IdeaAMBIG: Benchmarking Implementation-Critical Gaps in Research-Idea Specifications

Moving from an abstract research idea to faithful code often falters on omitted technical details. IdeaAMBIG benchmarks this codification readiness across 660 evidence-grounded gaps derived from GitHub issues and reproducibility artifacts. Across 13 evaluated language models, the strongest system achieved only a 9.6% defect recovery rate when locating omissions unaided. While models handle clarification well once a defect is explicitly marked, detecting what an author left unsaid remains their primary limitation.