Reproduce “Joint Constraint Aggregation for Qualifier-Sensitive Retrieval-Augmented QA” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jennifer Hsia, Zachary C. Lipton, Aarti Singh

Joint Constraint Aggregation (JCA) selects among answer candidates by deriving shared requirements from a question, checking each candidate against retrieved evidence, and jointly comparing the resulting reports. On the 294-instance AmbigQA hard slice with Llama4-Scout-17B, it achieves 92.5% accuracy, versus 82.3% for chain-of-thought. JCA ranks highest across nine hard-slice settings, conditional on the gold answer being a candidate.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.