SA-Bench: Evaluating Semantic Alignment in LLM-Based Paper Reproduction

• URL: http://arxiv.org/abs/2608.24252v1
• Date: Tue, 25 Aug 2026 08:48:48 GMT
• Title: SA-Bench: Evaluating Semantic Alignment in LLM-Based Paper Reproduction
• Abstract summary: LLM agents can generate reproduction code, yet often produce scientifically unfaithful implementations.<n>We define this failure mode as semantic drift, where generated code silently diverges from the paper's specifications.<n>We introduce SemanticAlign-Bench, a diagnostic benchmark covering 30 papers from ICLR, ICML and NeurIPS 2025.
• Score: 12.71956248550879
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: LLM agents can generate paper reproduction code, yet often produce scientifically unfaithful implementations. We define this failure mode as semantic drift, where generated code silently diverges from the paper's specifications. We introduce SemanticAlign-Bench(SA-Bench), a diagnostic benchmark covering 30 papers from ICLR, ICML and NeurIPS 2025. For each paper, we decompose its specifications into atomic and verifiable implementation claims, which we call Semantic Alignment Units (SAUs) and evaluate repositories along four diagnostic dimensions spanning numerical, methodological, protocol and ordering drift. In total, we construct 1,491 SAUs across five ML domains and evaluate 12 generator configurations (4 models $\times$ 3 scaffolds). Even the strongest configuration (Claude+PaperCoder) achieves a mean SAU score of only 0.301 out of 1.0, with an overall mean of 0.221 across 360 evaluations. A failure taxonomy reveals that agents attempt most requirements but implement them incorrectly, with implementation mismatch and stubs accounting for the majority of zero-scored claims. Our analysis further indicates that scaffolds optimized for executability provide limited leverage for scientific reproduction; narrowing the gap requires scaffolds that prioritize semantic specification verification. The benchmark, annotations and evaluation pipeline are publicly available.

