These findings position persona-conditioned judging as a controlled sensitivity probe for stress-testing LLM-based IR evaluation pipelines and identifying systems whose evaluation outcomes are sensitive to assessor framing.
Abstract
Large language models (LLMs) are increasingly used as relevance assessors in information retrieval (IR) evaluation, raising questions about how assessor framing affects judgment reliability and downstream system comparison. We study persona conditioning as a diagnostic mechanism for exposing LLM assessor sensitivity. Using task-oriented personas drawn from two complementary sources (PersonaHub and NVIDIA Nemotron-Personas-USA), we instantiate five assessor roles emphasizing intent interpretation, domain expertise, contrastive judgment, evidence verification, and global search-quality assessment, compared with a standard UMBRELA baseline. Across six LLM backbones on TREC DL20 and RAG24, our analyses reveal structured rather than uniform assessor sensitivity. Judgments usually remain close to the baseline while shifting assessment strictness, evidential threshold, or interpretation emphasis rather than producing widespread relevance reversals. At the system level, high-capacity models preserve system-ranking agreement, while smaller models amplify persona-induced instability. Local rank-displacement analysis shows sensitivity concentrates on particular retrieval systems and system types, especially neural ranking/reranking systems on DL20 and RAG-oriented pipelines on RAG24. Persona source matters less than assessor role and model capacity. These findings position persona-conditioned judging as a controlled sensitivity probe for stress-testing LLM-based IR evaluation pipelines and identifying systems whose evaluation outcomes are sensitive to assessor framing.
It is shown that prior scores, even when included only as context metadata, anchor judgments and systematically shift ratings toward their values, and effective mitigation must be validated for the intended model and task or domain.
A. Kapetanović, Kemal Altwlkany, Andro Merćep et al.· 0 citations
The rapid adoption of Large Language Models (LLMs) in educational assessment has reshaped scoring practices, yet evaluation remains tethered to aggregate reliability metrics like Quadratic Weighted Kappa, which obscure discrimination and rater effects. This study applies Signal Detection Theory to evaluate eight state-of-the-art LLMs (including Claude 3.5 Haiku, DeepSeek-V3, Gemini 3 Flash, GPT-4o, and Grok 4.1) against expert human raters across 1,726 essays. By decoupling discrimination from response criteria, I provide a diagnostic analysis of AI scoring behavior. Results indicate that human raters exhibit significantly superior evaluative precision, with average discrimination estimates approximately double those of the AI models. Furthermore, LLMs are prone to pronounced centrality effects and score compression, systematically failing to award the highest rubric tiers. These findings demonstrate that low human-machine agreement stems from both a deficit in discriminative accuracy and systematic shifts in response criteria. Ultimately, this research provides a robust framework for calibrating and selecting AI scoring systems based on specific pedagogical goals and fairness requirements.
Psychometric calibration for educational tests typically requires costly human response data. Large language models (LLMs) simulated examinees offer a promising route to early calibration, but their responses are too accurate and too uniform. We propose Cognitive Diagnostic Profiling (CDP), a zero-shot framework that prompts LLMs to simulate plausible examinees with diverse cognitive profiles: binary attribute-mastery patterns are rendered as natural-language profiles and sampled under an uninformative or an informative distribution. Using the Tatsuoka fraction-subtraction dataset (536 examinees, 15 items, five attributes), we evaluated eight LLM configurations under no-profile, uninformative-CDP, and informative-CDP conditions, assessing alignment with human examinees at the ability-distribution, mastery-profile, and item-difficulty levels. CDP improved all three levels: distributional overlap rose across configurations; weighted correlations between profile-level scores and human profile expectations reached 0.92 to 0.98; and item-difficulty recovery improved in rank order and absolute alignment, most for reasoning-enabled models; in the strongest case, Gemini 3.0 Flash (Thinking), one-parameter logistic (1PL) difficulty Spearman correlations rose from 0.24 to 0.86 and 0.90 and the root-mean-square error (RMSE) fell from 6.31 to 1.30 and 0.90; the informative condition helped most where profile-level alignment was strong. CDP brings LLM-simulated examinees into closer psychometric alignment with human examinees, making them practical for operational test development.
A fundamental disconnect is suggested between a model's capacity for factual accuracy and its ability to maintain social fairness, highlighting the need for multi-dimensional evaluation frameworks for small-scale systems.
M.J.F. Valdez, Arghir-Nicolae Moldovan· International Conference on...· 0 citations
Jailbreak robustness has become central to large language model (LLM) safety evaluation, yet prevailing methodologies rely primarily on refusal behavior, semantic resemblance, and intent-matching heuristics that emphasize linguistic plausibility rather than correctness. We identify a key limitation in existing evaluations: many jailbreak intents depend on instructional validity rather than epistemic factuality, allowing realistic-looking responses to be labeled successful despite being factually or procedurally incorrect. To address this gap, we propose Sequential Epistemic and Action-Level Validation (SEAV), a verification-centric jailbreak evaluation framework that decomposes responses into ordered steps and evaluates both validity and correctness. SEAV combines LLM-as-a-judge mechanisms for semantic interpretation with retrieval-grounded verification using external knowledge sources, assessing whether generated content is factually correct, structurally consistent, and operationally capable of advancing harmful objectives. Empirically, SEAV cuts the false-positive rate on SD-A (a curated strategic-dishonesty diagnostic) by 14.9\,pp vs. the strongest baseline, and reclassifies 22.1\%--51.0\% of sampled prior-labeled successes as invalid across three of four public benchmarks. Together, these results show that enforcing correctness substantially reshapes measured robustness: many previously labeled jailbreak successes are reclassified as invalid, and results are stable across the tested search backends and evaluator models. Code and data are available at https://github.com/Ardor-Wu/SEAV.
Qilong Wu, Sahil Wadhwa, Pranab Mohanty et al.· 0 citations
A consensus-based evaluation framework that measures relative preference among model-generated responses rather than absolute correctness rather than absolute correctness is introduced, offering an alternative perspective on response quality in scenarios where multiple valid answers exist.
Mohtashim Khan· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
