role-playing agents (RPAs) have become one of the most important consumer applications of large language models. Users engage in multi-turn conversations with RPAs for experiences such as emotional comfort, making reliable evaluation essential for measuring capability, comparing systems, and guiding further improvement. Existing benchmarks, however, typically require an RPA to continue a fixed dialogue history and then evaluate the continuation using a fixed rubric detached from the user. We identify and empirically demonstrate two limitations of this design. First, an RPA's output is shaped by the preceding dialogue history, preventing a scientifically grounded assessment of its role-playing ability in real multi-turn settings. Second, user experience varies substantially across individuals, and conventional fixed rubrics need not align with user satisfaction. We therefore introduce PALATE (Person-Aligned LLM-Simulated-User Assessment with Tailored Evaluation), a scalable RPA benchmark built on user simulators. PALATE is accompanied by a pool of 300 character profiles. Its main evaluation trains five per-user simulators and lets them engage candidate RPAs in free-form, multi-turn conversations over a pre-frozen panel of character profiles. Alongside a general quality rubric, we construct personalized rubrics to measure user satisfaction; on held-out annotated data, the personalized rubrics show higher agreement with human judgments than the general rubric. In the main evaluation of 16 candidates, PALATE separately characterizes generic turn quality, long-horizon session capability, and per-user experience on multi-turn trajectories co-constructed by each candidate. It thereby produces interpretable evaluations of specific user-RPA pairs rather than compressing systems into a single user-independent ranking.

Yuhang Zhu, Mingxuan Du, Benfeng Xu +3

University of Science and Technology of China · MetaStone · MetaStone Technology, Beijing, China

LLM Evaluation Large Language Model-Based Role-Play Simulation LLM Agent Evaluation Agent Evaluation Multi-Turn Dialogue Evaluation User Simulation

Jul 30, 2026cs.IR

Measuring Alignment With Reader Highlights Net of Position and Length

Context compression discards most of a document before a language model reads it, and is normally evaluated by downstream task accuracy - which makes another model the judge of what mattered. Naturalistic social highlighting offers a non-circular reference: many people independently marking passages on the same page. But the obvious metric, the fraction of crowd-marked sentences a compressor keeps, is confounded twice: crowd marks are front-loaded and crowd-marked sentences are longer, so any method favouring early or long sentences scores well regardless of readers. We remove both by matching each marked sentence against unmarked sentences of the same document at equal relative depth and equal within-document length rank, and we calibrate every estimator on synthetic nulls built from position and length alone - a step that matters, since depth-only stratification returns a false positive on 20-36% of nulls containing no effect. On 120 web documents (at least 12 independent readers each), a language-model importance ranking keeps 38.4% of crowd-marked sentences against 19.9% of their matched neighbours: an enrichment of +0.196 [+0.148, +0.239], at p = 0.0005 under an exact randomization test that assumes nothing about clustering, and replicated cross-vendor. Naive truncation, whose keep rule is position, correctly falls to +0.003. To give the number a scale: scored identically, on the same budget, against a crowd label recomputed to exclude them, a single human reader reaches +0.182 - indistinguishable from GPT-5.4 (+0.002 [-0.081, +0.088]) and below Claude Opus 5. Classical methods are not null - Luhn's 1958 heuristic reaches +0.088 - so reader selection is partly recoverable by counting words; conditioning additionally on lexical centrality removes only 0.010, so the agreement is not centrality. We also report that a claim in our own prior work does not reproduce on this corpus.

Kazuki Nakayashiki, Keisuke Watanabe

Glasp Inc.

LLM Evaluation Long-Context Language Model Evaluation Context Compression

Jul 30, 2026cs.CL

From Single- to Cross-Document: Benchmarking Multi-Granularity Event Analysis of Large Language Models

Event analysis is an essential and fundamental direction of information extraction, involving various event-centric tasks at different granularity of documents. While large language models (LLMs) have preliminarily achieved promising performance in part of these tasks individually, their capability in event analysis still lacks comprehensive understanding due to restricted document granularity, task designs, and data source of existing benchmarks. To address these limitations, we introduce MiGUE-Bench, a systematic benchmark for assessing the performance of LLMs in multi-granularity event analysis. To support large-scale evaluation, we first develop an LLM-driven self-correcting annotation framework called MiGUE-Pipeline, enabling scalable acquisition of high-quality source data of events with automatic labels. Then, we design four core tasks in our benchmark, i.e., event detection, relation reasoning, structure induction, and future prediction, to probe model competence at different levels, from atomic event details to complex cross-document narratives. Extensive experiments on state-of-the-art LLMs and retrieval-augmented generation (RAG) methods delineate the current capability boundary and identify critical deficiencies, providing insights into the future improvement of LLMs in challenging event analysis tasks.

Tao Wen, Shuai Shao, Pei Ke +7

Laboratory of Intelligent Collaborative Computing University of Electronic Science and Technology of China Chengdu, China · Department of Computer Science and Technology Tsinghua University Beijing, China · School of Computer Science and Engineering University of Electronic Science and Technology of China Chengdu, China +1

LLM Evaluation LLM Information Extraction