# Rethinking Atomic Decomposition for LLM Judges: A Prompt-Controlled Study of Reference-Grounded QA Evaluation

arXiv: https://arxiv.org/abs/2603.28005 · HTML: https://mlsanity.com/paper/2603.28005/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 5 of its 21 references that are in the map.
- First posted on arXiv 30 Mar 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 5 of 5)

- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Long-form factuality in large language models](https://mlsanity.com/paper/2403.18802/) (2024)
- [FaStfact: Faster, Stronger Long-Form Factuality Evaluations in LLMs](https://mlsanity.com/paper/2510.12839/) (2025)
- [Judging Against the Reference: Uncovering Knowledge-Driven Failures in LLM-Judges on QA Evaluation](https://mlsanity.com/paper/2601.07506/) (2026)
- [UFO: a Unified and Flexible Framework for Evaluating Factuality of Large Language Models](https://mlsanity.com/paper/2402.14690/) (2024)

## Built on by (most cited first, 5 of 5)

- [How Sensitive Are Safety Benchmarks to Judge Configuration Choices?](https://mlsanity.com/paper/2604.24074/) (2026)
- [PReMISE: Policy Rubrics as Measurement Specifications for LLM Judges](https://mlsanity.com/paper/2605.30803/) (2026)
- [From Holistic Evaluation to Structured Criteria: Rubrics Across the Evolving LLM Landscape](https://mlsanity.com/paper/2606.08625/) (2026)
- [VICT: Verifier-Instrumented Credit Tracing for Long-Horizon LLM Agent Reinforcement Learning](https://mlsanity.com/paper/2608.28128/) (2026)
- [Acmite: Mitigating Gender Bias in LLMs through Concept-Guided Mutual Information](https://mlsanity.com/paper/2610.01696/) (2026)

## Abstract (arXiv)

When an LLM judge only has to assign a three-way support label to a candidate answer given a reference, does asking it to decompose the answer into atomic claims help, and at what cost? We compare four single-call designs that share the judge model, the inputs, and the level of instruction detail: candidate-side atomic decomposition, a matched holistic rubric, reference-side decomposition that checks whether each reference claim is covered, and a bidirectional combination. Support labels are constructed from TruthfulQA, ASQA, and QAMPARI references (200 questions and 400 rows per dataset). All four designs are run with Opus-4.6, GPT-4.1, and Gemini Flash Lite; Sonnet-4.6 is added for the candidate-side and holistic designs. Candidate-side decomposition is weak where the label depends on completeness: the holistic rubric is more accurate on ASQA and QAMPARI for every judge while using fewer tokens. Reference-side decomposition is 12.5-21.3 points more accurate than holistic on ASQA for 28-29% more tokens, and stays near the holistic ceiling on QAMPARI at 63-69% more. On TruthfulQA misconceptions, candidate-side decomposition is competitive and significantly better for two judges. On a 60-row single-author subset whose labels are looser than strict reference completeness, candidate-side decomposition edges ahead of holistic for every judge, reversing the construction-label order. What a judge decomposes should follow what the label measures.