Skip to main content
Updates to the Vaquill Developer API: new endpoints, new corpora, and behavior changes. For live coverage counts, call GET /statutes/coverage.
​
NLRB Board decisions widen from a 15-year window to the full 90-year archive: 66,597 decisions back to 1935

​
NLRB Board decisions: the complete archive

The NLRB slice was a deliberate 15-year window of 2,762 decisions. It is now the full published archive: 66,597 decisions, 1935-12-30 to date, under source=nlrb_board_decision. That is 24x the previous coverage, and 1,705,304 retrievable passages.This is the Board’s entire body of unfair-labour-practice and representation decisions, the ones that bind later Board panels and the administrative law judges beneath them. The count is the publisher’s own: the volume listing enumerates 66,600 decisions across volumes 0 to 375, and 66,597 are served.Currency is still not asserted, and widening the corpus does not change that. goodLawStatus remains unknown on every one of these decisions. The Board publishes no overruling or supersession signal, and its CITENET Classified Index is a subject digest rather than a citator, with no reverse edge from an overruled decision to the one that overruled it. A decision is served as issued. Whether a later Board overruled it is a question we do not answer, and the older the decision the likelier that matters, because Board doctrine moves with Board composition. Each point carries a coverageNote saying so.Two decisions are excluded, and named rather than silently absent:
  • 227 NLRB 427 - the publisher’s PDF contains only the case caption and no body. Two independent extractors agree, so there is nothing to serve.
  • 321 NLRB 10 - a file no OCR engine will accept.
Both are re-tried if the Board re-issues them.Both citation forms are served and neither is normalized into the other, as before: NNN NLRB No. N while a decision is a slip opinion, NNN NLRB NNNN once the bound volume closes and assigns it a page. Across the full archive the balance shifts hard toward the bound form, because most of the archive is closed volumes. That is expected, not a regression.corpusType=AGENCY_ADJUDICATION now serves 100,920 sections across its seven bodies, up from 37,085.
​
IRS written determinations: the full 45,358-document archive, and they are not precedent

​
IRS Written Determinations join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers IRS written determinations released under 26 U.S.C. § 6110 (source=irs_written_determination): 45,358 documents across 402,546 retrieval passages, which makes it the largest source in this token by document count. It breaks down as 28,641 Private Letter Rulings, 5,573 Chief Counsel Advice memoranda, 5,482 Tax Exempt and Government Entities determination letters and 842 Technical Advice Memoranda, plus 4,820 the IRS does not type on the document itself.That is 45,358 of the 45,431 determinations the IRS lists, 99.84%.Each carries the publisher’s own Number, UILC code list, Subject and Release Date, so effectiveDate, effectiveYear and the UILC issue codes come from the IRS rather than from inference. Coverage runs 1999-01-08 to 2026-09-04, 28 distinct years.
A written determination binds one taxpayer and may not be cited as precedent. 26 U.S.C. § 6110(k)(3) provides that it “may not be used or cited as precedent” - that is the statute, not our caveat. Every one of the 45,358 therefore carries actStatus: "non_precedential" rather than in_force, and says so again in currencyNote. Read one as evidence of how the Service analysed one taxpayer’s facts, never as a rule of general application.
goodLawStatus is unknown on all of them, deliberately. The Service publishes no status or revocation field for written determinations, so whether a later determination revoked or modified any given one is not asserted here. We would rather say we do not know than imply currency we cannot measure.9,873 documents were read by OCR, and extractionMethod tells you which. The IRS publishes much of this series as scans of signed paper rather than as typeset text, so those carry extractionMethod: "ocr" and the rest carry "text_layer". Two consequences are worth knowing before you rely on the text. The section symbol survives scanning less reliably than the words around it, so a passage may read section 501(c)(3) where the page prints § 501(c)(3) - search for either form. And a stamped control number in a letterhead can lose a digit, which is why actId and citation are taken from the IRS index rather than from the scanned text.73 determinations are not served: 56 are scans no OCR could read, 14 failed a legibility check, and 3 are not PDFs at the publisher’s end. A document that could not be read is withheld rather than served as fragments.
​
CMS Medicare Internet-Only Manuals now covered

​
CMS Medicare Internet-Only Manuals join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the CMS Medicare Internet-Only Manuals (source=cms_iom, 11,018 sections across 21 publications, 26,153 retrieval passages), addressed at CMS’s own section numbering and cited Medicare Claims Processing Manual (Pub. 100-04), Ch. 12 SS 20.1. This is how Medicare actually runs day to day: Pub. 100-04 says what a provider may bill and how, Pub. 100-02 says what Medicare covers, and Pub. 100-08 is what a MAC reads before it opens an audit.Every section carries its own revision stamp. CMS prints (Rev. N, Issued, Effective, Implementation) on each revised section, so amendmentYears, lastAmendedYear, effectiveDate and transmittalNumber come from the publisher rather than from inference. 98.4% of sections carry at least one stamp, spanning 2000 to 2026 across 27 distinct years.
These manuals are agency procedure, not law. They instruct CMS’s own Medicare administrative contractors on how to administer the program. Courts have repeatedly held that manual provisions are not binding substantive rules; the binding requirements are in the Social Security Act and 42 C.F.R. Every passage carries that statement in currencyNote and goodLawStatus is unknown on all of them. Do not read a billing instruction here as a regulation.
Five of CMS’s 25 publications hold nothing or almost nothing, and those are the publisher’s gaps rather than ours. Pub. 100-12 (State Medicaid Manual), 100-13 (Medicaid CHIP) and 100-23 (Payment Error Rate Measurement) publish no chapters at all; 100-21 is titled Reserved; and 100-18, the Medicare Prescription Drug Benefit (Part D) Manual, is a two-page table of contents whose own text states that several chapters “have not been disseminated via HPMS or Pub. 100-18”. Pub. 100-14 does not exist.That scope statement is now a FIELD rather than a footnote: GET /statutes/coverage returns sourceScopeNotes on each corpusTypes entry, keyed by the source value you can filter on. It carries a per-source declaration for a token that fans out to many unrelated sources, where the existing whole-token scopeNote cannot describe any one of them.CMS replaces manual chapters in place and publishes no superseded editions, so asOf is not bounded for this corpus and every passage states the date we captured it.
​
FTC Part 3 administrative filings now covered

​
FTC Part 3 administrative filings join Administrative Adjudications

corpusType=AGENCY_ADJUDICATION now also covers the Federal Trade Commission’s Part 3 adjudicative docket (source=ftc_administrative_decision, 18,023 documents), 1995 to date. This is where the Section 5 doctrine on unfair and deceptive acts and unfair methods of competition actually lives: administrative complaints, initial decisions, Commission opinions, consent agreements and the orders that modify them, across the Franchise Rule, the Telemarketing Sales Rule, COPPA, Made in USA, HSR and the Safeguards Rule.Cited by docket (FTC Docket 9436) where the record prints one, and always by the Commission’s permanent File Number in caseNumber.The unit is a FILING, not a decision, which is the thing to know before querying it. A Part 3 docket is mostly documents written by the parties: measured on the corpus, 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. The whole docket is served; nothing is dropped on that basis.Each filing is stored with the Commission’s own label (“Complaint”, “Decision and Order”, “Agreement Containing Consent Order”) and a normalised class. Neither is exposed as a response field or a search filter yet, so today a retrieval over this corpus returns operative and procedural filings together.Scope is the Part 3 administrative body only. The FTC’s 2,039 federal-court records and ~2,636 records carrying neither action type are not held, and FTC rules are excluded because they are codified at 16 C.F.R. and served under corpusType=CFR. caseStatus and docketNumber are record-type dependent (88.2% and 73.6% respectively) rather than extraction gaps.goodLawStatus is unknown on every filing. A consent order’s standard 20-year sunset is stated in the order’s own text and the Commission publishes no machine-readable field for it, so whether an order still binds is not asserted.The FTC’s ADVISORY opinions decide no case and remain under corpusType=AGENCY_GUIDANCE.
​
Coverage is 4,983,685 sections, and every corpus type is now filterable

​
Coverage is 4,983,685 sections

Counts are on the definition GET /statutes/coverage publishes as sectionCount, across all 52 jurisdictions.Two counting units are in play and they are not interchangeable. A section is one provision or document, which is what sectionCount reports. A retrieval passage is one indexed chunk, and a long document has many.The CFR annual editions enter the total at their section count, 312,132, not at their stored versions. These are editions over time of the same regulations, not additional distinct ones.

