alea-institute/kl3m-data-ecfr

KL3M Data Project

Note: This page provides general information about the KL3M Data Project. Additional details specific to this dataset will be added in future updates. For complete information, please visit the GitHub repository or refer to the KL3M Data Project paper.

2026-09 rebuild

Why. The March 2025 upload of this dataset (revisions 849592b7/45a515d6; README updates through b48604bc) converted eCFR renderer HTML with a selection rule (kl3m-data d34c0bc) that favored an HTML-to-Markdown output which inserted spaces around inline elements and joined paragraphs with a single newline. In about 53% of documents, paragraph designators read ( a ) instead of (a) and paragraph breaks were lost, e.g. provisions: ( a ) Hydrogenated ... instead of provisions:\n\n(a) Hydrogenated .... The converter was fixed in alea-institute/kl3m-data#6, but the original renderer HTML was not archived, so those documents could not be reconverted.

Source. This revision is rebuilt from the official eCFR bulk XML (GPO/OFR ECFRBRWS format), a local snapshot of all titles downloaded on 2024-03-17. Each title carries its own amendment date (AMDDATE; most are Jan–Mar 2024). This is a different edition from the September 2024 renderer-HTML documents in earlier revisions, so section text can differ where regulations were amended in between.

Content. One document per CFR section (DIV8 TYPE="SECTION"), 229,868 documents (122.1M kl3m-004-128k-cased tokens):

• All titles in the snapshot, including Title 40 (Protection of Environment, 23,788 sections), which was absent from all earlier revisions. Title 35 is reserved and has no content.
• Appendices (DIV9 TYPE="APPENDIX") are excluded, as before. Part-level material (authority, source and part editorial notes) is not repeated into each section.
• Three DIV8 elements in Title 48 that are empty in the source XML are dropped.
• The XML is converted to Markdown with a dedicated XML-to-Markdown converter that follows the earlier conventions:
  • #### § 178.3610 Heading. section headings;
  • paragraphs separated by blank lines, with unspaced designators (a), (1);
  • *italic* / **bold** emphasis and [n] footnotes;
  • tables as GFM pipe tables;
  • images dropped.
• Text is then tokenized and scored with the kl3m-data pipeline code: postprocess_document, serialize_document and extract_tokens_and_metadata(include_score=True). The schema and dtypes are unchanged.
• The source's en dashes in section numbers and citations (e.g. § 60–4.6) are kept as-is; the renderer HTML used hyphens.

Identifiers. These documents did not come from the s3://data.kl3m.ai/documents/ecfr/... HTML objects, so identifiers use a new scheme: ecfr-bulk-xml/2024-03-17/title-<T>/<section>, e.g. ecfr-bulk-xml/2024-03-17/title-21/178.3610.

• <section> is the section number from the XML N attribute, without §/§§, with dashes normalized to -; reserved ranges look like 1402.104-1402.111.
• 101 documents whose section number is repeated within a title (e.g. multi-volume § 17.95 in Title 50, forms numbered Sec. 1 in Title 46) get @<NODE> appended, where NODE is the XML hierarchy id.
• dataset is ecfr and mime_type is text/markdown.

Validation.

• Documents with spaced designators (\( [A-Za-z0-9]{1,4} \)): 52.6% at b48604bc → 0 (0.0%).
• Newlines per 1k characters: 8.58. For comparison, b48604bc has 4.4 and the January 2025 revision has 9.33.
• No empty documents.
• Decode → re-encode round-trip is identical for 500/500 sampled documents.
• For the 204,726 sections also present in the January 2025 revision (matched by title and section), the median character similarity is 0.998 on raw text and 1.000 after normalizing Markdown marks, dashes and whitespace and dropping renderer-only editorial-note boxes. 97.0% of those sections are ≥ 0.9.
• Most low-similarity cases are tables, which the January revision dropped, or sections amended between the two editions.

Earlier revisions (for reproducibility).

• b48604bc2510d40cf9ae6191de4fdcbe1d8b8925: 206,781 docs, March-2025 conversion with the defect.
• 4a92ef1765fd12d8ed947e179a1bbc1c2003af2d: January 2025, clean conversion. It has 262,210 rows over several renderer dates, with no score column.

For example: load_dataset("alea-institute/kl3m-data-ecfr", revision="4a92ef1765fd12d8ed947e179a1bbc1c2003af2d").

Description

This dataset is part of the ALEA Institute's KL3M Data Project, which provides copyright-clean training resources for large language models.

Dataset Details

• Format: Parquet files containing document text and metadata
• License: CC BY 4.0
• Tokenizer: The tokens field uses the kl3m-004-128k-cased tokenizer, a case-sensitive 128K vocabulary tokenizer optimized for legal, financial, and enterprise documents

Abstract

Practically all large language models have been pre-trained on data that is subject to global uncertainty related to copyright infringement and breach of contract. This creates potential risk for users and developers due to this uncertain legal status. The KL3M Data Project directly confronts this critical issue by introducing the largest comprehensive training data pipeline that minimizes risks related to copyright or breach of contract.

The foundation of this project is a corpus of over 132 million documents and trillions of tokens spanning 16 different sources that have been verified to meet the strict copyright and licensing protocol detailed in the project. We are releasing the entire pipeline, including:

1. The source code to acquire and process these documents
2. The original document formats with associated provenance and metadata
3. Extracted content in a standardized format
4. Pre-tokenized representations of the documents
5. Various mid- and post-train resources such as question-answer, summarization, conversion, drafting, classification, prediction, and conversational data

All of these resources are freely available to the public on S3, Hugging Face, and GitHub under CC-BY terms. We are committed to continuing this project in furtherance of a more ethical, legal, and sustainable approach to the development and use of AI models.

Legal Basis

This dataset is fully compliant with copyright law and contractual terms. The content is included based on the following legal foundation:

• Public domain materials
• US government works
• Open access content under permissive licenses
• Content explicitly licensed for AI training

Papers

For more information about the KL3M Data Project, please refer to:

• The KL3M Data Project: Copyright-Clean Training Resources for Large Language Models
• KL3M Tokenizers: A Family of Domain-Specific and Character-Level Tokenizers for Legal, Financial, and Preprocessing Applications

Citation

If you use this dataset in your research, please cite:

@misc{bommarito2025kl3mdata,
  title={The KL3M Data Project: Copyright-Clean Training Resources for Large Language Models},
  author={Bommarito II, Michael J. and Bommarito, Jillian and Katz, Daniel Martin},
  year={2025},
  eprint={2504.07854},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

@misc{bommarito2025kl3m,
  title={KL3M Tokenizers: A Family of Domain-Specific and Character-Level Tokenizers for Legal, Financial, and Preprocessing Applications},
  author={Bommarito II, Michael J. and Katz, Daniel Martin and Bommarito, Jillian},
  year={2025},
  eprint={2503.17247},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

About ALEA

The ALEA Institute is a non-profit research organization focused on advancing AI for business, law, and governance. Learn more at https://aleainstitute.ai/.