ComplexConstraints and Beyond: Expert Rubrics for RLVR
2026-06-08 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors point out that as large language models (LLMs) get better, the usual ways to test them don’t keep up because real tasks are complicated and hard to check with simple tests. They suggest using detailed rubrics made by experts to better evaluate and train these models. They created a dataset called ComplexConstraints with many small checklist items for each task and showed that training models using these rubrics improves their instruction-following skills and helps them perform better on new tasks too. Overall, the authors show that expert-made rubrics are a useful tool both for measuring and improving LLM abilities.
large language modelsinstruction followingrubric-based evaluationComplexConstraintsatomic rubric criteriareinforcement learningmodel trainingbenchmarkingout-of-distribution generalizationagentic tasks
Authors
Sushant Mehta, Liudas Panavas, Edwin Chen
Abstract
As LLM capabilities advance rapidly, the evaluation methods used to assess them increasingly lag behind. Traditional benchmarks relied on programmatic verification of narrow, surface-level constraints, but real-world instruction following and agentic tasks demand assessment of nuanced, context-dependent behaviors that resist simple scripted checks. We present a systematic analysis of expert-curated rubric-based evaluation as an alternative paradigm, drawing on empirical evidence from two domains: complex instruction following and enterprise agentic tasks. We first articulate five design principles for constructing high-quality rubrics, including Maximum Viable Atomicity, intent-aware criterion design, and iterative LLM-judge calibration. To validate these principles, we introduce ComplexConstraints, a new expert-curated instruction-following dataset in which each prompt is paired with 10-40 atomic rubric criteria. We demonstrate that these expert rubrics are not only better evaluation instruments but also highly effective training signals: training on approximately 1,000 ComplexConstraints examples yields +15.5% improvement for a 4B-parameter model and +12.2% for a 235B-parameter model on instruction following, while single-epoch RL training on a rubric-graded enterprise environment produces gains that transfer to out-of-distribution benchmarks the model was never trained on (+4.5% BFCL, +7.4% Tau2-Bench, +6.8% Tool-Decathlon). Our findings establish that expert-authored rubrics improve both the measurement and the development of frontier LLM capabilities, serving as effective evaluation and RL training signals.