# Beyond Document Grounding: Span-Level Hallucination Detection over Code, Tool Output, and Documents

arXiv: https://arxiv.org/abs/2607.00895 · HTML: https://mlsanity.com/paper/2607.00895/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Hallucination detection in RAG, one of 15 directions in Retrieval-augmented generation.
- Builds on 16 of its 26 references that are in the map.
- First posted on arXiv 1 Jul 2026.

## Placed in

- Retrieval-augmented generation / [Hallucination detection in RAG](https://mlsanity.com/rag/hallucination-detection-in-rag/)

## Builds on (most cited first, 16 of 16)

- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [RAGTruth: A Hallucination Corpus for Developing Trustworthy Retrieval-Augmented Language Models](https://mlsanity.com/paper/2401.00396/) (2023)
- [MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documents](https://mlsanity.com/paper/2404.10774/) (2024)
- [Fine-grained Hallucination Detection and Editing for Language Models](https://mlsanity.com/paper/2401.06855/) (2024)
- [Lynx: An Open Source Hallucination Evaluation Model](https://mlsanity.com/paper/2407.08488/) (2024)
- [CodeHalu: Investigating Code Hallucinations in LLMs via Execution-based Verification](https://mlsanity.com/paper/2405.00253/) (2024)
- [LettuceDetect: A Hallucination Detection Framework for RAG Applications](https://mlsanity.com/paper/2502.17125/) (2025)
- [SemEval-2025 Task 3: Mu-SHROOM, the Multilingual Shared Task on Hallucinations and Related Observable Overgeneration Mistakes](https://mlsanity.com/paper/2504.11975/) (2025)
- [CodeMirage: Hallucinations in Code Generated by Large Language Models](https://mlsanity.com/paper/2408.08333/) (2024)
- [When Models Lie, We Learn: Multilingual Span-Level Hallucination Detection with PsiloQA](https://mlsanity.com/paper/2510.04849/) (2025)
- [Collu-Bench: A Benchmark for Predicting Language Model Hallucinations in Code](https://mlsanity.com/paper/2410.09997/) (2024)
- [Learning to Reason for Hallucination Span Detection](https://mlsanity.com/paper/2510.02173/) (2025)
- [Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks](https://mlsanity.com/paper/2605.07024/) (2026)
- [Squeez: Task-Conditioned Tool-Output Pruning for Coding Agents](https://mlsanity.com/paper/2604.04979/) (2026)
- [ACL-Verbatim: hallucination-free question answering for research](https://mlsanity.com/paper/2605.21102/) (2026)

## Built on by (most cited first, 3 of 3)

- [Uncertainty Quantification for LLM Agents: A Taxonomy, an Evaluation Protocol, and an Empirical Study](https://mlsanity.com/paper/2609.07395/) (2026)
- [Grounding Probes: Generator-Independent Hallucination Detection from Observer Model Hidden States](https://mlsanity.com/paper/2610.04642/) (2026)
- [Hallucination Across the Reasoning Lifecycle: Interface Visibility, Causal Evidence, and Release Control in Large Reasoning Models](https://mlsanity.com/paper/2610.05472/) (2026)

## Abstract (arXiv)

Hallucination detection for retrieval-augmented generation (RAG) is usually evaluated on natural-language document evidence. However, grounded generation systems increasingly rely on structured inputs: source code, developer-tool output, markdown documents, tables, and repository metadata. We introduce a unified benchmark for span-level hallucination detection over code, tool output, structured documents, and existing natural-language RAG datasets. The benchmark is built by starting from grounded correct answers, injecting localized hallucinations with exact character labels, and validating the code test split with evidence-based review. Our fine-tuned Qwen3.5-2B detector reaches 0.689 span-F1 on the unified test set and 0.60 on the code-agent source, where it substantially outperforms LettuceDetect-large (0.17) and the strongest zero-shot LLM judges we evaluated (at most 0.22). The same model remains competitive on established natural-language benchmarks, with 81.8 RAGTruth example-F1 and 0.724 English PsiloQA IoU.