Datasetpublic

incrediblecrab/crs-bill-summaries

US Bill and Resolution Summaries (Congressional Research Service) Every summary of a bill or resolution of the United States Congress that the Congressional Research Service (CRS) wrote and the Congress.gov API lists, from the 93rd Congress (1973-1974) on, with its text. CRS summarizes a measure when it is introduced and again at later actions, such as passing a chamber, so a bill can have several summaries; action_desc names the action. Nothing here is edited by hand. The… See the full description on the dataset page: https://huggingface.co/datasets/incrediblecrab/crs-bill-summaries.

sourceHugging Faceotherupdated 4h agoView on Hugging Face

1likes424downloads

Dataset Card

US Bill and Resolution Summaries (Congressional Research Service)

Every summary of a bill or resolution of the United States Congress that the Congressional Research Service (CRS) wrote and the Congress.gov API lists, from the 93rd Congress (1973-1974) on, with its text. CRS summarizes a measure when it is introduced and again at later actions, such as passing a chamber, so a bill can have several summaries; action_desc names the action.

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

421,845 of 421,845 summaries (100.0%) as of 2026-09-26T00:28:58Z UTC, when the API last listed 421,845. 216 of 216 slices (one congress and bill type each) have been read in full.

Last complete sync: 2026-09-25T22:33:30Z UTC.

Congress	Years	Rows	Listed	Slices read in full
119	2025-2026	5,914	5,914	8 of 8
118	2023-2024	14,257	14,257	8 of 8
117	2021-2022	21,270	21,270	8 of 8
116	2019-2020	20,007	20,007	8 of 8
115	2017-2018	16,501	16,501	8 of 8
114	2015-2016	15,359	15,359	8 of 8
113	2013-2014	13,404	13,404	8 of 8
112	2011-2012	14,856	14,856	8 of 8
111	2009-2010	17,483	17,483	8 of 8
110	2007-2008	18,620	18,620	8 of 8
109	2005-2006	17,159	17,159	8 of 8
108	2003-2004	14,833	14,833	8 of 8
107	2001-2002	13,382	13,382	8 of 8
106	1999-2000	12,419	12,419	8 of 8
105	1997-1998	10,393	10,393	8 of 8
104	1995-1996	9,046	9,046	8 of 8
103	1993-1994	11,075	11,075	8 of 8
102	1991-1992	13,522	13,522	8 of 8
101	1989-1990	13,503	13,503	8 of 8
100	1987-1988	13,292	13,292	8 of 8
99	1985-1986	13,381	13,381	8 of 8
98	1983-1984	14,280	14,280	8 of 8
97	1981-1982	14,971	14,971	8 of 8
96	1979-1980	17,167	17,167	8 of 8
95	1977-1978	23,378	23,378	8 of 8
94	1975-1976	25,290	25,290	8 of 8
93	1973-1974	27,083	27,083	8 of 8

Listed is the API's count for the slices when they were last read in full or checked.

Use

python

from datasets import load_dataset
summaries = load_dataset("incrediblecrab/crs-bill-summaries", split="train")

sql

-- DuckDB, straight from the Hub: every summary of H.R. 1 of the 119th Congress, oldest first
SELECT version_code, action_date, action_desc, text FROM 'hf://datasets/incrediblecrab/crs-bill-summaries/data/*.parquet' WHERE congress = 119 AND bill_type = 'hr' AND bill_number = 1 ORDER BY action_date, version_code;

Files

—data/{congress}-{type}.parquet: one row per summary of one congress and bill type, sorted by id, such as data/119-hr.parquet. The congress has three digits (093-hr) so the files sort in order. The 8 types are hr, s, hjres, sjres, hconres, sconres, hres, sres.
—manifest.json: per slice, the row count, SHA-256, the API's count, when it was last read in full and any windows it could not read; the last listing; the last 20 runs.

Schema

Column	Type	Description
`id`	string	congress-type-number-versionCode, for example 119-hr-8893-00
`congress`	int32	The Congress the bill or resolution was introduced in (119 for 2025-2026)
`bill_type`	string	hr, s, hjres, sjres, hconres, sconres, hres or sres
`bill_number`	int32	The bill's number
`version_code`	string	The API's versionCode: which of the bill's summaries this is (00 for the one at introduction)
`action_date`	string	The date of the action the summary describes (YYYY-MM-DD)
`action_desc`	string	The action the summary describes, such as "Introduced in House" or "Passed Senate amended"
`title`	string	The bill's title as the listing gives it
`origin_chamber`	string	House or Senate
`current_chamber`	string	The chamber the listing gives for the summary
`text`	large_string	The summary as plain text
`html`	large_string	The summary as the API gives it (HTML)
`summary_update_date`	string	The API's lastSummaryUpdateDate
`fetched_at`	string	When the run that wrote this row ran (UTC); a later read that found the summary unchanged is not recorded

How it stays current

A GitHub Actions job is scheduled every 12 hours, at 00:00 and 12:00 UTC. It reads the summaries whose updateDate is at most 48 hours older than the newest the previous run saw, and commits the slices they changed with this card. Once a day, on the first run at least 20 hours after the last check, it also asks the API for each slice's count, reads again in full each slice whose count differs from its rows here, and the 8 slices read longest ago once that was more than 30 days ago. A slice read in full loses the summaries the API no longer lists.

The listing orders summaries by updateDate, but summaries that share one come back in a different order from one request to the next, so paging by offset skips some and repeats others: one pass over the House bills of the 100th Congress read 6,644 distinct summaries of 6,765 on September 25, 2026. So the pipeline reads windows of updateDates instead, moving each window's end down to the oldest second on its page, and pages a second that fills a whole page in several orders until it has read that second's count. That read all 6,765 of those summaries in 102 requests.

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 summary can take more than 12 hours to appear. Each run downloads manifest.json and the slices it updates, 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

—Only the summaries: not the bills' text, actions, cosponsors or status, which the API lists elsewhere.
—The API lists no summaries before the 93rd Congress.
—title is the one title the listing gives with the summary; a bill has several, and the listing does not say which this is.
—In a second that holds more summaries than one page, every order the pipeline tries may still miss some. manifest.json records each such second with its count and how many were read, and each slice's failed is how many it lists that are not here; the status above totals both.
—A summary the pipeline cannot read keeps its earlier row, if it has one, and is listed with its error under failures in manifest.json; later runs try it again.

License

CRS summaries are works of the United States Government and are not subject to copyright in the United States (17 U.S.C. § 105).