Skip to main content
1,502,173 sections across 14 corpus types, measured 2026-09-07. The corpusType token for each is in Corpus types. The FAR and DFARS rows are inside the CFR row, not additional to it: they split existing Title 48 data by chapter.

​
Scope limits by corpus

Read the entry for any corpus you are about to rely on. Each one states what is deliberately not held.
The source is the federal publisher’s own annual USCODE package on GovInfo, issued once per edition rather than amended in place. A weekly job asks GovInfo whether a newer edition exists and re-ingests the whole Code when one appears. Between editions the text does not move.The edition served is 2024, current through 2025-01-06. For federal law enacted after that date use corpusType=SESSION_LAW, which does refresh weekly, or corpusType=FEDERAL_REGISTER for agency rulemaking.
Federal public and private laws as enacted, session by session, 1789-2026, from 137 of the 140 volumes. Volumes 7, 8 and 44 are not held. This is the source the US Code is compiled from, so a US Code section’s publicLawCites resolve here to the enacting text.Sections resolve by actId, for example SAL_PL116-136_DVA_TII_S1109 (CARES Act section 1109).Because these are laws as enacted they are a historical record rather than a statement of current law: they carry actStatus: "enacted" and goodLawStatus: "unknown" deliberately. For what the law says today, use corpusType=USC.
GPO’s officially published annual edition of the CFR, 31 editions from 1996 to 2026, all 49 titles. A section is stored once per distinct run of text rather than once per edition, so 312,132 distinct sections are held as 886,066 section versions and served as 2,844,809 passages.The section count is what enters the federal total. This is the same body of regulation as the CFR corpus observed at published dates, not additional breadth.Two behaviors a caller needs. editionsObserved is an enumeration, never a range: a version observed in 1998, 1999 and 2003 lists exactly those years, and a date in an unobserved gap comes back with asOf.isBounded: false rather than the nearest year held. A citation year is the volume’s own <REVISED> year, not the folder GPO filed it under, because GPO reprints unchanged volumes forward without re-dating them; 167 versions in title 26 sit in the 1997 folder stamped 1990.CFR_ANNUAL serves an edition as published, while asOf on GET /us/statutes/section/{actId}/body reconstructs a section from observed change history. See Point-in-time law for when each applies.
Held: notices that act on a rulemaking, which is withdrawing a proposed rule, extending or reopening a comment period, publishing significant agency guidance, and announcing negotiated rulemaking. That is 17,616 of the 766,073 notices published in the period, about 2.3%. Every point carries frNoticeClass naming the rule that admitted it.Not held: Paperwork Reduction Act information collections, meeting notices, permits and applications, antidumping proceedings, Privacy Act system-of-records notices, and the unclassified remainder. Do not read this token as complete notice coverage.Notices carry an empty CFR cross-reference graph (crossReferencesCfr and frTopics at 0%, frRegulationIdNumbers at 2%, against 100% and 94% on rules), so a notice is not reachable by a CFR-citation filter. frDocketIds at 73.5% is the join key that works. A notice amends nothing, so amendmentYears is empty and goodLawStatus is unknown.
Federal matters decided by an agency through a formal process, with precedential weight on that body’s later decisions.Citation forms: FTC Docket 9436, Exchange Act Release No. 34-106254, YYYY MSPB N, I&N Dec., NNN NLRB No. N, NN Op. O.L.C. NNN, Bureau File No. 2024-CFPB-0002.goodLawStatus is unknown on every body except the CFPB slice. No publisher here emits a machine-readable overruling signal, so a decision is served as issued and whether a later decision overruled it is not asserted.Five further scope limits:
  • The FTC slice is the Part 3 administrative body only, 1,431 records carrying the Commission’s own Administrative action type. Its 2,039 federal-court records and roughly 2,636 records carrying neither action type are not held. FTC rules are excluded because they are codified at 16 C.F.R. and served under corpusType=CFR.
  • The FTC unit is a FILING, not a decision. A Part 3 docket is mostly documents written by the parties: 21% of passages are the operative instruments (complaints, decisions and orders, consent agreements) and the rest are motions, notices of appearance, exhibits and public-comment analyses. Nothing is dropped on that basis, the whole docket is served, but the per-filing label is not yet a response field or a search filter.
  • The NLRB slice is now the full archive, 1935-12-30 to date: 66,597 of the 66,600 decisions the Board’s own volume listing enumerates. Two are excluded and named - 227 NLRB 427, whose PDF contains only the caption and no body under two independent extractors, and 321 NLRB 10, a file no OCR engine will accept. Neither is a coverage decision; both are re-tried if the Board re-issues the file.
  • OLC opinions issued in 2021 and later are not included. They have not appeared in a bound volume, and the Department’s slip-opinion listing cannot be paginated.
  • SEC ALJ initial decisions are not held, and neither are BIA reporter volumes 1 to 7, which EOIR does not publish.
Two sources under one corpusType=US_TAX_TREATY, 19,646 retrieval passages.The IRS mirror (173, us_tax_treaty and us_tax_treaty_technical_explanation): 71 bilateral income tax conventions for 68 countries, 71 Treasury Technical Explanations, 23 protocols, 6 exchanges of notes, 1 memorandum of understanding, 1 negotiation statement. A protocol is served as its own document rather than folded into the convention it amends, because a protocol is what moves withholding rates, limitation-on-benefits articles and entry-into-force dates. No estate or gift tax treaties.The Senate record (1,476): all 780 treaties the Senate has considered since 1949, tax and non-tax alike. senate_treaty (780) carries title, parties, subject index and dated procedural history; senate_treaty_resolution (406) is the resolution of advice and consent, where the reservations, understandings and declarations that narrow a treaty’s effect in US law live; senate_treaty_document (290) is the President’s transmittal message, from GPO.Two limits. About 60% of the Senate record is metadata with no instrument text from either publisher, which source states rather than hides. Termination is not tracked, because no source held states it, so goodLawStatus is unknown on every record.
An act’s own text as amended through a stated later public law, from GovInfo’s official compilations. Act-organized rather than codified, so a citation like “Section 210 of the Public Utility Regulatory Policies Act” resolves here even though that section was never codified into a positive-law title and has no USC address.Distinct from Statutes at Large, which holds the same acts as enacted. Each section states its own currency in currentThrough, ranging from Public Law 96-308 (1980) to Public Law 119-75 (2026). 2.6% of compilations state no currency at all and carry compilationCurrencyState: "unstated" rather than a guessed date. Cross-references into USC are captured as crossReferencesUsc.
The 302 current guideline sections, plus Appendix A (the statutory index) and Appendix B, whose Part III carries the text of FORMER guidelines and commentary as they read before amendment. The former versions are separate documents, not variants of the current section.
Federal Register and Executive Order cadence depends on federalregister.gov, which has been throttling and rate-limiting traffic in 2026. That can occasionally push a weekly pull later.

Agency guidance sources

The 51 named sources behind AGENCY_GUIDANCE, with per-source caveats.

State and territory coverage

Per-jurisdiction counts for all 52 jurisdictions.
Last modified on September 7, 2026