The Test Oracle Problem in Synthetic LLM-as-Judge Corpora: Disappearance, Distortion and a Validation Protocol

• URL: http://arxiv.org/abs/2607.13707v1
• Date: Wed, 15 Jul 2026 11:12:54 GMT
• Title: The Test Oracle Problem in Synthetic LLM-as-Judge Corpora: Disappearance, Distortion and a Validation Protocol
• Abstract summary: Studies of bias in LLM-as-judge systems typically build synthetic corpora by prompting an LLM to generate a hallucinated answer to pair with a factual one, then presenting both to a judge.<n>We report a case in which this generation step silently failed, and use it to argue that the failure mode is structural rather than incidental.<n>A positive control supports this claim directly: an analogous fault injected into a minimal perturbation-based corpus is caught with 100% accuracy by a zero-cost, zero-human gold-to-negative string comparison.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Studies of bias in LLM-as-judge systems typically build synthetic corpora by prompting an LLM to generate a hallucinated answer to pair with a factual one, then presenting both to a judge. We report a case in which this generation step silently failed, and use it to argue that the failure mode is structural rather than incidental. In a multilingual (Turkish/English) faithfulness-judgment corpus, a decoding-budget parameter shared between judging and generation calls truncated one producer's hallucinated answers to a few words. The resulting items produced a large, statistically robust effect: a 32-point cross-lingual collapse in one judge's selection accuracy, replicated from N=50 to N=500, explained by a three-layer mechanistic account, and confirmed by a controlled producer-swap experiment, none of which was real. The effect vanished to ceiling once the shared parameter was corrected, and only manual reading of the raw generations, not any aggregate statistical check, exposed the fault. A second measured bias (Markdown-formatting preference) was not fabricated but distorted by the same fault, its magnitude and in one case its sign shifting with stimulus length, a mode aggregate metrics cannot distinguish from the first. We frame the underlying vulnerability using the test oracle problem: corpora whose negative examples are LLM-generated carry no mechanical way to verify item integrity, while corpora built by deterministic perturbation of a gold answer carry an item-level oracle for free. A positive control supports this claim directly: an analogous fault injected into a minimal perturbation-based corpus is caught with 100% accuracy by a zero-cost, zero-human gold-to-negative string comparison. We close with a validation protocol, derived from our own case, for analysts working in the oracle-less regime that we argue describes most contemporary multilingual LLM-as-judge corpora.

