The Stability Trap: Evaluating the Reliability of LLM-Based Instruction Adherence Auditing

• URL: http://arxiv.org/abs/2601.11783v1
• Date: Fri, 16 Jan 2026 21:15:13 GMT
• Title: The Stability Trap: Evaluating the Reliability of LLM-Based Instruction Adherence Auditing
• Authors: Murtuza N. Shergadwala,
• Abstract summary: This study asks: To what extent does the instruction type of an Application Under Test (AUT) influence the stability of judge evaluations?<n>We introduce the Scoped Instruction Decomposition Framework to classify AUT instructions into Objective and Subjective types, isolating the factors that drive judge instability.<n>Our results reveal a Stability Trap'' characterized by a divergence between Verdict Stability and Reasoning Stability.
• Score: 1.5954459915735735
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The enterprise governance of Generative AI (GenAI) in regulated sectors, such as Human Resources (HR), demands scalable yet reproducible auditing mechanisms. While Large Language Model (LLM)-as-a-Judge approaches offer scalability, their reliability in evaluating adherence of different types of system instructions remains unverified. This study asks: To what extent does the instruction type of an Application Under Test (AUT) influence the stability of judge evaluations? To address this, we introduce the Scoped Instruction Decomposition Framework to classify AUT instructions into Objective and Subjective types, isolating the factors that drive judge instability. We applied this framework to two representative HR GenAI applications, evaluating the stability of four judge architectures over variable runs. Our results reveal a ``Stability Trap'' characterized by a divergence between Verdict Stability and Reasoning Stability. While judges achieved near-perfect verdict agreement ($>99\%$) for both objective and subjective evaluations, their accompanying justification traces diverged significantly. Objective instructions requiring quantitative analysis, such as word counting, exhibited reasoning stability as low as $\approx19\%$, driven by variances in numeric justifications. Similarly, reasoning stability for subjective instructions varied widely ($35\%$--$83\%$) based on evidence granularity, with feature-specific checks failing to reproduce consistent rationale. Conversely, objective instructions focusing on discrete entity extraction achieved high reasoning stability ($>90\%$). These findings demonstrate that high verdict stability can mask fragile reasoning. Thus, we suggest that auditors scope automated evaluation protocols strictly: delegate all deterministically verifiable logic to code, while reserving LLM judges for complex semantic evaluation.

