incrediblecrab/crs-constitution

US Constitution Annotated (Congressional Research Service)

The Constitution of the United States of America: Analysis and Interpretation, which the Congressional Research Service (CRS) prepares and the United States Congress prints as a Senate Document: every edition and supplement that GovInfo holds, with the full text of each part of the book and GovInfo's record of it.

Nothing here is edited by hand. The pipeline, its tests and its schedule are in github.com/incrediblecrab/crs-service-products, and this card is rendered from manifest.json in the same commit.

Status

495 of 495 units (100.0%) as of 2026-10-07T03:13:26Z UTC, when GovInfo last listed 495. 495 rows have text. 0 units failed 3 times and are retried every 24 hours; their errors are in manifest.json.

Last complete sync: 2026-10-07T03:13:26Z UTC.

Package	Rows	With text	Listed	Complete
GPO-CONAN-1992	49	49	49	yes
GPO-CONAN-1996	1	1	1	yes
GPO-CONAN-1998	1	1	1	yes
GPO-CONAN-2000	1	1	1	yes
GPO-CONAN-2002	48	48	48	yes
GPO-CONAN-2004	1	1	1	yes
GPO-CONAN-2006	1	1	1	yes
GPO-CONAN-2008	1	1	1	yes
GPO-CONAN-2010	1	1	1	yes
GPO-CONAN-2012	49	49	49	yes
GPO-CONAN-2013	49	49	49	yes
GPO-CONAN-2014	49	49	49	yes
GPO-CONAN-2014-SUPP	1	1	1	yes
GPO-CONAN-2017	49	49	49	yes
GPO-CONAN-2018-SUPP	1	1	1	yes
GPO-CONAN-2020-SUPP	2	2	2	yes
GPO-CONAN-2022	43	43	43	yes
GPO-CONAN-2024-SUPP	1	1	1	yes
GPO-CONAN-REV-2014	49	49	49	yes
GPO-CONAN-REV-2015	49	49	49	yes
GPO-CONAN-REV-2016	49	49	49	yes

Use

from datasets import load_dataset
conan = load_dataset("incrediblecrab/crs-constitution", split="train")

-- DuckDB, straight from the Hub: the parts of the newest edition, in the book's order
SELECT sequence, subsequence, title, pages FROM 'hf://datasets/incrediblecrab/crs-constitution/data/*.parquet' WHERE kind = 'edition' AND date_issued = (SELECT max(date_issued) FROM 'hf://datasets/incrediblecrab/crs-constitution/data/*.parquet' WHERE kind = 'edition') ORDER BY sequence, subsequence;

Files

• data/{package}.parquet: one row per unit of one GovInfo package, sorted by id. A unit is a granule, GovInfo's part of a book, or a whole package that GovInfo does not split into granules (most supplements).
• manifest.json: per package, the row count, SHA-256, rows with text and whether it is complete; failed units with their errors; the last listing; the last 20 runs.

Schema

Column	Type	Description
id	string	GovInfo's granule id (GPO-CONAN-2022-8), or the package id for a package without granules (GPO-CONAN-2024-SUPP)
package_id	string	The GovInfo package: one printed edition or supplement
package_title	string	The package's title as GovInfo's collection listing gives it
kind	string	edition or supplement, from the package title
title	string	The granule's title, a part of the book such as "Article I - Legislative Branch"; for a package without granules, its title
sequence	int32	The granule's number in its package (8 in GPO-CONAN-2022-8); null for a package without granules
subsequence	int32	For a granule under a heading, its number there (1 in GPO-CONAN-1992-9-1); else null
date_issued	string	GovInfo's dateIssued (YYYY-MM-DD)
pages	int32	Pages in the PDF, as pdftotext counts them
text	large_string	The PDF's text layer, pages separated by form feeds, so page n is text.split("\f")[n - 1]; null when the PDF has no text layer
url	string	The unit's page on govinfo.gov (the record's detailsLink)
pdf_url	string	The GovInfo API link the PDF was fetched from; the API needs an api.data.gov key
pdf_sha256	string	SHA-256 of the PDF's bytes
updated_at	string	The package's lastModified when this row was written. A changed value triggers a re-fetch; one that finds the unit unchanged keeps the row and records the new value in manifest.json instead
metadata	large_string	GovInfo's summary record for the unit, as JSON
fetched_at	string	When the fetch that wrote this row ran (UTC)

How it stays current

A GitHub Actions job is scheduled at 00:00 and 12:00 UTC. It lists the GPO-CONAN packages in GovInfo's collection of additional government publications (GPO) and the granules of each, then fetches the units that are new or whose package's lastModified changed, removes the ones GovInfo no longer has, and commits the changed packages with this card. lastModified is GovInfo's stamp, not the book's: on September 25, 2026, every package's was from March 7 to 8 or August 18, 2025, long after it was printed. A unit whose record and PDF did not change keeps its row, and its package file is not rewritten.

The job writes with Hugging Face Trusted Publishing, so no write token is stored anywhere. manifest.json names each run's writer: github-actions for this job, local for the same pipeline run from a computer. GitHub starts scheduled jobs late, or drops them, when it is busy, so a new supplement can take more than 12 hours to appear. Each run downloads manifest.json, and the Hub counts a download for each 5 minutes in which a run reads files, so part of this dataset's download count is this job.

Known gaps

• This is the printed book, not the web edition at constitution.congress.gov, which CRS updates between printings. That site answers automated requests with a Cloudflare challenge, which this pipeline does not try to get past.
• A granule that only heads others, such as GPO-CONAN-1992-9 ("The Constitution of the United States of America (With Annotations)") over GPO-CONAN-1992-9-1 to -9-8, has no PDF of its own and is not a row; its parts are. On September 25, 2026, 9 editions had 2 such headings each.
• text is the PDF's text layer: it keeps running heads, page numbers and footnote markers, and has no table structure. The first page of the 2020 Supplement's errata sheet (GPO-CONAN-2020-SUPP-2) has text in its text layer that the rendered page does not show, such as "115th" beside the visible "116th Congress", so its text interleaves the two.
• A PDF without a text layer gives a row with null text, fetched again every 7 days.

License

The Constitution Annotated is a work of the United States Government and is not subject to copyright in the United States (17 U.S.C. § 105).