Related papers

• Four Ledgers, Not One Score: Responsible Communication of LLM-Judge Calibration in Biomedical ML [0.39287497907611874]
  We formalize four distinct ledgers: planted perturbations, independent detector outputs, source-linked human dispositions, and human-added discoveries.<n>This single-workflow forensic case is an existence proof of a failure mode, not an estimate of its prevalence.
  arXiv  Detail & Related papers  (2026-09-14T04:27:58Z)
• Walking on the DARKSIDE [2.0052993723676895]
  We introduce DARKSIDE, a coherence auditing method on top of POLANYI++.<n>It formalises the trail as an explicit data structure of accumulated exclusions over discourse time.<n>We evaluate it as a steering layer over a Gemini 3 on BSBench, a 100-item corpus of sophisticated nonsense across software engineering, finance, healthcare, physics and law, with Sonnet 4.6 as an independent judge.
  arXiv  Detail & Related papers  (2026-08-24T15:17:58Z)
• Distilling Black-Box Machine Learning into a Small, Self-Explaining Language Model for Learning Analytics [0.0]
  We propose a two-stage fine-tuning pipeline that distills a fitted black-box estimator and its post hoc interpretation (the mentor) into a small, open-weight large language model (LLM)<n>The design is estimator-agnostic and paired with a faithfulness-first evaluation framework that audits every narration against the attribution it claims to describe.<n>We find that fluency is no evidence of correctness since narration quality is independent of signal quality, and decision quality collapses toward the majority action in severely imbalanced settings.
  arXiv  Detail & Related papers  (2026-08-21T14:36:23Z)
• PluginEval: A Diagnostic Benchmark for Fine-Grained Error Attribution in Function Calling [11.1017602726296]
  Large Language Models increasingly operate as autonomous agents.<n>Current benchmarks face three structural limitations.<n>Eval is a benchmark constructed through a two-stage framework.
  arXiv  Detail & Related papers  (2026-08-09T13:25:54Z)
• AgentFairBench: Do LLM Agents Discriminate When They Act? [2.3004655342211078]
  AgentFairBench is a cheap, reproducible, multi-domain benchmark for demographic disparity in the actions of LLM agents.<n>It spans three regulator-anchored domains: hiring, lending, and medical triage.<n>A NumPy-only harness computes counterfactual flip rate, mean absolute score difference (MASD), action-rate disparity, and tool-invocation disparity.
  arXiv  Detail & Related papers  (2026-06-15T13:50:26Z)
• Pramana: A Protocol-Layer Treatment of Claim Verification in Autonomous Agent Networks [0.0]
  Production verification today splits into two halves: Probabilistic verdict patterns (self-consistency voting, reviewer LLM ensembles) produce judgments, not artifacts.<n>Pramana defines the missing wire format. Every consequential agent output is wrapped in a typed ClaimAttestation with one of four variants.<n>Pramana exhaustively verified under TLC across three symmetry-reduced models: 38,563 distinct reachable states, zero invariant violations.
  arXiv  Detail & Related papers  (2026-05-19T17:00:33Z)
• BiAxisAudit: A Novel Framework to Evaluate LLM Bias Across Prompt Sensitivity and Response-Layer Divergence [22.315546054051143]
  Bias audits of large language models now operate within governance frameworks such as the EU AI Act.<n>We introduce textscBiAxisAudit, a protocol that reports each bias score together with a reliability estimate on two axes.
  arXiv  Detail & Related papers  (2026-05-09T16:26:49Z)
• From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks [49.57538588967748]
  Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks.<n>We uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma.<n>We show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph.
  arXiv  Detail & Related papers  (2026-04-01T11:40:12Z)
• Blind to the Human Touch: Overlap Bias in LLM-Based Summary Evaluation [89.52571224447111]
  Large language model (LLM) judges have often been used alongside traditional, algorithm-based metrics for tasks like summarization.<n>We provide an LLM judge bias analysis as a function of overlap with human-written responses in the domain of summarization.
  arXiv  Detail & Related papers  (2026-02-07T19:39:28Z)
• A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness [57.510025257780306]
  We show that existing validation protocols fail to account for substantial distribution shifts inherent to red-teaming.<n>We propose ReliableBench, a benchmark of behaviors that remain more consistently judgeable, and JudgeStressTest, a dataset designed to expose judge failures.
  arXiv  Detail & Related papers  (2026-02-04T15:13:35Z)
• 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)
• MQM-APE: Toward High-Quality Error Annotation Predictors with Automatic Post-Editing in LLM Translation Evaluators [53.91199933655421]
  Large Language Models (LLMs) have shown significant potential as judges for Machine Translation (MT) quality assessment.<n>We introduce a universal and training-free framework, $textbfMQM-APE, based on the idea of filtering out non-impactful errors.<n>Experiments show that our approach consistently improves both the reliability and quality of error spans against GEMBA-MQM.
  arXiv  Detail & Related papers  (2024-09-22T06:43:40Z)
• Fact-Checking the Output of Large Language Models via Token-Level Uncertainty Quantification [116.77055746066375]
  Large language models (LLMs) are notorious for hallucinating, i.e., producing erroneous claims in their output. We propose a novel fact-checking and hallucination detection pipeline based on token-level uncertainty quantification.
  arXiv  Detail & Related papers  (2024-03-07T17:44:17Z)
• Sentiment Analysis through LLM Negotiations [58.67939611291001]
  A standard paradigm for sentiment analysis is to rely on a singular LLM and makes the decision in a single round. This paper introduces a multi-LLM negotiation framework for sentiment analysis.
  arXiv  Detail & Related papers  (2023-11-03T12:35:29Z)

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.