README.md

metadata

dataset_info:
  features:
    - name: paper_id
      dtype: string
    - name: title
      dtype: string
    - name: abstract
      dtype: string
    - name: publication_date
      dtype: string
    - name: paper_link
      dtype: string
    - name: code_available
      dtype: bool
    - name: code_link
      dtype: string
    - name: source
      dtype: string
    - name: dataset
      list:
        - name: kind
          dtype: string
        - name: dataset_name
          dtype: string
        - name: data_instructions
          dtype: string
    - name: execution_requirements
      struct:
        - name: code_language
          dtype: string
        - name: dependencies
          sequence: string
        - name: needs_gpu
          dtype: bool
    - name: other_instructions
      dtype: string
    - name: full_text
      dtype: string
  splits:
    - name: papers
      num_bytes: TBD
      num_examples: TBD
    - name: tasks
      num_bytes: TBD
      num_examples: TBD
  download_size: TBD
  dataset_size: TBD
configs:
  - config_name: default
    data_files:
      - split: papers
        path: papers.jsonl
      - split: tasks
        path: tasks.jsonl

ReplicationBench

arXiv: ReplicationBench: Can AI Agents Replicate Astrophysics Research Papers?

GitHub: https://github.com/Christine8888/replicationbench-release

Dataset Description

The ReplicationBench dataset contains 111 astrophysics research replication tasks, spanning complete replications of 20 research papers. The dataset includes:

• Original and masked manuscript text
• Metadata (title, abstract, publication info, etc.)
• Pointers to datasets and dataset access instructions
• Additional specifications from the authors
• Execution requirements
• Detailed descriptions and grading guidelines for each task

Usage

from datasets import load_dataset

# Load papers
papers_ds = load_dataset("ChristineYe8/replicationbench", split="papers")

# Load tasks
tasks_ds = load_dataset("ChristineYe8/replicationbench", split="tasks")

You can load the dataset from HuggingFace into native ReplicationBench format using this script. However, if using RB's native formats, we recommend using the native data loading instead, described here.

Citation

If you use ReplicationBench in your research, please cite:

@misc{ye2025replicationbenchaiagentsreplicate,
      title={ReplicationBench: Can AI Agents Replicate Astrophysics Research Papers?},
      author={Christine Ye and Sihan Yuan and Suchetha Cooray and Steven Dillmann and Ian L. V. Roque and Dalya Baron and Philipp Frank and Sergio Martin-Alvarez and Nolan Koblischke and Frank J Qu and Diyi Yang and Risa Wechsler and Ioana Ciuca},
      year={2025},
      eprint={2510.24591},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.24591},
}

License

MIT License