​
Every counted corpus type is now filterable

Checked against the deployed OpenAPI document on 2026-09-07: the published corpusType enum carries all 19 tokens, CFR_ANNUAL, FEDERAL_REGISTER_NOTICE, STATUTE_COMPILATION and AGENCY_ADJUDICATION among them. A corpus that /statutes/coverage counts is a corpus you can scope a search to; there is no longer a token that is counted but rejected.The six per-body source values are live in the same document: nlrb_board_decision, sec_commission_opinion, bia_precedent, ag_precedent, cfpb_enforcement_action and irs_irm.

​
Things that stay true when the number goes up

Nothing in this entry widens a claim, and four scope limits travel with these figures wherever they are quoted.
  • Federal Register notices are 17,616 of the 766,073 published in the period, 2.30%, four operative classes by design. That is not notice coverage and is never described as such.
  • The Internal Revenue Manual is agency procedure, not law. It is the IRS’s own instruction manual for its employees and does not bind the agency against a taxpayer.
  • The adjudication corpora (FTC Part 3, NLRB, MSPB, BIA, SEC, OLC, CFPB) carry no verified doctrinal currency. goodLawStatus is unknown by design on all but the CFPB slice, and nothing here implies good-law assurance. A live corpusType=AGENCY_ADJUDICATION search returns nlrb_board_decision hits with goodLawStatus: "unknown", which is the intended answer and not a gap.
  • Statutes at Large is held from 137 of the 140 volumes. Volumes 7, 8 and 44 are not held.
​
OFAC sanctions FAQs join AGENCY_GUIDANCE, with their prior versions as separately citable documents

​
OFAC Frequently Asked Questions, and their revision history

source=ofac_faq under corpusType=AGENCY_GUIDANCE. 989 current FAQs, cited OFAC FAQ 401, covering all 38 sanctions programs the Office of Foreign Assets Control administers. This is the interpretive layer over 31 C.F.R. chapter V: how the 50 Percent Rule reaches through an ownership chain (FAQ 401), whether a non-U.S. person is exposed for a transaction a General License authorizes (FAQ 1247), whether an institution may wait for its screening vendor’s overnight report (FAQ 43). None of it is in the CFR.

​
Prior versions are separate documents, not a field

361 prior versions are served as their own citable documents with actStatus: "superseded", taken from OFAC’s own FAQ Archive and stamped with the publisher’s timestamps, so the earlier text is retrievable rather than described. A superseded version never competes with the current answer in search; ask for it by act_id.238 FAQs (24.1%) carry at least one observed revision; the deepest is FAQ 999 with 12. A version whose text is byte-identical to the one before it is NOT counted as a revision - OFAC re-saves FAQs without changing them, and 183 of its 1,533 archive rows are exactly that.

​
Two bounds, served on every passage

The revision trail begins 2023-03-22 and goes no further back. OFAC migrated its FAQs to a new website that day and states that FAQs active at the time display the migration date rather than the date it first published that version. Those 860 version records carry observedDate and no effectiveDate, because that date is when we observed the text and not when OFAC published it.279 FAQ numbers have been rescinded and OFAC publishes nothing for them - no text, no date, no reason. A rescinded FAQ is absent from this corpus rather than marked withdrawn, so a gap in the numbering is a rescission, not a hole in our coverage. Both bounds are in currencyNote and coverageNote on every passage.

​
asOf answers from the stored version, not a reconstruction

GET /statutes/section/{act_id}/body?asOf=2026-04-01 returns the version OFAC published on that date, verbatim, with asOf.source = stored_edition. There is no diff walk and no basisChangeId: the earlier text is a document we hold, so we serve it rather than rebuild it.Three answers are distinguished, and the difference matters:That last row is the one to read. isBounded: true means the text is the earliest we hold and NOT the text in force on your date. Rendering it as point-in-time law makes a claim we did not make. Note isBounded does not mean the same thing on the stored_edition engine used by CFR_ANNUAL; read engine first.Sanctions LISTS (SDN, consolidated) and General Licenses are deliberately not here. General License links are served as cross-references.
​
FinCEN AML Act of 2020 implementation material is live as source=fincen_amla_material

​
FinCEN Anti-Money Laundering Act of 2020 implementation material

source=fincen_amla_material adds 13 documents across 159 retrieval passages: the AML/CFT National Priorities and the interagency statements issued with them, the Arts and Antiquities notice, threat-pattern and trend reports, the Section 6305 Report to Congress on no-action letters, the whistleblower bulletin, and the AML/CFT program NPRM fact sheets. Cited as FinCEN, Statement for Banks (AML Act of 2020 Sec. 6101, June 30, 2021).The AML Act rulemakings themselves are not duplicated here. The NPRMs and final rules implementing each section are Federal Register documents and are already served under corpusType=FEDERAL_REGISTER. This source carries only FinCEN’s own standalone publications.Completes the FinCEN Corporate Transparency Act and AML Act batch begun in the previous entry.
​
State financial-institutions regulator bulletins are live as source=state_financial_bulletin under corpusType=STATE_AGENCY_GUIDANCE

​
State financial-institutions regulator bulletins (Arizona, Oregon)

STATE_AGENCY_GUIDANCE now fans out into two named sources rather than one, and takes a source filter:155 new documents across 584 retrieval passages, from four regulators that each share an index with their state’s insurance regulator: New Jersey DOBI banking (81), Arizona DIFI (35, covering banks, credit unions, trust companies, escrow agents, consumer lenders, money transmitters, loan originators, collection agencies and sales finance), Vermont DFR banking and securities (29), and Oregon DFR financial services (10, covering investment advisers, collection agencies and pawnbrokers, student loan servicers, income share agreements and debt management).New Jersey and Vermont are the clearest illustration of why this is scoped the way it is. DOBI runs ONE numbered series across insurance, banking AND the Real Estate Commission; the real-estate bulletins are in neither corpus, because licensing brokers is neither insurance nor a financial institution. Vermont runs a SEPARATE series per division, so Banking Bulletin #50 and Insurance Bulletin #50 are different documents and each carries its series in its identifier.Nothing you already query changes. Existing insurance-bulletin results keep their citations, their actId values and their counts, and a search with no source filter now spans both. A search for insurance guidance will never be handed a credit-union bulletin: the two are separate categories, and each document’s source says which it is.This is not a claim to cover state banking law. These two states are served because their banking regulator is combined with their insurance regulator and publishes both series on one index, under numbers we can cite. Most states’ financial regulators publish no numbered series at all, so source=state_financial_bulletin returns nothing for the other 47 jurisdictions and no wider rollout is planned.state_financial_bulletin is served goodLawStatus: "unknown", deliberately. Neither regulator has rescinded anything we hold (0 of 35 in Arizona, 0 of 10 in Oregon), so there is no measured rescission signal to certify against. Arizona’s INSURANCE bulletins keep good_law on their own signal, which is 15 of 199 marked - same state, same corpusType, two publishers, two verdicts. Read unknown as “we make no claim”, not as “probably still current”.Every document carries an issuedDate from the publisher’s own index, and both sources refresh weekly alongside the rest of the state-guidance registry.
​
FinCEN Corporate Transparency Act beneficial-ownership guidance is live under corpusType=AGENCY_GUIDANCE

