Accuracy Hides How Language Models Fail: Measuring Failure States Under Matched Output Budgets

• URL: http://arxiv.org/abs/2607.24268v2
• Date: Thu, 30 Jul 2026 06:23:36 GMT
• Title: Accuracy Hides How Language Models Fail: Measuring Failure States Under Matched Output Budgets
• Abstract summary: Language-model benchmarks collapse two distinct measurement questions into a single accuracy score.<n>We introduce a two-layer evaluation framework that separates scorer-independent execution evidence.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Language-model benchmarks collapse two distinct measurement questions into a single accuracy score: whether a response reached an evaluable state, and whether its answer was judged correct. We introduce a two-layer evaluation framework that separates scorer-independent execution evidence, including termination, answer exposure, parseability, and completion length, from scorer-dependent correctness. Across 2,550 outputs from five fixed Qwen and DeepSeek configurations on MATH and ARC-Challenge, matched 2,048-token limits produce sharply different execution mixtures: 49 of 450 Qwen MATH outputs terminate without a final answer, compared with 5 of 300 DeepSeek MATH outputs and none of the 750 ARC outputs. Among the same 300 DeepSeek MATH question-model pairs, no missing-final length termination is observed at 8,192 tokens. A coverage-audited targeted verification study further shows that candidate-selection and aggregation policies can substantially alter comparative accuracy estimates. These results demonstrate that accuracy conflates execution case mix with verification policy. Evaluations of test-time methods should therefore report pre-intervention execution states, verification coverage, and scorer provenance alongside accuracy.

