README.md

---
license: cc0-1.0
language: [en]
task_categories: [text-retrieval, question-answering]
size_categories: [100K<n<1M]
tags: [legal, law, regulations, cfr, federal-law, us-law, administrative-law, public-domain, rag, legal-nlp]
pretty_name: US Federal Regulations — Code of Federal Regulations (verbatim)
configs:
- config_name: all
  data_files: data/*/regulations.jsonl.gz
  default: true
dataset_info:
  config_name: all
  features:
  - name: id
    dtype: string
  - name: level
    dtype: string
  - name: jurisdiction
    dtype: string
  - name: code
    dtype: string
  - name: section
    dtype: string
  - name: citation
    dtype: string
  - name: title
    dtype: string
  - name: text
    dtype: string
  - name: source
    dtype: string
  - name: license
    dtype: string
  - name: retrieved_at
    dtype: string
  - name: extra
    dtype: string
---

# US Federal Regulations — the CFR, held word for word

> **Code & tools:** [github.com/docketx](https://github.com/docketx) — [legal-scrambler](https://github.com/docketx/legal-scrambler) pseudonymises a case file on your own hardware before a frontier model sees it; [claude-for-legal](https://github.com/docketx/claude-for-legal) is the Claude Code plugin (`docketx-open-law`) that loads these datasets and checks citations against them.

**222,767 regulations** — 219,061 CFR sections and 3,706 appendices — across **all 49 titles
of the Code of Federal Regulations**, each one as the agency publishes it.

Statutes say what Congress wrote. The CFR says what the agencies actually require, and for most regulated
work the regulation is the operative text: an employer's duty under 29 C.F.R. § 1910.1200, a broker's under
17 C.F.R. § 240.10b-5. This is the other half of federal law, and it is here verbatim — a paraphrase is not
usable in a filing.