​
FinCEN Corporate Transparency Act beneficial-ownership guidance

Three new AGENCY_GUIDANCE sources, 145 documents across 220 retrieval passages, all from fincen.gov:This takes AGENCY_GUIDANCE to 40 named sources, 23,921 documents.Two of the three predate the current rule, and we serve them anyway, labeled. On August 11, 2026 FinCEN issued a final rule that permanently narrowed BOI reporting: entities created in the United States and their beneficial owners are exempt, and only certain companies formed under foreign law and registered to do business in a US State or tribal jurisdiction still report. The FAQs (issued 2023-2024) and the compliance guide (March 2025) both predate it, and FinCEN’s own pages say some of their content is outdated.We do not drop them, because they remain the authoritative FinCEN text and much of it is untouched by the scope change: the beneficial-owner and substantial-control definitions in FAQ category D read the same either way. Instead, every passage from these sources carries FinCEN’s own notice verbatim in currencyNote, with currencyYear: 2026. Disregard any statement in them that US companies or their beneficial owners must report BOI, that BOI must be reported for US persons, that US persons must update or correct FinCEN ID information, or that reports were due before April 25, 2025.For the current position, use source=fincen_boi_rule_qa. The 12 final-rule Q&As are current. The 10 interim-rule Q&As are served with actStatus: "superseded" and are excluded from good-law retrieval, because FinCEN’s own final-rule Q&A states the final rule adopted the interim changes permanently.These sources are not offered as a change-alert board; they refresh weekly with the rest of the agency-guidance registry.
​
CFR annual editions 1996-2026 are indexed under corpusType=CFR_ANNUAL

​
CFR annual editions, 1996 to 2026

corpusType=CFR_ANNUAL covers GPO’s officially published annual editions of the Code of Federal Regulations: 31 editions across all 49 titles. The corpus is indexed and counted by /statutes/coverage; the corpusType filter itself ships with the next API deploy.Read the headline number as section VERSIONS, not sections. A section is stored once per distinct run of text, not once per edition. A section whose wording never changed between 1996 and 2026 is ONE record whose editionsObserved lists every year it was observed in, rather than thirty-one near-identical copies.editionsObserved is an enumeration, never a range. A version observed in 1998, 1999 and 2003 lists exactly those three years. It does not imply 2000 to 2002, and a date falling in an unobserved gap is answered with asOf.isBounded: false rather than served from the nearest year we happen to hold.Citation years are each volume’s own <REVISED> year, not the folder GPO filed it under. GPO reprints unchanged volumes into later folders without re-dating them, so only the date inside the volume is authoritative. That is why 167 versions in title 26 (parts 502, 503 and 509) carry 1990 while sitting in the 1997 folder.This corpus is deliberately not offered as a change-alert board. The existing cfr board already carries the daily change signal for the same regulations, so a second board would fire once a year on a whole edition and repeat what subscribers had already been told the day each change happened.CFR_ANNUAL and the asOf parameter are both point-in-time routes and are not interchangeable. The corpus serves an edition as published; asOf=YYYY-MM-DD on GET /us/statutes/section/{actId}/body reconstructs a section from observed change history against the current CFR.
​
New corpusType: FEDERAL_REGISTER_NOTICE, a curated 17,612-notice slice of the Federal Register notice series

​
FEDERAL_REGISTER_NOTICE, and what it deliberately does not hold

Federal Register notices, 1994 to present, under the new corpusType=FEDERAL_REGISTER_NOTICE. 17,612 documents, 128,564 retrieval passages.🔴 This is a curated slice, not the notice series. The Federal Register published 766,073 notices in this period and we hold 2.3% of them. Saying so is the point: a corpus is worth what its limits are known to be, and a token that looked like complete notice coverage would be wrong in the one way that matters.What is held is the four classes of notice that act on a rulemaking:
  • withdrawal of a proposed rule,
  • extension or reopening of a comment period,
  • publication of significant agency guidance,
  • announcement of negotiated rulemaking.
Every point names the rule that admitted it in frNoticeClass, so the scope is readable from the data rather than only from this page.What is not held is everything else, which is mostly Paperwork Reduction Act information collections, meeting notices, permits and applications, antidumping proceedings and Privacy Act system-of-records notices. None of those states or changes a legal obligation. The 748,461 excluded documents are each recorded by Federal Register document number in our candidate ledger, so the scope can be widened later without re-crawling the series.A separate token from FEDERAL_REGISTER on purpose. Rules and proposed rules are ingested whole. Folding notices in would attach this partial-scope statement to 5.5M points that are not partial, which is a false claim in the other direction.⚠️ Notices carry an empty CFR cross-reference graph. crossReferencesCfr and frTopics are 0% and frRegulationIdNumbers is 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.amendmentYears is empty by design and goodLawStatus is unknown: a notice amends nothing, and the publisher prints no amendment credit for one, so there is no claim to make either way.The token takes no source filter. It has one internal source, so the value would be an exact synonym for the corpusType itself.
​
NLRB Board decisions join AGENCY_ADJUDICATION: 2,762 decided cases on a bounded 15-year window, with currency deliberately not asserted

​
NLRB Board decisions

The unfair-labor-practice and representation cases decided by the five-member National Labor Relations Board itself, which bind later Board panels and the administrative law judges beneath them. 2,762 decisions, 2012 to date, under source=nlrb_board_decision.This is a bounded 15-year window of a 90-year, 66,612-decision archive. That is a scope decision, not an omission: the window covers essentially every decision a practicing labor lawyer searches day to day, and the full backfill is tracked as a named follow-up.Two citation forms, both served, neither normalized into the other. A Board decision is cited NNN NLRB No. N as a slip opinion and NNN NLRB NNN once the bound volume closes and assigns it a page. These are different facts rather than variants, because the page number does not exist until the volume closes: 57.45% carry a slipOpinionNumber, 42.55% a pageNumber, and which one a decision carries is itself the record of where it sits in its own lifecycle. 56.07% carry the Board’s own notice that the opinion “is subject to formal revision before publication in the bound volumes”.Every decision carries caseName, caseNumber (the Board’s own docket, e.g. 28-CA-230115), volume and issuedDate at 100%, plus the source PDF and extracted text in R2.

​
goodLawStatus is unknown, and that is the point

Board doctrine swings hard with Board composition, and the NLRB publishes no overruling or supersession signal for its decisions. Its CITENET Classified Index is a subject digest rather than a citator: it records an overruling only as editorial prose on the decision doing the overruling, with no reverse edge on the decision that was overruled. Measured against the landed corpus, no decision carries a dead status.So every decision is served as the Board issued it, with goodLawStatus: "unknown" and a coverageNote stating the limit. Do not read that as “still good law”. It means we can show you faithfully what the Board decided and we will not assert whether a later Board has since overruled it.Unpublished Board decisions are summary dispositions with no independent precedential value, are published as a separate series, and are out of scope by design.
​
SEC Commission opinions join AGENCY_ADJUDICATION: 5,790 adjudicatory rulings, 1996 to date, with the File-Number proceeding graph

​
SEC Commission opinions and adjudicatory orders

The Securities and Exchange Commission’s own adjudicated rulings: appeals from an administrative law judge, review of a self-regulatory organization’s disciplinary action (FINRA, an exchange), and applications for review. 5,790 documents, 1996-04-05 to 2026-09-01, under source=sec_commission_opinion.These bind the Commission’s own later adjudications and the ALJs beneath it, and they are the securities bar’s primary source on what the SEC believes the securities laws mean in a contested posture. Cited by release number: Exchange Act Release No. 34-106254. One action issued under several securities statutes at once carries every release number it was given in citableAs.The File-Number proceeding graph. Every document carries its Administrative Proceeding File Number in fileNumber (99.6%; 112 documents name more than one, because the Commission consolidates proceedings). Documents sharing a File Number are ONE proceeding, and they are linked in date order through supersedes / supersededBy - 2,353 proceedings, 65.4% of them holding more than one document, 3,641 links across 86.7% of the series.🔴 Those links record a SEQUENCE, not a currency verdict, and no SEC document is served with a dead actStatus because of one. A docket says a later document exists; it does not say the earlier one stopped being good law. This series is overwhelmingly interlocutory - 839 orders to show cause, 820 orders extending time to issue a decision, 227 orders scheduling briefs, against 901 opinions of the Commission - so a later document routinely affirms the earlier one, schedules briefing, or merely notices an initial decision as final. goodLawStatus is unknown for the series.What is not here, stated rather than implied. ALJ initial decisions are deliberately not included: an initial decision is what one judge concluded before the Commission acted, and the SEC publishes no supersession field, so serving one as current authority would misstate its weight. Where an opinion names the initial decision it reviews, the citation is carried in relatedCitations and the referenced document is not in the corpus - that happens on 3.3% of the series. The 18,637-item settled-order and Fair Fund docket is also excluded, because a consent order reciting allegations is not a decision holding them proven. Federal court review under 15 U.S.C. 78y is not published on the SEC’s adjudication listings and is not reflected here. Every section states these bounds in its own coverageNote.
​
BIA and Attorney General precedent decisions join AGENCY_ADJUDICATION: the full I&N Dec. reporter, 3,292 decisions