Related papers

• IdeaAMBIG: Benchmarking Implementation-Critical Gaps in Research-Idea Specifications [52.570108663867046]
  A research idea may be novel, coherent, and scientifically plausible, yet its proposed method may remain insufficiently specified for faithful implementation.<n>We study the codification readiness of implementation-facing research-method specifications, defined by whether they provide sufficient methodological information for a competent implementer or coding agent to construct the intended method without unsupported assumptions.<n>We introduce IdeaAMBIG, a benchmark of 660 evidence-grounded instances: 163 real-world gaps from reports and GitHub issues, and 497 controlled synthetic gaps injected into codification-ready references.
  arXiv  Detail & Related papers  (2026-09-09T17:59:04Z)
• Structuring Semantic Embeddings for Principle Evaluation: A Prototype-Guided Contrastive Learning Approach [58.15427952222424]
  General-purpose text embeddings are widely deployed for such tasks, but broad semantic similarity can place semantically similar yet task-distinct examples.<n>We introduce Prototype-Guided Contrastive Learning (PGCL), a prototype-guided geometric regularization module built on top of frozen text embeddings.
  arXiv  Detail & Related papers  (2026-08-15T13:19:25Z)
• Benchmarking and Enhancing LLMs for Rule-Intensive Review of National Standard Documents [7.397429318828859]
  Large language models (LLMs) increasingly support complex professional tasks, yet their capabilities in rule-intensive document review remain insufficiently evaluated.<n>Existing benchmarks focus on domain knowledge and question answering, largely overlooking quality review for professional documents.<n>We introduce GB/T-Bench, the first benchmark for the structured review of national standard documents.
  arXiv  Detail & Related papers  (2026-08-06T17:27:23Z)
• 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)
• Humanizing Automatically Generated Unit Test Suites with LLM-Based Refactoring [18.727291516223115]
  Search-based test generation tools such as EvoSuite produce compilable and high-coverage unit tests at scale, but their suites are often hard to read and maintain.<n>We introduce TestHumanizer, a hybrid SBST+LLM approach that uses LLMs as controlled layers over compilable SBST suites to improve naming, structure, and developer-oriented clarity.<n>TestHumanizer reaches 88-98% compilation rates, close to EvoSuite's 100% baseline and clearly above direct generation.
  arXiv  Detail & Related papers  (2026-06-26T16:18:56Z)
• Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025) [0.0]
  Existing surveys treat code security, jailbreak taxonomy, or vulnerability detection as the central object and mention these corpora only in passing.<n>This paper treats the prompt datasets themselves as the unit of analysis.<n>The synthesis surfaces three recurring methodological gaps.
  arXiv  Detail & Related papers  (2026-05-19T18:05:51Z)
• Cross-Dialect Generalization Without Retraining: Benchmarks and Evaluation of Schema-Derived Constrained Decoding for MLIR [0.0]
  Multi-Level Intermediate Representation (MLIR) underlies modern ML compiler infrastructure (TensorFlow, JAX/StableHLO, PyTorch Inductor, IREE)<n>MLIR is also by design: new dialects ship per application domain, so a fine-tuned model per dialect does not scale.<n>We release four natural-language-to-MLIR benchmarks across three dialects.
  arXiv  Detail & Related papers  (2026-05-14T02:08:13Z)
• LiveFMBench: Unveiling the Power and Limits of Agentic Workflows in Specification Generation [75.05397479715576]
  Large language models (LLMs) and agents have shown promising progress, but their true capabilities and failure modes remain unclear.<n>We present the first systematic and contamination-aware study of LLM- and agent-based formal specification generation for C programs.
  arXiv  Detail & Related papers  (2026-05-02T11:31:33Z)
• SPARC: Scenario Planning and Reasoning for Automated C Unit Test Generation [1.0010193170880752]
  We introduce a neuro-symbolic, scenario-based framework that bridges the gap between high-level program intent and the rigid syntactic constraints of pointer arithmetic and manual memory management.<n>We evaluate it on 59 real-world and algorithmic subjects, where it outperforms the vanilla prompt generation baseline by 31.36% in line coverage, 26.01% in branch coverage, and 20.78% in mutation score, matching or exceeding the symbolic execution tool KLEE.
  arXiv  Detail & Related papers  (2026-02-18T18:09:03Z)
• SciReplicate-Bench: Benchmarking LLMs in Agent-driven Algorithmic Reproduction from Research Papers [16.80818230868491]
  This study evaluates large language models (LLMs) in generating code from algorithm descriptions in recent NLP papers.<n>To facilitate rigorous evaluation, we introduce SciReplicate-Bench, a benchmark of 100 tasks from 36 NLP papers published in 2024.<n>Building on SciReplicate-Bench, we propose Sci-Reproducer, a dual-agent framework consisting of a Paper Agent that interprets algorithmic concepts from literature and a Code Agent that retrieves dependencies from repositories and implements solutions.
  arXiv  Detail & Related papers  (2025-03-31T22:02:24Z)
• SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal [64.9938658716425]
  SORRY-Bench is a proposed benchmark for evaluating large language models' (LLMs) ability to recognize and reject unsafe user requests.<n>First, existing methods often use coarse-grained taxonomy of unsafe topics, and are over-representing some fine-grained topics.<n>Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations.
  arXiv  Detail & Related papers  (2024-06-20T17:56:07Z)
• Factcheck-Bench: Fine-Grained Evaluation Benchmark for Automatic Fact-checkers [121.53749383203792]
  We present a holistic end-to-end solution for annotating the factuality of large language models (LLMs)-generated responses. We construct an open-domain document-level factuality benchmark in three-level granularity: claim, sentence and document. Preliminary experiments show that FacTool, FactScore and Perplexity are struggling to identify false claims.
  arXiv  Detail & Related papers  (2023-11-15T14:41:57Z)

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.