## What is in it

| CFR title | name | records | of which appendices | source file last built by GPO |
|--:|---|--:|--:|---|
| 1 | General Provisions | 271 | 0 | 11-Aug-2026 20:55 |
| 2 | Federal Financial Assistance | 1,577 | 70 | 19-Aug-2026 20:28 |
| 3 | The President | 20 | 0 | 20-May-2024 17:04 |
| 4 | Accounts | 221 | 1 | 19-Jul-2024 22:33 |
| 5 | Administrative Personnel | 5,340 | 35 | 04-Sep-2026 20:29 |
| 6 | Domestic Security | 589 | 11 | 13-Aug-2026 21:09 |
| 7 | Agriculture | 16,373 | 124 | 10-Sep-2026 21:29 |
| 8 | Aliens and Nationality | 923 | 0 | 10-Sep-2026 21:27 |
| 9 | Animals and Animal Products | 2,340 | 0 | 20-Aug-2026 21:15 |
| 10 | Energy | 5,344 | 193 | 10-Sep-2026 21:32 |
| 11 | Federal Elections | 560 | 0 | 09-Jun-2026 20:48 |
| 12 | Banks and Banking | 7,171 | 255 | 09-Sep-2026 21:53 |
| 13 | Business Credit and Assistance | 1,657 | 6 | 01-Sep-2026 21:44 |
| 14 | Aeronautics and Space | 6,370 | 166 | 04-Sep-2026 20:32 |
| 15 | Commerce and Foreign Trade | 2,348 | 175 | 31-Aug-2026 22:19 |
| 16 | Commercial Practices | 2,194 | 83 | 01-Sep-2026 21:48 |
| 17 | Commodity and Securities Exchanges | 3,394 | 79 | 09-Sep-2026 21:50 |
| 18 | Conservation of Power and Water Resources | 2,026 | 11 | 25-Aug-2026 21:21 |
| 19 | Customs Duties | 3,132 | 35 | 27-Aug-2026 21:06 |
| 20 | Employees' Benefits | 5,589 | 31 | 10-Sep-2026 21:29 |
| 21 | Food and Drugs | 8,333 | 4 | 09-Sep-2026 21:50 |
| 22 | Foreign Relations | 3,311 | 41 | 01-Sep-2026 21:46 |
| 23 | Highways | 1,015 | 41 | 02-Sep-2026 20:48 |
| 24 | Housing and Urban Development | 4,848 | 23 | 21-Aug-2026 21:35 |
| 25 | Indians | 4,710 | 9 | 10-Dec-2025 00:35 |
| 26 | Internal Revenue | 6,013 | 0 | 09-Sep-2026 21:51 |
| 27 | Alcohol, Tobacco Products and Firearms | 3,857 | 0 | 18-Aug-2026 21:40 |
| 28 | Judicial Administration | 3,051 | 50 | 02-Sep-2026 20:50 |
| 29 | Labor | 7,201 | 102 | 10-Sep-2026 21:31 |
| 30 | Mineral Resources | 5,793 | 13 | 09-Sep-2026 21:50 |
| 31 | Money and Finance: Treasury | 5,345 | 70 | 27-Aug-2026 21:08 |
| 32 | National Defense | 3,702 | 114 | 18-Aug-2026 21:41 |
| 33 | Navigation and Navigable Waters | 4,619 | 25 | 10-Sep-2026 21:30 |
| 34 | Education | 3,170 | 32 | 10-Sep-2026 21:30 |
| 36 | Parks, Forests, and Public Property | 3,081 | 27 | 25-Aug-2026 21:21 |
| 37 | Patents, Trademarks, and Copyrights | 1,267 | 11 | 27-Aug-2026 21:07 |
| 38 | Pensions, Bonuses, and Veterans' Relief | 2,871 | 16 | 28-Aug-2026 21:24 |
| 39 | Postal Service | 1,125 | 6 | 15-Jun-2026 20:53 |
| 40 | Protection of Environment | 24,639 | 1,276 | 10-Sep-2026 22:07 |
| 41 | Public Contracts and Property Management | 2,945 | 11 | 01-Sep-2026 22:05 |
| 42 | Public Health | 7,244 | 25 | 14-Aug-2026 21:38 |
| 43 | Public Lands: Interior | 5,207 | 14 | 13-Aug-2026 21:13 |
| 44 | Emergency Management and Assistance | 907 | 6 | 24-Jun-2026 14:49 |
| 45 | Public Welfare | 5,652 | 33 | 01-Sep-2026 22:05 |
| 46 | Shipping | 8,372 | 51 | 02-Sep-2026 20:52 |
| 47 | Telecommunication | 4,734 | 15 | 04-Sep-2026 20:31 |
| 48 | Federal Acquisition Regulations System | 10,128 | 10 | 02-Sep-2026 20:52 |
| 49 | Transportation | 8,944 | 238 | 10-Sep-2026 21:45 |
| 50 | Wildlife and Fisheries | 3,244 | 168 | 02-Sep-2026 21:20 |

**All 49 CFR titles are complete.** CFR Title 35 does not appear because it is *Reserved* — the
eCFR's own title index marks it `reserved: true`, so 49 titles is the whole Code, not a subset.

Every count in that table was measured by reading the file in `data/` on 2026-09-11. Nothing is estimated and
nothing is taken from an index page's own total.

## As of

The text is the **current** CFR as republished by the U.S. Government Publishing Office from the
Electronic Code of Federal Regulations, pulled **2026-09-11**. The eCFR title index reported itself
`up_to_date_as_of 2026-09-09` at pull time. GPO rebuilds a title's bulk file when that title
changes, so each title carries its own build date — the right-hand column above, and
`extra.bulk_last_modified` on every row. Titles that agencies rarely touch carry older build dates by
design: Title 3 (*The President*) was last rebuilt 20-May-2024, and the eCFR title index gives its
`latest_amended_on` as 2015-03-17.

This is a snapshot, not a live feed. For a section that has been amended since its title's build date,
check the eCFR URL in `source`.

