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

arXiv: https://arxiv.org/abs/2608.24252 · HTML: https://mlsanity.com/paper/2608.24252/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Web, search, and research agents, one of 16 directions in LLM agents and tool use.
- Builds on 21 of its 26 references that are in the map.
- First posted on arXiv 25 Aug 2026.

## Placed in

- LLM agents and tool use / [Web, search, and research agents](https://mlsanity.com/agents/web-search-and-research-agents/)

## Builds on (most cited first, 21 of 21)

- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge](https://mlsanity.com/paper/2411.16594/) (2024)
- [RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems](https://mlsanity.com/paper/2306.03091/) (2023)
- [PaperBench: Evaluating AI's Ability to Replicate AI Research](https://mlsanity.com/paper/2504.01848/) (2025)
- [MinerU: An Open-Source Solution for Precise Document Content Extraction](https://mlsanity.com/paper/2409.18839/) (2024)
- [CORE-Bench: Fostering the Credibility of Published Research Through a Computational Reproducibility Agent Benchmark](https://mlsanity.com/paper/2409.11363/) (2024)
- [Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning](https://mlsanity.com/paper/2504.17192/) (2025)
- [NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents](https://mlsanity.com/paper/2512.12730/) (2025)
- [SciReplicate-Bench: Benchmarking LLMs in Agent-driven Algorithmic Reproduction from Research Papers](https://mlsanity.com/paper/2504.00255/) (2025)
- [SlopCodeBench: Benchmarking How Coding Agents Degrade Over Long-Horizon Iterative Tasks](https://mlsanity.com/paper/2603.24755/) (2026)
- [LMR-BENCH: Evaluating LLM Agent's Ability on Reproducing Language Modeling Research](https://mlsanity.com/paper/2506.17335/) (2025)
- [AutoP2C: An LLM-Based Agent Framework for Code Repository Generation from Multimodal Content in Academic Papers](https://mlsanity.com/paper/2504.20115/) (2025)
- [Know When To Stop: A Study of Semantic Drift in Text Generation](https://mlsanity.com/paper/2404.05411/) (2024)
- [Reflective Paper-to-Code Reproduction Enabled by Fine-Grained Verification](https://mlsanity.com/paper/2508.16671/) (2025)
- [SciCoQA: Quality Assurance for Scientific Paper--Code Alignment](https://mlsanity.com/paper/2601.12910/) (2026)
- [Asymmetric Goal Drift in Coding Agents Under Value Conflict](https://mlsanity.com/paper/2603.03456/) (2026)
- [Enhancing Automated Paper Reproduction via Prompt-Free Collaborative Agents](https://mlsanity.com/paper/2512.02812/) (2025)

## Built on by (most cited first, 1 of 1)

- [RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers?](https://mlsanity.com/paper/2609.28850/) (2026)

## Abstract (arXiv)

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 × 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.