DraftNEPABench: A Benchmark for Drafting NEPA Document Sections with Coding Agents

Dataset Description

DraftNEPABench is a novel benchmark designed to evaluate the capabilities of Large Language Models (LLMs) and LLM-based agents in drafting sections of Environmental Impact Statements (EIS). This benchmark is curated by subject matter experts (SMEs) to ensure that the tasks reflect realistic, domain-relevant drafting challenges encountered in environmental planning and compliance processes.

Dataset Structure

Each task is organized in the following folder structure:

task-<N>/
├── References/
│   └── reference.pdf
├── task.docx
├── ground_truth.docx  (not included for held-out tasks)
└── rubric.docx

Files in Each Task

• task.docx: Task description and instructions
• ground_truth.docx: The authoritative expected answer/content (not provided for held-out evaluation tasks)
• rubric.docx: Grading criteria for evaluation
• References/: Folder containing reference documents (typically PDFs)

All three .docx files are required for standard tasks. Held-out tasks do not include ground_truth.docx.

Metadata

The dataset includes a metadata.json file that contains summary information for all tasks, including:

• agency: The government agency responsible for the EIS
• eis_title: Title of the Environmental Impact Statement
• eis_section: The specific section type (e.g., "AE_Air Quality", "EC_Socioeconomics", "AE/EC_Biological Resources")
• difficulty: Task difficulty rating (1.0-4.0)
• total_ref: Number of reference documents provided
• ground_truth_length: Length of the ground truth document in words
• holdout_set: Boolean indicating whether this is a held-out evaluation task (true/false)
• task_number: Task identifier

The held-out set contains tasks where ground truth documents are not provided, allowing for unbiased evaluation of model performance.

Using This Dataset

To use this dataset with the DraftNEPABench benchmark:

1. Download the dataset from HuggingFace
2. Clone the benchmark repository: https://github.com/pnnl/PermitAI/tree/master/DraftNEPABench
3. Place the tasks under run-and-grade/tasks/task-<N> in the benchmark repository

The benchmark provides:

• Scripts to generate EIS section drafts using coding agents (Codex, Claude, Gemini) and vanilla RAG
• Scripts to grade generated drafts using LLM-judges (GPT, Sonnet, Gemini)
• Analysis notebooks for result aggregation and visualization

Citation

Please cite this work as follows:

@misc{acharya_draftnepabench,
  title        = {DraftNEPABench: A Benchmark for Drafting NEPA Document Sections with Coding Agents},
  author       = {Acharya, Anurag and Lakha, Bishal and Meyur, Rounak and Nuttall, Rohan and Chaturvedi, Sarthak and Halappanavar, Anika and Hare, Leah and Zeng, Lin and Parker, Mike and Munikoti, Sai and Horawalavithana, Sameera},
  note         = {Manuscript},
  year         = {2026}
}

License

Copyright Battelle Memorial Institute 2026

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Disclaimer

This material was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor the United States Department of Energy, nor Battelle, nor any of their employees, nor any jurisdiction or organization that has cooperated in the development of these materials, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness or any information, apparatus, product, software, or process disclosed, or represents that its use would not infringe privately owned rights.

             PACIFIC NORTHWEST NATIONAL LABORATORY
                          operated by
                            BATTELLE
                            for the
               UNITED STATES DEPARTMENT OF ENERGY
                under Contract DE-AC05-76RL01830