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

Hugging Face Daily Papers Papers

Summary

Introduces IG-Bench, a benchmark for scientific lineage reasoning and idea generation, representing scientific works as Idea Genome objects. Evaluates 14 LLM-based scientists, finding a compositional bottleneck with the strongest system achieving only 27.3% exact accuracy on lineage reasoning.

Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks still say little about whether AI systems can follow this inheritance structure. We present IdeaGene-Bench (IG-Bench), a benchmark for scientific lineage reasoning and lineage-grounded idea generation. IG-Bench is organized around the IdeaGene framework: each paper or proposal is represented as a set of minimal, typed, evidence-grounded Idea Genome objects, and a GenomeDiff aligns these objects to record inheritance, mutation, loss, external import, and novel insertion under six operational evolutionary dynamics. The benchmark contains 1,961 golden lineage traces, 1,085 curated Idea Genome objects, and 920 pairwise GenomeDiff records across 10 scientific domains. It supports two evaluations. IG-Exam (42 task types, 1,029 instances) tests closed-form lineage reasoning across Idea Genome abstraction, inheritance tracing, evolutionary reasoning, and lineage verification. IG-Arena evaluates generation with a lineage-conditioned Population-Evolution Score(PES), asking whether a proposal can be inserted as a coherent descendant of a given lineage population: it should inherit the right Idea Genome objects, vary meaningfully from nearby work, and offer selection value for future research. Experiments on 14 LLM-based scientists expose a compositional bottleneck. The strongest system reaches only 27.3% exact accuracy on lineage reasoning, and structured lineage context reshuffles system rankings rather than helping every participant uniformly.
Original Article
View Cached Full Text

Cached at: 07/10/26, 06:17 AM

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

Source: https://huggingface.co/papers/2607.08758 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

A benchmark for scientific lineage reasoning and idea generation is introduced, organizing scientific works as genetic-like Idea Genome objects and evaluating both reasoning and generation capabilities.

Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks still say little about whether AI systems can follow this inheritance structure. We present IdeaGene-Bench (IG-Bench), a benchmark for scientificlineage reasoningand lineage-groundedidea generation.IG-Benchis organized around theIdeaGene framework: each paper or proposal is represented as a set of minimal, typed, evidence-groundedIdea Genome objects, and aGenomeDiffaligns these objects to record inheritance, mutation, loss, external import, and novel insertion under six operationalevolutionary dynamics. The benchmark contains 1,961 golden lineage traces, 1,085 curatedIdea Genome objects, and 920 pairwiseGenomeDiffrecords across 10 scientific domains. It supports two evaluations.IG-Exam(42 task types, 1,029 instances) tests closed-formlineage reasoningacross Idea Genome abstraction, inheritance tracing, evolutionary reasoning, and lineage verification.IG-Arenaevaluates generation with a lineage-conditionedPopulation-Evolution Score(PES), asking whether a proposal can be inserted as a coherent descendant of a given lineage population: it should inherit the rightIdea Genome objects, vary meaningfully from nearby work, and offer selection value for future research. Experiments on 14LLM-based scientistsexpose a compositional bottleneck. The strongest system reaches only 27.3% exact accuracy onlineage reasoning, and structured lineage context reshuffles system rankings rather than helping every participant uniformly.

View arXiv pageView PDFProject pageGitHub2Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.08758 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.08758 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.08758 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

GENEB: Why Genomic Models Are Hard to Compare

arXiv cs.CL

GENEB is a large-scale diagnostic benchmark that evaluates 40 genomic foundation models across 100 tasks in 13 functional categories under a unified probing protocol, exposing that aggregate leaderboards are unstable and that architectural alignment often outweighs model scale. The work addresses the fragmented evaluation landscape in genomic machine learning, analogous to what MTEB did for NLP.

Measuring the Gap Between Human and LLM Research Ideas

Hugging Face Daily Papers

This research paper introduces a framework to measure the distributional gap between human-generated and LLM-generated research ideas, finding that LLM ideas are concentrated around specific opportunity patterns and synthesis methods, while human ideas are more diverse.