# Share the Judge, Learn the Deferral: Where Specialization Helps LLM Evaluation

arXiv: https://arxiv.org/abs/2607.27984 · HTML: https://mlsanity.com/paper/2607.27984/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use.
- Builds on 19 of its 26 references that are in the map.
- First posted on arXiv 30 Jul 2026.

## Placed in

- LLM agents and tool use / [General LLM agent frameworks and benchmarks](https://mlsanity.com/agents/general-llm-agent-frameworks-and-benchmarks/)

## Builds on (most cited first, 19 of 19)

- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://mlsanity.com/paper/2411.15124/) (2024)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Prometheus: Inducing Fine-grained Evaluation Capability in Language Models](https://mlsanity.com/paper/2310.08491/) (2023)
- [Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Models](https://mlsanity.com/paper/2405.01535/) (2024)
- [HealthBench: Evaluating Large Language Models Towards Improved Human Health](https://mlsanity.com/paper/2505.08775/) (2025)
- [Consistent Estimators for Learning to Defer to an Expert](https://mlsanity.com/paper/2006.01862/) (2020)
- [Predict Responsibly: Improving Fairness and Accuracy by Learning to Defer](https://mlsanity.com/paper/1711.06664/) (2017)
- [Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy](https://mlsanity.com/paper/2507.01352/) (2025)
- [RewardBench 2: Advancing Reward Model Evaluation](https://mlsanity.com/paper/2506.01937/) (2025)
- [MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA-based Mixture of Experts](https://mlsanity.com/paper/2404.15159/) (2024)
- [Trust or Escalate: LLM Judges with Provable Guarantees for Human Agreement](https://mlsanity.com/paper/2407.18370/) (2024)
- [Mixture-of-LoRAs: An Efficient Multitask Tuning for Large Language Models](https://mlsanity.com/paper/2403.03432/) (2024)
- [SCOPE: Selective Conformal Optimized Pairwise LLM Judging](https://mlsanity.com/paper/2602.13110/) (2026)
- [Ask a Strong LLM Judge when Your Reward Model is Uncertain](https://mlsanity.com/paper/2510.20369/) (2025)
- [AdaRubric: Task-Adaptive Rubrics for Reliable LLM Agent Evaluation and Reward Learning](https://mlsanity.com/paper/2603.21362/) (2026)
- [Gatekeeper: Improving Model Cascades Through Confidence Tuning](https://mlsanity.com/paper/2502.19335/) (2025)
- [Quantitative LLM Judges](https://mlsanity.com/paper/2506.02945/) (2025)
- [SLMJury: Can Small Language Models Judge as Well as Large Ones?](https://mlsanity.com/paper/2606.07810/) (2026)

## Abstract (arXiv)

Agentic systems generate outputs faster than human review. We contrast two LLM evaluator specialization strategies: specialized judge weights, or rule-based deferral policies for safe judgment acceptance. On 99,952 rubric-conditioned samples, correct rubrics improve accuracy by 2.11 points, while incorrect rubrics reduce performance by 2.66 points. Splitting training data across eight criterion-specific LoRA experts lowers accuracy by 10.05 points and reduces 5% error-bound coverage from 24.44% to 5.43%. This loss is independent of model size and training settings, with most performance recoverable by warm-starting experts from a unified judge. Sweeping data budgets confirms scratch expert specialization yields no empirical gains. Warm-started splits appear competitive with unified models, yet under limited data, unified training outperforms split training, with specialization beneficial only after unified training plateaus. Results hold on HealthBench, where physician rubrics improve accuracy while flawed rubrics degrade performance. Unlike weight specialization, deferral policies enable efficient evaluation. On RewardBench 2, lightweight deferral heads form a 0.6B-4B-8B reward cascade with no core scoring modification. Across 20 splits, the cascade achieves 89.40% accuracy versus 84.75% for a standalone 8B judge at 41.5% compute, satisfying 95% risk constraints. Margin-based deferral matches accuracy at far higher compute cost. The design generalizes across models, improving Tulu-3-8B and Skywork-8B performance with a lightweight DeBERTa frontend. We derive simple, robust evaluator design rules: unify judgment training or warm-start split models, and use audited deferral cascades for low-cost, reliable LLM evaluation.