## How complete is it — checked against the live eCFR

Completeness is not asserted, it is measured. For every one of the 49 titles we asked the eCFR's own
structure endpoint (`/api/versioner/v1/structure/2026-09-09/title-{n}.json`, an allowed path) for the list of
section identifiers it currently publishes, and compared it with what is in these files:

| | count |
|---|--:|
| section identifiers the eCFR listed as live on 2026-09-09 | 220,526 |
| section identifiers held here | 219,055 (99.33%) |
| listed by eCFR, not held here | 1,520 |
| held here, no longer listed by eCFR | 49 |

Of the 1,520 not held, **1,476 carry no text of their own in the source** — they are undesignated
group headings whose substance lives in their subsections (48 C.F.R. 1.105 *Issuance*, 29 C.F.R. 2530.200a
*Scope*) or pure cross-reference headings (29 C.F.R. 1917.28). The remaining 44, together with the
49 we hold that the eCFR no longer lists, are the lag between a title's GPO build date and 2026-09-09.

No title is a partial pull. Every title is whole as of its build date.

## The record

Twelve columns, all strings — the same shape as `docketx/us-statutes`, so the two can be read together:

`id`, `level` (`federal`), `jurisdiction` (`us`), `code` (e.g. `29 C.F.R.`), `section` (e.g. `1910.1200`),
`citation` (e.g. `29 C.F.R. § 1910.1200`), `title`, `text` (verbatim), `source` (the eCFR URL for that
section), `license`, `retrieved_at`, `extra`.

`extra` is a JSON **string** (parse it with `json.loads`) carrying the place of the section in the Code —
`cfr_title`, `cfr_title_name`, `chapter`, `chapter_name`, `subchapter`, `part`, `part_name`, `subpart`,
`subpart_name`, `subject_group`, the eCFR `node` id, the `source_credit` (the Federal Register history note
printed with the section), and the `bulk_file` / `bulk_last_modified` it came from.

```python
from datasets import load_dataset
ds = load_dataset("docketx/us-regulations")                       # all 49 titles
osha = ds["train"].filter(lambda r: r["code"] == "29 C.F.R.")     # one title
```

## Where it came from, and how

- **Source:** the official bulk XML at `https://www.govinfo.gov/bulkdata/ECFR/title-{n}/ECFR-title{n}.xml`
  — GPO's bulk republication of the current eCFR, one file per title, 824,347,237 bytes in total.
- **robots first.** `https://www.govinfo.gov/robots.txt` places no `Disallow` on `/bulkdata/`, and GPO
  publishes a bulkdata sitemap for the ECFR collection. Every request identified itself as
  `DocketRouterBot/1.0 (+https://docketrouter.ai/bot)` and was paced at 1.5s.
- **What we did not do.** eCFR's own full-text endpoints — `/api/versioner/v1/full/` and
  `/api/renderer/v1/content/` — are `Disallow`ed in `https://www.ecfr.gov/robots.txt`, so they were not
  used and no user-agent was switched to get around them. Only `…/api/versioner/v1/titles.json`, which no
  rule covers, was read, for title names and the as-of date.
- **Not text:** 7,171 sections and appendices marked `[Reserved]` in the source carry no text and are
  not included;
  a further 1,683 entries are heading-only stubs (an undesignated group heading such as 48 C.F.R. 1.105,
  whose text lives in its subsections) or appendices that exist only as a scanned form image. Nothing was
  dropped that had extractable text. Tables are rendered as pipe-separated rows; images are not
  transcribed, and the figure filenames are kept in `extra.graphics`.
- **Appendices follow the source.** Appendices attached to a *part* or *subpart* are their own records
  (3,706 of them). Where the source nests an appendix inside a section — as it does with the mandatory
  Appendices A-F to 29 C.F.R. § 1910.1200 — that text is part of that section's record, which is why the
  1910.1200 row is 307,615 characters long.

## Licence

The Code of Federal Regulations is a work of the **United States Government** — published by the Office of
the Federal Register (National Archives and Records Administration) and the Government Publishing Office —
and is **in the public domain** in the United States. Every row carries
`license: public-domain-government-work`. No claim of copyright is made over this compilation.

Part of the [DocketX](https://docketx.ai) open legal corpus, alongside `docketx/us-statutes`,
`docketx/court-rules`, `docketx/us-caselaw`, `docketx/oral-arguments-us` and `docketx/us-judges`.