Related papers

• Guardrailed Meta-Agent Loops: Stress-Testing Policy Pinning, Budget Bounds, and Crash Recovery [7.414694666903067]
  GuardrailLoop is a simulation-based testbed that makes three operational contracts jointly testable.<n>A hash-pinned policy fixes goals, scope, evaluation identity, budget, and release conditions.<n>It separates useful adaptation, state recovery, and repeated execution.
  arXiv  Detail & Related papers  (2026-09-10T21:22:08Z)
• Correct Tests Are Not Enough: Measuring and Training Oracle Conversion in Specification-Based Test Generation [14.44743150931186]
  Generating tests from a natural-language specification requires both an input that exposes faulty behavior and a correct expected output.<n>We study this interaction through executable reward decomposition and suite-level oracle-conversion measurement.<n>Our generator jointly emits five input--output tests in one response.
  arXiv  Detail & Related papers  (2026-09-05T04:56:31Z)
• MemToC: Benchmarking Memory-Tool Conflict Resolution in Large Language Models [49.47300047353726]
  We introduce MemToC, a benchmark for post-tool-return arbitration with executable tools.<n>MemToC comprises 6,504 evaluation episodes constructed from 542 quality-controlled factual questions.<n>Across five open-weight 7-9B models, tool returns strongly dominate elicited closed-book answers.
  arXiv  Detail & Related papers  (2026-08-26T18:22:03Z)
• Analysis of Federated Aggregation under Model Poisoning and Backdoor Attacks: A Reconstructed Cross-Dataset and Cross-Architecture Benchmark [0.0]
  A 500-cell seed-1 evaluation matrix was reconstructed across five aggregation methods.<n>Execution logs were identified for 454 original runs and 36 repaired or rerun executions.<n>Krum attained the highest recorded accuracy under both sign-flipping and Gaussian configurations.
  arXiv  Detail & Related papers  (2026-08-11T20:42:56Z)
• When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs [42.32694162421078]
  We introduce CALVER, a training-free symbolic verifier that scores structured traces against Pearl's causal criteria.<n>On CLEAR find-one-valid queries that admit multiple graph-valid answers, CALVER reaches 42.1% where plurality, a reward model, an LLM judge, and model confidence remain near 30%.
  arXiv  Detail & Related papers  (2026-08-04T11:45:46Z)
• Fantastic Adaptive Taxonomies and How to Use Them [100.20614173157708]
  AdaMAST converts execution traces into a compact, evidence-grounded failure taxonomy.<n>No code is hand-authored, no trace human-annotated.<n>In agent-system search, taxonomy-coded diagnoses of failed candidates outperform free-form.
  arXiv  Detail & Related papers  (2026-07-17T17:41:16Z)
• Benchmarking the Benchmarks: A Validity Audit of Tool-Calling Evaluation [3.079997053226693]
  We present a systematic validity and audit of four major tool-calling benchmark families.<n>Across 496 expert-reviewed benchmark tasks, we find 92 evaluator-human disagreements, corresponding to an 18.5% misalignment rate.<n>We introduce a unified taxonomy of tool-calling evaluation failures, release trace-level audit artifacts and corrected evaluation components, and argue for metrics that separately measure tool invocation, task completion, and outcome verification.
  arXiv  Detail & Related papers  (2026-06-30T21:10:42Z)
• Efficient Visual Pointing for Embodied AI:Agent-Driven Data Synthesis, Cross-Block Attention, and Iterative Correction [55.11480729304395]
  We describe our PointArena 2026 solution, which achieves 77.2% overall accuracy and ranks second on the benchmark.<n>The ap proach targets three failure modes. First, agent-driven syn thesis builds large semantic and anchor-relative candidate pools.<n>Second, a determinis tic steerable-data pipeline creates a verified 10,000-sample main set, plus reserve samples, using masks, templates, and path verification.
  arXiv  Detail & Related papers  (2026-06-29T06:39:03Z)
• Accuracy, Stability, and Repeated-Run Reliability of Large Language Models on Deterministic Programming Tasks [0.8699677835130409]
  Run-level pass rate overstates retry-free coverage by up to 17.8 percentage points.<n>We present a repeated-run evaluation protocol with metrics for run-level accuracy, retry-free coverage, and per-problem variability.
  arXiv  Detail & Related papers  (2026-05-30T23:03:05Z)
• FineVerify: Scaling Test-Time Compute with Fine-Grained Self-Verification for Agentic Search [88.16262636915975]
  FineVerify is a fine-grained self-verification framework for agentic search.<n>It decomposes each question into checkable sub-questions, verifies sampled candidates, and selects the candidate with the highest aggregated score.<n>FineVerify consistently outperforms standard scaling baselines.
  arXiv  Detail & Related papers  (2026-05-30T10:21:20Z)
• Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning [40.342912574072024]
  Large Language Models produce structured outputs such as travel plans or code solutions.<n>Individual reasoning steps may appear correct while the output as a whole violates budgets, fails test cases, or contradicts earlier deductions.<n>We propose a learned quality scorer with deterministic analytical constraint penalties for verifying structured LLM outputs.
  arXiv  Detail & Related papers  (2026-05-15T17:08:27Z)
• Pause and Reflect: Conformal Aggregation for Chain-of-Thought Reasoning [8.024041325202612]
  Chain-of-thought (CoT) reasoning with self-consistency improves performance by aggregating multiple sampled reasoning paths.<n>We introduce a conformal procedure for CoT reasoning that directly addresses aggregation uncertainty.<n>Our approach replaces majority voting with weighted score aggregation over reasoning paths and calibrates an abstention rule using conformal risk control.
  arXiv  Detail & Related papers  (2026-05-13T20:33:59Z)
• Multi-Dimensional Behavioral Evaluation of Agentic Stock Prediction Systems Using Large Language Model Judges with Closed-Loop Reinforcement Learning Feedback [1.2362187555287152]
  Forecast evaluation in finance has relied on aggregate accuracy metrics and predictive-accuracy tests built on point-forecast errors.<n>We propose a behavioral forecast-evaluation methodology that complements accuracy tests by assessing the intermediate decision process itself.
  arXiv  Detail & Related papers  (2026-05-07T06:31:34Z)
• When Correct Isn't Usable: Improving Structured Output Reliability in Small Language Models [2.064923532131528]
  Deployed language models must produce outputs that are both correct and format-compliant.<n>We study this structured-output reliability gap using two mathematical benchmarks -- GSM8K and MATH.<n>We develop AloLab, an iterative system-prompt requiring only black-box API access to the target model.
  arXiv  Detail & Related papers  (2026-05-04T09:07:44Z)
• HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam [63.84155758655084]
  Humanity's Last Exam (HLE) has become a widely used benchmark for evaluating frontier large language models.<n>We introduce HLE-Verified, a verified and revised version of HLE with a transparent verification protocol and fine-grained error taxonomy.<n>We evaluate seven state-of-the-art language models on HLE and HLE-Verified, observing an average absolute accuracy gain of 7--10 percentage points.
  arXiv  Detail & Related papers  (2026-02-15T02:50:15Z)
• Shrinking the Generation-Verification Gap with Weak Verifiers [42.538675831498715]
  Verifiers can improve language model capabilities by scoring and ranking responses from generated candidates.<n>Weaver is a framework for designing a strong verifier by combining multiple weak, imperfect verifiers.
  arXiv  Detail & Related papers  (2025-06-22T23:38:15Z)
• Latent Veracity Inference for Identifying Errors in Stepwise Reasoning [78.29317733206643]
  We introduce Veracity Search (VS), a discrete search algorithm over veracity assignments.<n>It performs otherwise intractable inference in the posterior distribution over latent veracity values.<n>It generalizes VS, enabling accurate zero-shot veracity inference in novel contexts.
  arXiv  Detail & Related papers  (2025-05-17T04:16:36Z)
• Detecting Errors and Estimating Accuracy on Unlabeled Data with Self-training Ensembles [38.23896575179384]
  We propose a principled and practically effective framework that simultaneously addresses the two tasks. One instantiation reduces the estimation error for unsupervised accuracy estimation by at least 70% and improves the F1 score for error detection by at least 4.7%. On iWildCam, one instantiation reduces the estimation error for unsupervised accuracy estimation by at least 70% and improves the F1 score for error detection by at least 4.7%.
  arXiv  Detail & Related papers  (2021-06-29T21:32:51Z)

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.