​
The I&N Dec. reporter, under AGENCY_ADJUDICATION

The Board of Immigration Appeals’ published precedent decisions, the Attorney General’s decisions reviewing the Board, and the legacy INS and USCIS administrative decisions the same reporter carries. 3,292 decisions across volumes 8 to 30, 1958 to 2026.A published Board decision binds every immigration judge and every DHS officer nationwide under 8 C.F.R. 1003.1(g) unless and until it is overruled. This is the operative body of immigration law between the statute and the circuit courts, and it is not in any general case-law database.Three sources under the existing token, filterable with source:decidedBy carries the deciding authority verbatim as EOIR prints it, across 37 distinct authorities, and a decision the Board decided and the Attorney General then reviewed carries both.The overruling graph. The Board states its own effect on earlier decisions in the holdings prose beside each entry. 343 of those statements resolve into supersedes / supersededBy links, with the publisher’s own verb in dispositionVerb and the sentence itself in history. overruled_in_part is stored distinctly and does not carry a dead actStatus. 70 decisions are served dead on this evidence; the corpus as a whole stays goodLawStatus: "unknown".What is not here, stated rather than implied. Volumes 1 to 7 are not published on EOIR’s precedent page. Holdings prose begins at volume 19, so an overruling of a pre-1984 decision by a pre-1984 decision is invisible. A court of appeals can reject a precedent for its own circuit while it remains in force elsewhere, and EOIR publishes nothing about that. Unpublished single-member Board orders are not precedent and are out of scope. Every point states these in its own coverageNote.
​
CFPB compliance guidance joins AGENCY_GUIDANCE: 16 circulars and 95 supervisory-guidance documents

​
CFPB compliance guidance, under AGENCY_GUIDANCE

The other half of the Bureau’s output, alongside the enforcement actions below. 111 documents across two new sources:
  • cfpb_circular (16) - Consumer Financial Protection Circulars, cited Circular 2023-01. These are the Bureau telling every enforcer and every firm how it reads a statute, so they are the most broadly applicable thing it publishes.
  • cfpb_supervisory_guidance (95) - bulletins, policy statements, interpretive rules and rescissions.
This completes a shelf that was missing its most-read agency: we already carried OCC bulletins and interpretive letters, FDIC FILs, Federal Reserve SR letters and FinCEN guidance, and nothing from the one federal consumer-finance regulator a bank compliance officer reads every week.The Bureau’s enforcement actions are NOT here. They are adjudications and live under corpusType=AGENCY_ADJUDICATION: an enforcement action binds one named respondent, while a circular addresses the whole market. Folding them together would make an AGENCY_GUIDANCE search mean two different things.Two things worth knowing about coverage. The two series have opposite shapes - a circular publishes its full text on its page, while a supervisory-guidance page is a landing stub whose real document is the linked PDF - and both are served as full text regardless. And Circular 2023-03 is deliberately absent: the Bureau archived it, publishes no text for it, and leaves only a third-party mirror, which we do not source from.goodLawStatus is unknown on both sources. Neither publishes a structured rescission signal, so we do not claim currency we cannot verify.
​
CFPB enforcement actions join AGENCY_ADJUDICATION: 386 actions, and the first measured supersession signal in that corpusType

​
CFPB enforcement actions, under AGENCY_ADJUDICATION

386 Consumer Financial Protection Bureau enforcement actions, filterable with source=cfpb_enforcement_action. This closes a hole in a shelf we already had: we carry OCC bulletins, FDIC FILs, Federal Reserve SR letters and FinCEN guidance, and carried nothing from the one federal consumer-finance regulator whose output a bank compliance officer reads every week.Both forums are included and are separable on the new forum field. 225 actions are administrative proceedings carried on a Bureau File No. such as 2024-CFPB-0002; 162 are civil actions the Bureau files in federal district court, which carry the court’s own number in caseNumber and no Bureau citation, because that identifier belongs to the court rather than to the Bureau. One action carries both labels.The unit is the ACTION, not the filing. A consent order and the stipulation consenting to it are one record, because a stipulation is not a separate legal instrument from the order it stipulates to and splitting them makes every search return two half-answers. relatedDocuments names every filing whose text a record carries - 993 documents in total, a mean of 2.6 per action and a maximum of 17. Twelve actions are served with the Bureau’s own summary alone, because it posts no document for them.

​
185 of the 386 are excluded from ordinary search, on purpose

The Bureau publishes its own three-value status enum and defines it publicly. An action it marks Expired/Terminated/Dismissed - meaning the matter “did not result in, or is no longer subject to, ongoing compliance obligations” - is served with a dead actStatus and is therefore outside ordinary search results. That is correct for an order that no longer binds anyone, and it is close to half the set, so it is stated here rather than left to be discovered.That status bundles three different endings and the Bureau does not separate them: a matter it lost or withdrew is not the same event as an order that ran its term. Every record therefore carries the Bureau’s own status string verbatim in history, so you can see the ambiguity we inherited rather than only our reading of it.

​
The first AGENCY_ADJUDICATION source with a real goodLawStatus

No publisher in this corpusType ships a structured overruling field - an adjudication is superseded by another decision, and the document itself never changes when it stops being good law. MSPB and OLC records therefore carry goodLawStatus: "unknown". The CFPB is the exception: its status enum is defined, complete and published, so this source’s goodLawStatus is a real verdict. MSPB and OLC are unchanged.New fields on this corpus: forum, court and relatedDocuments.
​
New corpusType: STATUTE_COMPILATION, 80,276 sections across 2,680 GovInfo compilations

​
New corpusType: STATUTE_COMPILATION

A new federal corpus token for a public law’s own text as amended through a stated later public law - GovInfo’s official compilations. 80,276 sections across 2,680 compilations, sourced from GovInfo’s COMPS collection.This fills a real citation gap. A section like “Section 210 of the Public Utility Regulatory Policies Act” has no USC address, because it was never codified into a positive-law title; STATUTE_COMPILATION is where it resolves.
Distinct from both of its neighbors, deliberately. USC is the codified, subject-organized arrangement - only for provisions that were codified at all. SESSION_LAW (the Statutes at Large) holds the same acts as enacted, a frozen historical record that carries actStatus: "enacted" and goodLawStatus: "unknown" by design. STATUTE_COMPILATION makes the opposite claim: “this is the act as it reads today, amended through Public Law 117-58.” Folding it into either neighbor would mean a customer reading a decades-old snapshot as current law.Currency is stated per document, and read it before relying on the text. Each section carries currentThrough (e.g. Public Law 117-58), naming exactly which amending law the compilation reflects. Across the collection that ranges from Public Law 96-308 (1980) to Public Law 119-75 (2026) - a 45-year span in the same corpus. 2.6% of compilations state no currency at all, and those carry compilationCurrencyState: "unstated" with currentThrough: null rather than a guessed date. STATUTE_COMPILATION does not carry a goodLawStatus certification.The join into USC is captured, not resolved. Where GovInfo publishes a structured cross-reference into the Code (16 U.S.C. 2603, for example), it’s served as crossReferencesUsc, so a caller can look up the codified version of a compiled provision directly.Full text is served by GET /statutes/section/{actId}/body, exactly like every other corpus.
​
New corpusType: AGENCY_ADJUDICATION, with 5,455 MSPB decisions and 1,371 OLC opinions