Related papers

• VI-CuRL: Stabilizing Verifier-Independent RL Reasoning via Confidence-Guided Variance Reduction [55.04308051033549]
  Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a dominant paradigm for enhancing Large Language Models (LLMs) reasoning.<n>We introduceVerifier-Independent Curriculum Reinforcement Learning (VI-CuRL), a framework that leverages the model's intrinsic confidence to construct a curriculum independent from external verifiers.
  arXiv  Detail & Related papers  (2026-02-13T03:40:52Z)
• Gaming the Judge: Unfaithful Chain-of-Thought Can Undermine Agent Evaluation [76.5533899503582]
  Large language models (LLMs) are increasingly used as judges to evaluate agent performance.<n>We show this paradigm implicitly assumes that the agent's chain-of-thought (CoT) reasoning faithfully reflects both its internal reasoning and the underlying environment state.<n>We demonstrate that manipulated reasoning alone can inflate false positive rates of state-of-the-art VLM judges by up to 90% across 800 trajectories spanning diverse web tasks.
  arXiv  Detail & Related papers  (2026-01-21T06:07:43Z)
• A Comprehensive Evaluation of LLM Reasoning: From Single-Model to Multi-Agent Paradigms [20.241519889633285]
  Large Language Models (LLMs) are increasingly deployed as reasoning systems, where reasoning paradigms play a critical role.<n>We conduct a comprehensive and unified evaluation of reasoning paradigms, spanning direct single-model generation, CoT-augmented single-model reasoning, and representative MAS.<n>We introduce MIMeBench, a new open-ended benchmark that targets two foundational yet underexplored semantic capabilities.
  arXiv  Detail & Related papers  (2026-01-19T17:23:45Z)
• RULERS: Locked Rubrics and Evidence-Anchored Scoring for Robust LLM Evaluation [15.787947727055611]
  We introduce RULERS, a compiler-executor framework that transforms natural language rubrics into executable specifications.<n>RULERS operates by compiling criteria into versioned immutable bundles, enforcing structured decoding with deterministic evidence verification, and applying lightweight Wasserstein-based post-hoc calibration.
  arXiv  Detail & Related papers  (2026-01-13T15:31:42Z)
• Rubric-Conditioned LLM Grading: Alignment, Uncertainty, and Robustness [4.129847064263056]
  We systematically evaluate the performance of Large Language Models for rubric-based short-answer grading.<n>We find that alignment is strong for binary tasks but degrades with increased rubric granularity.<n>Experiments reveal that while the model is resilient to prompt injection, it is sensitive to synonym substitutions.
  arXiv  Detail & Related papers  (2025-12-21T05:22:04Z)
• ReasonBENCH: Benchmarking the (In)Stability of LLM Reasoning [2.1461777157838724]
  We introduce ReasonBENCH, the first benchmark designed to quantify the underlying instability in large language models (LLMs) reasoning.<n>Across tasks from different domains, we find that the vast majority of reasoning strategies and models exhibit high instability.<n>We further analyze the impact of prompts, model families, and scale on the trade-off between solve rate and stability.
  arXiv  Detail & Related papers  (2025-12-08T18:26:58Z)
• SSR: Socratic Self-Refine for Large Language Model Reasoning [78.62319252287938]
  Socratic Self-Refine (SSR) is a novel framework for fine-grained evaluation and precise refinement of Large Language Models (LLMs)<n>Our proposed SSR decomposes model responses into verifiable (sub-question, sub-answer) pairs, enabling step-level confidence estimation.<n> Empirical results across five reasoning benchmarks and three LLMs show that SSR consistently outperforms state-of-the-art iterative self-refinement baselines.
  arXiv  Detail & Related papers  (2025-11-13T18:47:07Z)
• TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them [58.04324690859212]
  Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks.<n>We identify two fundamental types of inconsistencies: Score-Comparison Inconsistency and Pairwise Transitivity Inconsistency.<n>We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations.
  arXiv  Detail & Related papers  (2025-09-25T13:04:29Z)
• Aurora: Are Android Malware Classifiers Reliable and Stable under Distribution Shift? [51.12297424766236]
  AURORA is a framework to evaluate malware classifiers based on their confidence quality and operational resilience.<n>AURORA is complemented by a set of metrics designed to go beyond point-in-time performance.<n>The fragility in SOTA frameworks across datasets of varying drift suggests the need for a return to the whiteboard.
  arXiv  Detail & Related papers  (2025-05-28T20:22:43Z)
• Retrieval is Not Enough: Enhancing RAG Reasoning through Test-Time Critique and Optimization [58.390885294401066]
  Retrieval-augmented generation (RAG) has become a widely adopted paradigm for enabling knowledge-grounded large language models (LLMs)<n>RAG pipelines often fail to ensure that model reasoning remains consistent with the evidence retrieved, leading to factual inconsistencies or unsupported conclusions.<n>We propose AlignRAG, a novel iterative framework grounded in Critique-Driven Alignment (CDA)<n>We introduce AlignRAG-auto, an autonomous variant that dynamically terminates refinement, removing the need to pre-specify the number of critique iterations.
  arXiv  Detail & Related papers  (2025-04-21T04:56:47Z)
• DAFE: LLM-Based Evaluation Through Dynamic Arbitration for Free-Form Question-Answering [12.879551933541345]
  We propose the Dynamic Arbitration Framework for Evaluation (DAFE) to evaluate large language models.<n>DAFE employs two primary LLM-as-judges and engages a third arbitrator only in cases of disagreements.<n>We show DAFE's ability to provide consistent, scalable, and resource-efficient assessments.
  arXiv  Detail & Related papers  (2025-03-11T15:29:55Z)
• Debate, Deliberate, Decide (D3): A Cost-Aware Adversarial Framework for Reliable and Interpretable LLM Evaluation [0.0]
  We present Debate, Deliberate, Decide (D3), a cost-aware, adversarial multi-agent framework that orchestrates structured debate among role-specialized agents.<n>We develop a probabilistic model of score gaps that characterizes reliability and convergence under iterative debate.<n>We show state-of-the-art agreement with human judgments, reduced positional and verbosity biases via anonymization, and a favorable cost-accuracy frontier.
  arXiv  Detail & Related papers  (2024-10-07T00:22:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.