⌁ Awesome Auto Research Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

PEARL: Auditable Repair for Scientific Reasoning Graph Extraction

arXiv 2026 59.9 method, application

TLDR

PEARL repairs noisy LLM-generated scientific reasoning graphs into auditable, valid structures, achieving high accuracy on ARCHE benchmark.

Reasoning

The paper presents a clear, training-free method with strong empirical results (300/350 gate passes), but its scope is limited to reasoning graph extraction rather than full scientific discovery. The audit trail feature is a notable strength for transparency.

Read-first score

Read-first score 59.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 8%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Reproducibility 25%
85

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=artifact,code,github

Methodology quality 25%
60

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,benchmark

Topical relevance 42%
36.7

Uses existing LLM keyword relevance scores normalized to 0-100. AI scientist,automated scientific discovery,autonomous research agent,automated research,literature review agent,survey generation,automated experimentation,experiment design agent,AI for scientific research,paper writing agent,research automation,scientific discovery agent

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 115.

Keyword Scores

AI for scientific research
7
AI scientist
6
autonomous research agent
6
automated scientific discovery
5
research automation
5
scientific discovery agent
5
automated research
4
literature review agent
2
survey generation
1
automated experimentation
1
experiment design agent
1
paper writing agent
1

Deep Analysis

Innovations

  • PEARL framework: training-free repair of noisy LLM graph responses into auditable reasoning graphs
  • Materialization of graph content under a closed Peircean schema
  • Evidence-grounded judge feedback to repair rejected edge types, local inference steps, and terminal roots
  • Preservation of an audit trail for inspectability

Methodology

PEARL is a training-free framework that first materializes explicit graph content under a closed Peircean schema, then uses matched evidence-grounded judge feedback to repair rejected edge types, local inference steps, and terminal roots while preserving an audit trail. It is evaluated on five 70-paper model archives from the ARCHE benchmark for latent reasoning-chain extraction.

Key Results

PEARL raises strict gate passes from 0/350 for the LLM baseline to 300/350, with average REA improving from 0.339 to 0.906.

Resources

PaperCode

Tags