LexReward: A Taxonomy-Driven Reward Framework for Legal Language Models

Reward models are central to preference alignment for legal tasks, yet existing reward model benchmarks are still primarily centered on general-domain preferences. In this work, we introduce LegalRewardBench, the first human-verified benchmark for evaluating reward models in the legal domain. The benchmark organizes legal preference evaluation into five capability levels: knowledge, understanding, analysis, decision, and writing, and preference examples are constructed through rubric-guided automatic judging combined with human verification. We evaluate three families of judges, covering both general and fine-tuned models, and generative and discriminative architectures. Experimental results show that strong general-purpose generative models are currently the most robust legal reward models, while existing fine-tuned reward models transfer less reliably to legal preference evaluation. We also conduct detailed comparisons of input organizations and training strategies.

Large language models are increasingly used in high-stakes domains such as law, where systems must ground their reasoning in retrieved evidence and abstain when that evidence is insufficient. However, existing reward models are largely optimised for general preferences rather than contextual grounding, limiting their ability to evaluate these behaviours in retrieval-augmented generation (RAG) settings. We introduce a framework for transforming existing legal QA datasets into contextual preference data and use it to construct LegalRewardBench (LRB), a benchmark for evaluating grounded legal generation under noisy and insufficient retrieval conditions. Across general and legal contextual evaluation, we find that contextual DPO improves grounded evaluation, but performance is sensitive to preference-data construction. Length-balanced augmentation substantially improves grounded legal evaluation, with the strongest configuration combining length-balanced legal and general contextual preference data and improving performance by up to $\mathbf{+25.6}$pp over baseline. We further find evidence of cross-jurisdiction transfer: models contextually refined primarily on Victorian criminal-law data improve grounded evaluation on external US legal benchmarks, including a $\mathbf{+16.2}$pp improvement on Housing Statute QA. Together, these results provide a reproducible foundation for constructing and evaluating grounded legal reward models in retrieval-augmented settings.

What happens when a legal AI model learns to look like a lawyer instead of reasoning like one? We fine tune Qwen3-8B with Group Relative Policy Optimisation (GRPO) against a proxy built from three surface features: citation count, legalese density, and response length. The model does not learn to reason more effectively. It learns to withhold commitment. Across 16 yes or no legal reasoning tasks from LegalBench (N=320), overall accuracy collapses from 0.500 (chance) to 0.072 (McNemar p < 10^-36), driven entirely by the rate of properly formatted answers falling from 0.900 to 0.109. The model stops committing to answers. Yet when it does commit, accuracy rises from 0.556 to 0.657, showing that the collapse is not a failure of capability but a strategic response: the model has learned that verbose responses packed with citations but empty of a direct answer score higher than terse correct ones. We term this the Saul Goodman effect, a policy that becomes maximally lawyerly while becoming maximally noncommittal, and prove formally that it is the optimal response to any surface feature proxy that attaches no penalty to abstention. We further show that 89.3% of citations produced after training are structurally implausible hallucinations, many of them subtly corrupted names of real landmark cases, constructed in effect to survive a casual read and fail under scrutiny. To detect this failure mode before deployment, we introduce three diagnostic tools: the Confidence Theater Score (CTS), the Citation Plausibility Rate (CPR), and the Regret Gap (RG). In a domain where a confidently wrong answer can constitute malpractice, the broader lesson is direct: a reward function that measures how legal a response looks will produce a model that is maximally photogenic and minimally useful.