​
New corpusType: AGENCY_ADJUDICATION

A new federal corpus token for administrative adjudication: matters decided by a federal body through a formal process, carrying precedential weight on its later decisions. It is deliberately separate from AGENCY_GUIDANCE, which is interpretive rather than adjudicative, so a corpusType=AGENCY_GUIDANCE search keeps meaning one thing.1,371 OLC opinions, cited NN Op. O.L.C. NNN, filterable with source=olc_opinion. OLC opinions are the executive branch’s own controlling legal advice: an opinion binds executive agencies unless the Attorney General or the President overrides it. Sourced entirely from the Department’s own official bound volumes at justice.gov.
Each opinion carries its volume and starting page, the canonical citation, the requester it was addressed to, the signing official, its date, and the statutes and regulations it cites. Full text is served by GET /statutes/section/{actId}/body.Coverage ends with volume 44 (2020). Opinions issued in 2021 and later have not yet been published in a bound volume and are not included; every section states this in its own coverageNote. Supplemental volume 1 (NN Op. O.L.C. Supp. NNN) adds opinions written 1933-1977 and published later.Currency. OLC opinions default to goodLawStatus: "unknown", because an opinion is superseded by a later opinion rather than amended in place and the publisher ships no structured overruling field. Where OLC printed an editor’s note that an opinion was withdrawn, superseded or reversed, that note is served in currencyNote and the section carries actStatus: "superseded".

​
Merit Systems Protection Board decisions

5,455 MSPB decisions, November 2018 to September 2026, sourced from the Board’s own published JSON manifests at mspb.gov. The MSPB adjudicates federal-employee appeals: removals, suspensions, demotions, reduction-in-force challenges, Individual Right of Action whistleblower claims under the Whistleblower Protection Act, and USERRA cases.The two series are separately filterable, and that distinction is load bearing. 106 precedential decisions, cited YYYY MSPB N, bind administrative judges across the Board’s regional offices. 5,349 nonprecedential orders do not: the Board states in the orders themselves that they “do not add significantly to the body of MSPB case law”. Filter with source=mspb_precedential / source=mspb_nonprecedential, or on the precedential boolean carried by every decision.
Each decision carries its docket number, the respondent federal agency, the appellant caption, its issue date, and the Title 5 statutes and regulations it cites (5 U.S.C. 7513, 2302, 1221 and 5 C.F.R. 1201 are the common ones). Only the precedential series has a citation. A nonprecedential order is identified by its docket number, which is the only identifier the Board assigns it; nothing is synthesized to fill the gap.Currency. MSPB decisions default to goodLawStatus: "unknown". A decision is superseded by a later decision on the same docket rather than amended in place, and the Board publishes no supersession or reconsideration field. Where two decisions share a docket they are linked with supersedes / supersededBy, which records the sequence only: the Board does not say whether the later decision affirmed, modified or reversed the earlier one, so neither do we.One gap, stated plainly. MSPB decisions are reviewable by the Federal Circuit under 5 U.S.C. 7703, and a reversed decision is no longer good law. The Board publishes nothing about this, and a decision cannot mention its own later appeal, so a decision reversed on appeal is not marked as such here. Every section says so in its currencyNote.
​
State regulations reach all 52 jurisdictions; corpus passes 4.1M sections

​
Regulations for every US jurisdiction

State regulations now cover all 52 jurisdictions, 50 states plus DC and Puerto Rico: about 1.51M rules, up from 48 states and about 1.22M. No jurisdiction is now without a regulations corpus.Three are entirely new, each from the jurisdiction’s own publisher:DC’s online edition is the official copy by its own rule, 1 DCMR 300.7. Puerto Rico’s reglamentos are published in Spanish, the language they are enacted in, and language_code is set accordingly; free access is required by LPAU 2.10, 3 L.P.R.A. 9620.

​
Ten jurisdictions re-sourced onto their own government publisher

Each of these now comes from the jurisdiction’s own official publisher, with amendment history and a resolvable sourceUrl on every section:Oklahoma, Pennsylvania and Massachusetts are not yet re-sourced. Each is documented as such rather than partially ingested.

​
Coverage corrections

New Mexico was serving 9 of the 22 NMAC titles. All 22 are now live, at 3,116 parts and 36,310 sections, up from 1,074 parts.Colorado was serving 610 of the 1,155 CCR rules its publisher lists. All 1,155 are now live, including Medicaid, workers’ compensation, taxation, water quality and election rules, which were absent entirely. Colorado amendment coverage goes from none to 99%.

​
Amendment history across eight more states

lastAmendedYear and amendmentYears are now populated for Idaho, Illinois, Minnesota, Ohio, South Carolina, Virginia, Washington and Wisconsin, where the publisher records them. Roughly 350,000 sections gained derived amendment history. Texas and Iowa are unchanged: their publishers do not record an amendment for every rule, and the API reports null rather than inferring one.

​
Corpus total

Over 4.1M distinct sections, indexed as over 11.9M retrieval passages, up from 3.9M and 11.1M.
Rhode Island and Colorado are currently keyed one section per Part and per rule respectively, rather than per subsection. Citation lookup and search are unaffected; a finer-grained key for both is in progress.
​
New corpusType: SESSION_LAW, the US Statutes at Large

​
Read a law as Congress enacted it, not as it was codified

A fifteenth corpusType lands: SESSION_LAW, the US Statutes at Large. 5,256 public and private laws, 110,287 sections, 1999 to 2026.The US Code is a compilation. It reorganizes enacted text by subject and drops the structure the enacting Congress used, so a US Code section does not tell you what the act it came from actually said. SESSION_LAW does.
The join is the point. A US Code section’s publicLawCites resolve directly into this corpus, so you can go from codified text to enacting text in one hop. That join resolves for every US Code section we serve.
These are laws as enacted, a historical record rather than a statement of current law, so they carry actStatus: "enacted" and goodLawStatus: "unknown" by design. For what the law says today, use corpusType=USC.
​
Tennessee regulations added; state regulations reach 49 states

​
Tennessee regulations

Tennessee regulations are live: 11,608 rules, taking state regulations to 49 states, about 1.23M rules. Source is the Official Compilation Rules and Regulations (Tennessee Secretary of State).
Metadata coverage: 98.5% statutory authority, 98.4% amendment years and history, 97.6% effective date. priorEffectiveDates is populated where the publisher records more than one.Watchable as a law-change board (corpusType=STATE_REGULATION, state=tn), refreshing monthly.
Section numbers are zero-padded as Tennessee cites them (0880-05-.01, not 0880-5-.01). Filter sectionNumber with the padded form.
​
State regulations reach 48 states, court rules reach 46 jurisdictions

​
Regulations and court rules move onto official sources

State regulations now cover 48 states, about 1.22M rules, with Utah and Kansas joining the corpus.State court rules now cover 46 jurisdictions, about 47.6K rules, with New Mexico (3,462 rules across 47 rule sets), New Jersey and Kentucky added. New Mexico, New Jersey and Kentucky are also now watchable as law-change boards.Counts in your own integration are worth re-reading from GET /statutes/coverage rather than carried forward: a re-source can move a jurisdiction’s count in either direction.
​
GET /boards returned only the first 200 sources

​
Fixed: /boards silently truncated its own catalogue

GET /boards defaulted to limit=200 and returned no total, so with 216 watchable sources every client was reading a truncated list with nothing in the response to say so. Sources sorting last were invisible, including usc, the US Code.The response now carries a total, and paging is explicit. If you built a source picker from this endpoint, re-fetch it: sources you never saw are not new, they were always there.GET /watches also gained filters, so a large watch list can be narrowed server-side instead of client-side.
​
Five new endpoints: change history, citators, definitions, neighbors, cross-state

