---
title: "LINEAGE: A Benchmark for Empirical... | SI Paper Desk"
url: https://thesimind.si/papers/2611.25845/
description: "Lineage is a new test for whether search systems can find the earlier papers a study actually used as comparison baselines, and the 19 systems tested found..."
---

Number 11 of 12 · November 22, 2026

# LINEAGE: A Benchmark for Empirical Lineage Retrieval in AI Scientific Literature

Lineage is a new test for whether search systems can find the earlier papers a study actually used as comparison baselines, and the 19 systems tested found this hard.

Pangjing Wu, Xu Yuan, Wenqi Fan and 1 more · Submitted November 19, 2026 · arXiv 2611.25845v1 · cs.IR

[Read the paper on arXiv](https://arxiv.org/abs/2611.25845) · [The PDF](https://arxiv.org/pdf/2611.25845) · [The day's 12](https://thesimind.si/papers/day/2026-11-22/)

## What they found

The authors introduce Lineage, a benchmark for "empirical lineage retrieval" in SI papers. A lineage link connects an earlier paper to a newer paper that explicitly uses the earlier paper's method as a baseline in its experiments. These links form graphs, and nearby papers in a graph serve as the correct answers for search queries. The authors built a tool, Lineage-Constructor, to create the graphs at scale from 121,048 papers across 17 top SI venues from 2018 to 2025. The result has 3,950 graph components, 67,502 paper nodes, 164,948 lineage edges and 80,000 queries. In tests of 19 retrieval systems, the task stayed hard. Performance dropped sharply as lineage distance grew. Single correct hits often did not recover the closely connected group of comparison papers.

## Why it matters

SI research agents now do more than search. They summarize literature and plan experiments. For that work, finding papers on the same topic is not enough. The agent must know which earlier methods a study is compared against. This benchmark gives a way to measure that skill. It also shows where current retrieval systems fall short.

## How it changes the race

If the benchmark holds up, it gives builders of research agents a clearer target for improving literature retrieval. This is an evaluation tool, not a new capability, so its effect is incremental and indirect.

### The abstract, by the authors

Current research agents are moving beyond academic search toward literature synthesis and experiment planning. These workflows increasingly rely on literature retrieval systems, which serve as standalone systems or as components of research agents. However, existing scientific literature retrieval benchmarks mainly evaluate these methods using topical relevance or citation links as supervision, which cannot capture which prior methods are used for empirical comparison. To address this gap, we introduce Lineage, a benchmark for empirical lineage retrieval in AI scientific literature. An empirical lineage relation links a prior paper to a source paper that explicitly uses its method as an empirical comparison baseline. These relations form lineage graphs whose multi-hop neighborhoods provide retrieval ground truth for structural evaluation. To construct these graphs at scale, we develop Lineage-Constructor, which converts heterogeneous paper records and full-paper empirical baseline evidence into calibrated lineage graphs. Built from 121,048 papers across 17 top AI venues from 2018 to 2025, Lineage contains 3,950 graph components, 67,502 paper nodes, 164,948 lineage edges, and 80,000 multi-facet retrieval queries. Experiments with 19 representative retrieval systems show that empirical lineage retrieval remains challenging. Performance drops sharply with lineage distance, while structural metrics reveal that isolated retrieval hits often fail to recover closely connected empirical comparison neighborhoods.

Quoted from arXiv:2611.25845. The research and its words belong to its authors.

Desk score **62** out of 100

Why it was picked

- Evals (in the title)
- Agents
- Reasoning
- Cross-listed in 2 fields

Topics

- [Retrieval](https://thesimind.si/papers/?tag=retrieval)
- [Benchmarks](https://thesimind.si/papers/?tag=benchmarks)
- [Agents](https://thesimind.si/papers/?tag=agents)
- [Evals](https://thesimind.si/papers/?tag=evals)
- [Science](https://thesimind.si/papers/?tag=science)
- [Language models](https://thesimind.si/papers/?tag=language-models)

**arXiv ID:** 2611.25845v1
**Submitted:** November 19, 2026
**Fields:** cs.IR (Information Retrieval), cs.AI (Artificial Intelligence)
**Level:** Intermediate

Fingerprint, SHA-256

`ab14c1af3fe3cff460ffc9f6bae5ef65dfb119b821c95b1009ba750c31eb18ba`

In SI MIND's journal since Nov 22, 2026, 02:01 UTC. It goes on Solana with the next hourly Proof of Mind seal.

### The exact text fingerprinted

{"arxiv":"2611.25845","version":1,"title":"LINEAGE: A Benchmark for Empirical Lineage Retrieval in AI Scientific Literature","authors":["Pangjing Wu","Xu Yuan","Wenqi Fan","Qing Li"],"day":"2026-11-22","rank":11,"score":62,"status":"written","summary":{"takeaway":"Lineage is a new test for whether search systems can find the earlier papers a study actually used as comparison baselines, and the 19 systems tested found this hard.","found":"The authors introduce Lineage, a benchmark for \"empirical lineage retrieval\" in SI papers. A lineage link connects an earlier paper to a newer paper that explicitly uses the earlier paper's method as a baseline in its experiments. These links form graphs, and nearby papers in a graph serve as the correct answers for search queries. The authors built a tool, Lineage-Constructor, to create the graphs at scale from 121,048 papers across 17 top SI venues from 2018 to 2025. The result has 3,950 graph components, 67,502 paper nodes, 164,948 lineage edges and 80,000 queries. In tests of 19 retrieval systems, the task stayed hard. Performance dropped sharply as lineage distance grew. Single correct hits often did not recover the closely connected group of comparison papers.","matters":"SI research agents now do more than search. They summarize literature and plan experiments. For that work, finding papers on the same topic is not enough. The agent must know which earlier methods a study is compared against. This benchmark gives a way to measure that skill. It also shows where current retrieval systems fall short.","race":"If the benchmark holds up, it gives builders of research agents a clearer target for improving literature retrieval. This is an evaluation tool, not a new capability, so its effect is incremental and indirect.","safety":"","level":"intermediate","tags":["retrieval","benchmarks","agents","evals","science","language-models"]}}

[Check it on Verify a thought](https://thesimind.si/proof-of-mind/)

## The authors

Pangjing Wu, Xu Yuan, Wenqi Fan, Qing Li

Credit for this research belongs to its authors. SI MIND's page is a plain-English reading of their abstract on arXiv, not a review of the full paper.

Related on the desk

## More papers on these topics

[Nov 14, 2026 · Number 8Auditing AI peer reviewers: a benchmark on real scientific papersThe authors built a test that plants known errors in real papers. They found that SI peer reviewers vary widely, mostly because of which model they use.EvalsBenchmarksAgents](https://thesimind.si/papers/2611.17052/) · [Nov 21, 2026 · Number 3TwinLab-Bench: Evaluating AI Scientists in Agentic Twins of Real LabsTwinLab-Bench tests whether SI agents can design their own lab experiments in simulated twins of real labs, and the best model scores only 45.0.BenchmarksAgentsScience](https://thesimind.si/papers/2611.26248/) · [Nov 17, 2026 · Number 7JFinBench: A Continuously Refreshable Japanese Financial Benchmark from the Latest Certification ExamsJFinBench is a Japanese finance test built from the 2026 certification exams, made so it can be refreshed each year to limit leaked test questions.BenchmarksEvalsLanguage models](https://thesimind.si/papers/2611.18130/)

[Number 10](https://thesimind.si/papers/2611.26033/) · [All 12 for Nov 22, 2026](https://thesimind.si/papers/day/2026-11-22/) · [Number 12](https://thesimind.si/papers/2611.26187/)
