# ReFACT: A Benchmark for Scientific Confabulation Detection with Positional Error Annotations

arXiv: https://arxiv.org/abs/2509.25868 · HTML: https://mlsanity.com/paper/2509.25868/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 9 of its 27 references that are in the map.
- First posted on arXiv 30 Sep 2025.

## Builds on (most cited first, 9 of 9)

- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation](https://mlsanity.com/paper/2305.14251/) (2023)
- [Factuality Enhanced Language Models for Open-Ended Text Generation](https://mlsanity.com/paper/2206.04624/) (2022)
- [Fine-grained Hallucination Detection and Editing for Language Models](https://mlsanity.com/paper/2401.06855/) (2024)
- [Confabulation: The Surprising Value of Large Language Model Hallucinations](https://mlsanity.com/paper/2406.04175/) (2024)
- [I Don't Know: Explicit Modeling of Uncertainty with an [IDK] Token](https://mlsanity.com/paper/2412.06676/) (2024)

## Built on by (most cited first, 4 of 4)

- [Does Finetuning with Scientific Data Increase Hallucinations? A Multi-domain Factuality Evaluation of LLMs](https://mlsanity.com/paper/2606.21359/) (2026)
- [Averaging Bias: Human Faithfulness Annotations are not Locally Faithful](https://mlsanity.com/paper/2608.00205/) (2026)
- [Auditing the Synthetic Memoir: Measuring Scene-Level Confabulation in LLM-Generated Autobiography Against the Documented Record of the Life It Describes](https://mlsanity.com/paper/2608.23640/) (2026)
- [PetQA: Benchmarking Veterinary Knowledge and Clinical Reasoning](https://mlsanity.com/paper/2609.04598/) (2026)

## Abstract (arXiv)

The mechanisms underlying scientific confabulation in Large Language Models (LLMs) remain poorly understood. We introduce ReFACT (Reddit False And Correct Texts), a benchmark of 1,001 expert-annotated question-answer pairs with span-level error annotations derived from Reddit's r/AskScience. Evaluating 9 state-of-the-art LLMs reveals two critical limitations. First, models exhibit a dominant "salient distractor" failure mode: 61% of incorrect span predictions are semantically unrelated to actual errors. Crucially, this pattern persists across all model scales (1B to 70B), indicating a fundamental semantic grounding deficit that scaling alone fails to resolve. Second, we find that comparative judgment is paradoxically harder than independent detection, even GPT-4o's F1 score drops from 0.67 to 0.53 when comparing answers side-by-side. These findings directly challenge the reliability of LLM-as-Judge paradigms for scientific factuality. Code and data are released at https://github.com/ddz5431/ReFACT.