​
Five questions about a section you previously had to answer yourself

Each takes an actId and answers one question a section lookup could not./changes is the one that removes a workaround. Every change our refreshes have observed to a section: when it was added, each time its text was replaced, and whether it has since been removed. Answering that used to require creating a board watch and waiting, because a watch can only show changes that postdate it. This reads the captured history directly./related returns the sections a reader would actually turn to, not the numerically adjacent row, which in most codes belongs to an unrelated part.Not-found and failed lookups are not charged. An empty result for a section that exists is the answer, and is charged.
​
Every jurisdiction now reports when we last pulled it

​
Coverage answers “how old is this?” without you having to assume

GET /statutes/coverage gained a retrieval array covering every (jurisdiction, corpusType) pair, maintained ones included:
  • lastRetrievedAt - when we last pulled that body of law from its publisher and finished cleanly
  • cadence - how often we do
  • status - current, stale, failing, or never_retrieved
Read it alongside a section count so the count carries an age. Note this is when we last checked, not when the law last changed, and not a claim that the text is current as of that date. Where a publisher states its own currency, that rides on the section itself as currentThrough.The freshness array is the companion: it names only the pairs that are not being updated, each with a plain-language reason, so a large count can be read as maintained rather than merely large.
​
Watch a citation instead of a whole source

​
Scoped alerts: watch a citation, not a source

POST /watches and PATCH /watches/{watchId} now accept an optional scope so an alert can follow one title, chapter, part or section instead of an entire corpus. Watching cfr meant every change to every federal regulation; now it can mean 21 CFR part 314.
A scoped alert stays quiet when nothing inside its scope changed, rather than sending an empty “the source was updated” on every refresh. Deliveries carry the scope, changes contains only what falls inside it, and GET /watches/{watchId}/changes applies the same filter so the read-back agrees with what we sent.GET /boards gained scopable (whether that source accepts a scope) and scopeBrowseCorpus (the corpusType to pass to GET /statutes/divisions to enumerate its titles and parts). Sources whose documents are numbered items rather than citations, such as the Federal Register, are not scopable.Nothing changes for existing alerts. An omitted scope means the whole source, exactly as before, and every watch created before today keeps behaving identically. See Alerts.
​
Nine new response fields: version history, currency and amendment provenance

​
Follow a superseded section to the text that replaced it

Sections now return the provenance the corpus has always stored. Nine fields, all additive, all optional: nothing in your existing integration changes.The one that changes what you can build: supersededBy. A section with actStatus=superseded used to be a dead end. You could tell it was not current law, but nothing in the response pointed at what replaced it. Now:
Point-in-time research is one hop in either direction.currentThrough is set on every US Code section, so a currency check no longer depends on the publisher having written prose. ruleSet matters more than it looks for court rules: rule numbers repeat across families, and a Civil Rule 12 and a Criminal Rule 12 both exist in most states, so this is what tells them apart.Every field is omitted when the publisher does not record it, rather than returned empty. Absence means the source is silent, never that we did not look.
​
State court rules expand from 10 to 43 jurisdictions

​
Court rules: 10 jurisdictions to 43

corpusType=STATE_RULES now covers 43 jurisdictions and about 43,500 rules, up from 10 states and roughly 13,000. Added: AK, AL, CT, DC, DE, GA, HI, IA, ID, IN, KS, LA, MA, ME, MI, MS, MT, NC, ND, NE, NH, OH, OR, PR, RI, SC, TN, UT, VA, WA, WI, WV, WY.Every jurisdiction is ingested from its own court system or legislature, New York included, which is served from nycourts.gov. Coverage spans civil, criminal, appellate, evidence, professional conduct and local rules; the District of Columbia alone carries 21 separately numbered rule families across its Superior Court divisions.Rules are refreshed monthly and carry actStatus, goodLawStatus, amendmentYears, amendmentsCount and lastAmendedYear where the publisher prints them. Amendment coverage varies by source and is a property of the publisher, not the ingest: some courts print a dated credit under every rule, others print none at all.A state absent from this list is usually not a backlog item. Some publish no consolidated rules text from a government source; several codify their rules into the statute book, where they are already served under corpusType=STATE (Nebraska’s evidence rules are a good example, sitting in the statutes as Neb. Rev. Stat. 27-401 and deliberately not duplicated here); and at least one restricts automated access by published court policy, which we honor. Call GET /statutes/coverage for the live per-jurisdiction picture rather than assuming a state is missing.
​
State insurance regulator guidance expands to 40 jurisdictions

​
State agency guidance now covers 40 jurisdictions

STATE_AGENCY_GUIDANCE holds each state insurance regulator’s own bulletins and circular letters: the interpretive and enforcement guidance that sits between the statute and the regulation. It is the state-level analog of federal AGENCY_GUIDANCE, and it is queryable with corpusType=STATE_AGENCY_GUIDANCE paired with a state.Coverage is now 8,986 documents across 40 jurisdictions, up from 5 states. Per-jurisdiction counts are in API Coverage, and GET /statutes/coverage always returns the live figures.Every document is sourced from the issuing regulator’s own government site. Where a state’s official index is itself incomplete, the confirmed subset ships and the gap is recorded.A small number of states publish no browsable bulletin archive at all; those are documented as such rather than partially ingested.
​
Search ranks current law above superseded and rescinded versions

​
Dead law no longer competes with live law on relevance alone

Search now ranks a provision’s current text above superseded or rescinded versions of the same provision.Eight actStatus values carried no ranking penalty, so 44,199 passages were ordered as though they were current law: 29,901 superseded, 12,041 rescinded, and smaller counts of revoked, expired, deleted, recodified, unconstitutional and rejected. The effect was largest wherever a corpus keeps prior versions as their own addressable records, so a court rule with a version history or a rescinded state regulation was where you were most likely to see it. A superseded version is near-identical to the live one by construction, which is precisely what makes a relevance score unable to separate the two.Dead law is still returned and still searchable. That is deliberate: research routinely needs the text that was in force when the events happened, so a repealed section is a legitimate result rather than noise. It now ranks below live law of comparable relevance instead of being hidden, and every result still carries actStatus and goodLawStatus, so you can filter it out or surface it on purpose.No request or response shape changed, and there is nothing to update on your side. If you were relying on the previous ordering to reach prior versions, filter on actStatus explicitly.
​
NLRB General Counsel and Advice Memoranda now covered

​
NLRB joins Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the National Labor Relations Board’s two guidance vehicles. General Counsel Memoranda (300 memos, the full archive back to 1973) are citable as NLRB General Counsel Memorandum GC 22-04; a rescinded memo carries actStatus=rescinded and a supersededBy pointer to the memo that rescinded it, and the rescinding memo itself lists every memo it rescinded. Division of Advice Memoranda (339 memos) are citable as NLRB Division of Advice Memorandum, Case No. 13-CA-182757 (Savage Fueling); this is a rolling 10-year window, not the full ~2,500-2,600-document archive back to 1969, which remains a future addition. Filter to either with source=nlrb_gc_memo or source=nlrb_advice_memo.
​
New: Law Change Alerts, with section-level diffs

​
Subscribe to a jurisdiction or corpus and get notified when it changes

A new, free capability: GET /boards lists every source you can watch (a jurisdiction’s statutes, a corpus like the Federal Register or the eCFR), and POST /watches subscribes you to one on a webhook, an email address, or both. When that source refreshes with real changes, you get a signed webhook delivery (X-Vaquill-Event: board.updated, X-Vaquill-Signature when a secret is set) and/or an email naming the exact sections that were added, amended, or removed — not just a count.GET /watches/{watchId}/changes lists that same detail on demand, scoped to your watch’s own lifetime. For any amended section where hasDiff is true, GET /watches/{watchId}/changes/{changeId}/diff returns the section’s text before and after that specific change, so you can show exactly what moved rather than sending someone back to reread the whole section.PATCH /watches/{watchId} pauses, resumes, or rotates a destination. POST /watches/{watchId}/test fires a one-off test delivery before you rely on it. GET /watches/{watchId}/deliveries is webhook delivery history. See the Law Change Alerts guide for the full walkthrough, payload shape, and signature verification.
​
FCC, FERC, and DOE interpretive guidance now covered

