Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Lineage-Grounded Idea Generation

The paper introduces IdeaGene-Bench (IG-Bench), a benchmark for tracing scientific idea lineage and generating proposals grounded in that lineage. Papers and proposals are decomposed into minimal, typed, evidence-grounded Idea Genome objects, while GenomeDiff records inheritance, mutation, loss, external import, and novel insertion. Across 10 scientific domains, the benchmark includes 1,961 golden lineage traces, 1,085 curated genome objects, and 920 pairwise differences. Its IG-Exam covers 42 task types, and IG-Arena uses a lineage-conditioned Population-Evolution Score. Tests with 14 LLM-based scientists reveal a compositional bottleneck: the best system achieves only 27.3% exact accuracy on lineage reasoning, while structured lineage context changes system rankings unevenly.