​
FCC, FERC, and DOE join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers three federal agencies’ interpretive-guidance vehicles. FCC Declaratory Rulings (169 rulings, discovered from the FCC’s own EDOCS “Commission Documents” listing, citable as “DA-26-680: (07/07/26)”, source=fcc_declaratory_ruling) resolve a petitioner’s request for the Commission to state its interpretation of a statute, rule, or order in a specific factual context. FERC Policy Statements (45 statements back to 1996, citable by the real practitioner ”___ FERC ¶ _____” Reports form when a statement has been paginated into a bound volume, falling back to “Docket No. (): ” otherwise, source=ferc_policy_statement) are FERC’s generally-applicable interpretive vehicle, deliberately narrower than the docket-based eLibrary universe of case-specific orders. DOE Appliance Standards Guidance and FAQs (60 final, non-archived documents, citable as “DOE Appliance Standards Guidance: ()”, source=doe_appliance_guidance) interpret the Energy Policy and Conservation Act program DOE administers; documents the source itself marks “draft” are excluded, never ingested as if they were settled guidance. See Coverage for the updated count.
​
BIS and DDTC export control guidance now covered

​
BIS Advisory Opinions and DDTC export control guidance join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers export control interpretive guidance from the two federal agencies that administer nearly all U.S. export controls. BIS (Bureau of Industry and Security, U.S. Department of Commerce) Advisory Opinions interpreting the Export Administration Regulations (33 opinions, 2003-2023, citable as “BIS Advisory Opinion, 09/25/23: Export guidance and BIS interpretation of ‘gene’ related to ECCN 1C353”, source=bis_advisory_opinion) are a curated, agency-discretion archive. DDTC (Directorate of Defense Trade Controls, U.S. Department of State) Commodity Jurisdiction Determinations (5,879 ITAR/USML-vs-EAR rulings, 2010-present, citable as “DDTC Commodity Jurisdiction Determination: (), ”, source=ddtc_commodity_jurisdiction) and DDTC Policy Guidance (7 named documents including Guidelines for Preparing Agreements and the ITAR Compliance Program Guidelines and Risk Matrix, source=ddtc_guidance) are both sourced directly from DDTC’s own public portal. See Coverage for the updated count.
​
CPSC Advisory Opinions and Small Entity Compliance Guides now covered

​
CPSC Advisory Opinions and Small Entity Compliance Guides join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers two Consumer Product Safety Commission sources. CPSC Office of General Counsel Advisory Opinions (139 opinions with a legible text layer, 1973-2012, citable as “CPSC OGC Advisory Opinion No. 82”, source=cpsc_advisory_opinion) are a historical interpretive archive: the program has not issued a new public opinion in over a decade, so this is not presented as current guidance. CPSC Small Entity Compliance Guides (34 actively maintained per-rule guides required under SBREFA, citable as “CPSC Small Entity Compliance Guide: Safety Standard for Neck Floats (90 FR 58096)”, source=cpsc_secg) track new safety-standard final rules as CPSC publishes them. CPSC’s recall notices (product recalls, safety alerts) are news/safety-communication content, not legal-interpretive text, and remain out of scope. See Coverage for the updated count.
​
FinCEN Administrative Rulings and Guidance now covered

​
FinCEN Administrative Rulings and Guidance join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers FinCEN’s BSA/AML interpretive material. FinCEN Administrative Rulings (86 rulings back to 1988, each with precedential value, citable as “FinCEN Ruling FIN-2009-R001”, source=fincen_ruling) and FinCEN Guidance (127 alerts, advisories, notices, bulletins, and fact sheets, citable as “FinCEN Guidance FIN-2019-G001”, source=fincen_guidance). Corporate Transparency Act beneficial-ownership guidance and Anti-Money Laundering Act of 2020 material are not yet covered (they live on separate FinCEN hub pages). See Coverage for the updated count.
​
Federal Reserve / OCC / FDIC banking guidance now covered

​
Federal Reserve, OCC, and FDIC banking guidance join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers four banking-regulator sources: Federal Reserve Supervision and Regulation (SR) / Consumer Affairs (CA) Letters (336 letters back to 1990, citable as “SR 21-4 / CA 21-2”, source=frb_sr_letter), OCC Bulletins (606 back to 1994, source=occ_bulletin), OCC Interpretive Letters (448 reachable online from 1996-present, citable as “OCC Interpretive Letter No. 1188”, source=occ_interpretive_letter), and FDIC Financial Institution Letters (2,313 back to 1994, spanning both FDIC’s active and inactive FIL indexes so superseded letters stay queryable with act_status=superseded rather than silently dropped, source=fdic_fil). See Coverage for the updated count.
​
CFTC Staff Letters now covered

​
CFTC Staff Letters join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers CFTC Staff Letters (942 no-action, exemptive, interpretative, and advisory letters, 2008-present), citable as “CFTC Letter No. 26-17”. Each letter carries the regulation parts and topic tags CFTC itself publishes, filterable via the standard search fields. Filter to just this source with source=cftc_staff_letter. Pre-2008 letters are not yet covered (CFTC’s own site does not browse them separately today). See Coverage for the updated count.
​
FAR/DFARS-native citation resolution, plus DFARS PGI

​
FAR and DFARS are now natively citable, and DFARS PGI joins Agency guidance

The Federal Acquisition Regulation (FAR, 48 C.F.R. ch. 1) and Defense FAR Supplement (DFARS, ch. 2) were already in the corpus as part of CFR (Title 48). /statutes/resolve and citation-shaped search queries now accept the forms practitioners actually use — FAR 52.219-14, DFARS 252.225-7001, with or without periods or a pinpoint subsection — and resolve them straight to the existing CFR record. Filter to just FAR or just DFARS within corpusType=CFR with source=far / source=dfars.corpusType=AGENCY_GUIDANCE now also covers DFARS PGI (Procedures, Guidance, and Information), DoD’s own non-codified procedural companion to the DFARS, citable as “PGI 225.370”. Filter to just it with source=dfars_pgi. See Coverage for the updated counts and Corpus Types for the full source list.
​
USPTO MPEP and TMEP now covered

​
USPTO patent and trademark manuals join Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the USPTO Manual of Patent Examining Procedure (MPEP, 2,024 sections across all 29 chapters, citable as “MPEP § 2106”) and the Trademark Manual of Examining Procedure (TMEP, 2,109 sections across all 19 chapters, citable as “TMEP § 1207.01”). Filter to just one with source=mpep or source=tmep. See Coverage for the updated count.
​
DOJ Justice Manual now covered

​
DOJ Justice Manual joins Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the DOJ Justice Manual (1,548 sections across all 9 Titles), citable by section as “Justice Manual § 9-27.420”. Filter to just it with source=doj_justice_manual. See Coverage for the updated count.
​
USCIS Policy Manual now covered

​
USCIS Policy Manual joins Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the USCIS Policy Manual (456 chapters across all 12 volumes), citable by chapter as “USCIS Policy Manual, Vol. 12, Pt. D, Ch. 3”. Filter to just it with source=uscis_policy_manual. See Coverage for the updated count.
​
Rules of the Supreme Court of the United States now covered

​
Supreme Court Rules join Federal Rules of Procedure

/statutes/resolve and corpusType=FEDERAL_RULES search now cover the Rules of the Supreme Court of the United States (Sup. Ct. R. N), ingested from the official supremecourt.gov committee print alongside the five Federal Rules of Procedure sets already served. See Coverage for the updated count.
​
New agency-guidance sources: Merger Guidelines, DOJ Leniency Policy, Copyright Circulars

​
New sources under Agency guidance

corpusType=AGENCY_GUIDANCE now also covers the DOJ & FTC Merger Guidelines (2023), the DOJ Antitrust Division Leniency Policy, and US Copyright Office Circulars, each independently filterable and citation-labeled, folded into the same AGENCY_GUIDANCE token you already query. See Coverage for the updated count.
​
New endpoint: catalog the bodies of law

​
New endpoint: GET /statutes/laws

A catalog of the distinct bodies of law available, each a whole corpus within a jurisdiction (a state’s statutes, a state’s regulations, the US Code, the US Constitution, federal court rules, …). Every entry carries a stable key, its corpusType (pass straight to /statutes/search), a lawType (statutes, regulations, constitution, court rules, …), a human name, and a sectionCount.
  • GET /statutes/laws lists every jurisdiction.
  • GET /statutes/laws?jurisdiction=ca (or federal) scopes the list.
Use it to build a picker of what is queryable or to ground an LLM in the available bodies of law. To drill a state’s statutes into its individual codes (Penal, Civil, …), call GET /statutes/codes?state=XX. Free (an API key, rate-limited, no credits).
​
Citation resolver now covers constitutions, court rules, and annotated forms

​
Resolve constitutions, court rules, and more citation forms

GET /statutes/resolve (and citation-as-query search) now resolves several citation families it previously missed:
  • US and state constitutions: U.S. Const. art. I, § 8, U.S. Const. amend. XIV, § 1, U.S. Const. pmbl., Cal. Const. art. I, § 7.
  • Federal court rules: Fed. R. Civ. P. 12 (incl. a pincite like Fed. R. Civ. P. 12(b)(6)), Fed. R. Crim. P. 11, Fed. R. Evid. 702, Fed. R. App. P. 4, Fed. R. Bankr. P. 9011, and the compact FRCP 12 form.
  • Annotated / prefixed federal statutes: the annotated 42 U.S.C.A. § 1983 and 42 U.S.C.S. § 1983 forms, Title 42 U.S.C. § 1983, and a doubled §§.
This is additive and precision-focused: a Bluebook citation names one exact section, so it stays the most reliable way to reach or verify a section. See Resolve and verify a citation.
​
Scope a search to a chapter or part

​
Search within a hierarchy subtree

POST /statutes/search gains two filters that scope a search below a title or code, additive and optional:
  • chapter (USC/state) and part (CFR): confine results to one or more chapters or parts, e.g. titleNumber: 42, chapter: "21" or titleNumber: 17, part: "240". Both accept a single value or a list.
These are the search-side of the parent object on every result: take a hit’s parent and pass it straight back to search that section’s neighbors. Because a chapter number repeats across every title, chapter/part must be paired with titleNumber or code (which parent always carries); an unpaired one is rejected with 422. See Search within a container.
​
Search controls, navigable parent, amendment provenance, deeper paging

​
Control what search matches and returns

POST /statutes/search gains two request options, both additive:
  • matchType: any (default) is intent-ranked hybrid search. all requires every query term to appear; phrase requires the exact phrase. Use the stricter modes when you need precision over recall.
  • excerptChars: how much matching text each result carries in excerpt (default 500, range 100 to 4000). Raise it for more inline context.

​
Every result now points to its neighbors

Each result carries a parent object: the ready-made GET /statutes/divisions query that lists the section’s siblings (its containing chapter or part). Walk up the hierarchy from any hit without constructing the query yourself.

​
Amendment provenance on statute results

Statute results now surface amendment history already in the corpus: sourceCredit (the source’s credit line), amendmentYears, lastAmendedYear, amendmentsCount, publicLaws (e.g. ['Pub. L. 102-166']), and federalRegisterCitations. All are nullable and additive; existing fields are unchanged.

​
Page deeper

The offset cap on POST /statutes/search is raised from 40 to 70, so with the largest page size you can reach the entire reranked candidate pool. See Pagination.
​
Structured body: Markdown and a pincite subsection tree

​
/body can now return a structured subsection tree

GET /statutes/section/{actId}/body?structured=true adds two fields to the body response, at no extra cost:
  • markdown: the section rendered as nested Markdown lists, one level per subsection depth.
  • subsections: a tree parsed by the drafting convention ((a) subsection, (1) paragraph, (A) subparagraph, (i) clause). Each node carries its label, a pincite path (e.g. (b)(2)), a full pinpoint citation (e.g. 42 U.S.C. § 1983(b)(2)), its own text, and nested children.
Use it to address, deep-link, or quote a specific subsection. It is additive: the default /body response is unchanged. See Response Formats.
​
Citation resolver and hierarchy browse

​
Two new endpoints: resolve a citation, browse the hierarchy

  • GET /statutes/resolve (2 credits): resolve a Bluebook citation string (42 U.S.C. § 1983, 16 C.F.R. § 444.1, Del. Code Ann. tit. 13, § 1301) straight to the exact section, confirmed, with an official source link. A pinpoint subsection resolves to the parent section and is echoed in subsection. resolved reports whether the citation is real, so it is the one-call way to verify a citation an LLM produced. See the citation recipe.
  • GET /statutes/divisions (1 credit): walk the statutory hierarchy one level at a time, in statutory order. Pass the deepest level you know (corpusType, titleNumber, chapter/part, or a state code) and get the children below it, down to sections with their actId. Supports USC, CFR, STATE, and REGULATION. See the browse recipe.
Both are additive and use the existing API key and credit model.
​
Richer statute results: status, currency, hierarchy, and provenance

​
Statute results now carry status, currency, and fuller metadata

Every statute result (from POST /statutes/search, GET /statutes/section/{actId}, and POST /statutes/sections) now includes additional fields that were already in the corpus. This is additive: existing fields are unchanged, so no client change is required.
  • Status and currency: actStatus (in_force, repealed, renumbered, transferred, omitted, reserved, vacant, unconstitutional), a derived goodLawStatus (good_law / not_good_law / not_operative / unknown), currencyNote (the source’s “current through …” line), and renumberedTo / transferredTo pointers.
  • Fuller hierarchy: subchapter / subchapterName, part / partName, subpart / subpartName, and popularName (e.g. “Truth in Lending Act”).
  • Provenance: issueDate, crossReferencesCfr (parsed CFR references as a list of section strings, e.g. ['1.6011-4']), and govinfo granuleId / packageId.
goodLawStatus is conservative: jurisdictions without a reliable repeal signal report unknown rather than over-claiming that a section is current.
​
Regulations, coverage discovery, and pagination

​
Regulations, Federal Register, and agency guidance are now queryable

Three new corpusType tokens open up corpora that were already in the corpus but not filterable:
  • REGULATION: state administrative regulations (176K+ rules across 12 states).
  • FEDERAL_REGISTER: Federal Register agency rules, final and proposed (100K+ documents).
  • AGENCY_GUIDANCE: agency guidance, including SSA rulings and IRS notices, revenue rulings, and procedures.
Scope a search to any of them with the corpusType filter (pair REGULATION with a state). See Corpus Types.

​
New endpoint: GET /statutes/coverage

A self-describing coverage matrix. It returns a legend of every corpusType token (its meaning and whether it is federal or state scoped) plus, for each jurisdiction, the per-corpusType counts.Call it to learn exactly what is queryable where, then pass a token as corpusType. See Jurisdictions and State Codes and the coverage monitoring recipe.

​
GET /statutes/states now returns a per-corpusType breakdown

Each jurisdiction row now includes a corpora object, for example {"STATE": 176778, "REGULATION": 51142, "STATE_CONSTITUTION": 179}, so you can see which corpora a jurisdiction has without a separate call.

​
Pagination on POST /statutes/search

Search now accepts an offset parameter (0-40) alongside limit, and the response includes offset and a hasMore flag so you can page through results. total is the count returned on the current page. See Pagination.

​
Fix: STATE_RULES now returns results

The STATE_RULES (state court rules) corpus is now correctly served. Court rules are available for a handful of states, discoverable via /statutes/coverage.

​
New documentation

A full developer guide: Quickstart, Core Concepts, Pagination, Rate Limits, Best Practices, a Grounding LLMs guide, eight Recipes, and a Use Cases catalog.
Last modified on September 7, 2026