Resolve and verify a statute citation

What this gets you: a way to catch a fabricated or wrong citation before it ships, whether it came from an LLM, a user, or your own database - one call tells you if a citation is real, and hands you the verified text if it is. Given a Bluebook citation string like 42 U.S.C. § 1983, 16 C.F.R. § 444.1, or Del. Code Ann. tit. 13, § 1301, resolve it to the exact section and get back the section metadata with an official source link. One route does the whole job: /us/statutes/resolve parses the citation server-side and tells you whether it is real. Use the GET form for a single citation and the POST form for up to 50 at once, which is what you want when you are checking the citations in a document rather than looking one up.

Bluebook-formatted citations resolve most precisely. These all resolve, on either form of /us/statutes/resolve:

Federal statute: 42 U.S.C. § 1983 (annotated 42 U.S.C.A. § 1983 / 42 U.S.C.S. § 1983 and Title 42 U.S.C. § 1983 also resolve)
Federal regulation: 16 C.F.R. § 444.1
State statute: Del. Code Ann. tit. 13, § 1301
US Constitution: U.S. Const. art. I, § 8, U.S. Const. amend. XIV, § 1
State constitution: Cal. Const. art. I, § 7
Federal court rule: Fed. R. Civ. P. 12, Fed. R. Evid. 702
Federal Acquisition Regulation / DFARS: FAR 52.219-14, DFARS 252.225-7001 (with or without periods, and with a pinpoint like FAR 52.219-14(b)), resolving straight into the existing CFR Title 48 records, no separate corpus

The form a state’s own courts print resolves too, not just the Bluebook form: O.C.G.A. § 51-1-6, RCW 59.18.060, R.C. 2305.10, § 292.290, RSMo. The safest string to send is the citation value we returned to you.Agency guidance resolves as well, and it has no Bluebook form at all. Send its citation exactly as we printed it: NY Insurance Circular Letter No. 4 (2022), TDI Commissioner's Bulletin B-0004-22, IL Company Bulletin 2021-03. This covers STATE_AGENCY_GUIDANCE (state insurance bulletins and circular letters) and AGENCY_GUIDANCE (IRS Notices, OCC Bulletins, CFTC staff letters and the rest). Because these are matched on the published string rather than parsed, the citation we returned is not merely the safest input here, it is the only one guaranteed to work.Scope corpusType (and state) when the citation names the corpus, so a same-numbered section in another corpus cannot rank first. Scope is a constraint, not a hint: a citation naming a different jurisdiction returns resolved: false rather than being forced into the scope, so leave the scope off unless you are disambiguating.See Citation Formats for every form that resolves, the shared acronyms refused on purpose, and how state unlocks them.

This recipe covers statutes and regulations only (USC, CFR, state codes, state regulations, constitutions, court rules, and other statute-corpus types). It is not for case law citations. Case law is not part of the Vaquill Developer API.

Endpoints used: GET /us/statutes/resolve for one citation (2 credits), POST /us/statutes/resolve for up to 50 at a time (2 credits each).

Resolve in one call

GET /us/statutes/resolve does the whole thing for you: it parses the Bluebook citation, resolves it to the exact section, and returns the section metadata with an official source link. The resolved flag tells you whether the citation is real, which is exactly the signal you want when verifying a citation an LLM produced.

curl -G https://api.vaquill.ai/api/v1/us/statutes/resolve \
  -H "Authorization: Bearer vq_key_..." \
  --data-urlencode "cite=42 U.S.C. § 1983"

import requests

resp = requests.get(
    "https://api.vaquill.ai/api/v1/us/statutes/resolve",
    headers={"Authorization": "Bearer vq_key_..."},
    params={"cite": "42 U.S.C. § 1983"},
)
data = resp.json()
if data["resolved"]:
    sec = data["section"]
    print(sec["citation"], "->", sec["actId"])
else:
    print("Unverified citation:", data["inputCitation"])

const url = new URL("https://api.vaquill.ai/api/v1/us/statutes/resolve");
url.searchParams.set("cite", "42 U.S.C. § 1983");
const data = await (
  await fetch(url, { headers: { Authorization: "Bearer vq_key_..." } })
).json();
if (data.resolved) {
  console.log(data.section.citation, "->", data.section.actId);
} else {
  console.warn("Unverified citation:", data.inputCitation);
}

A resolved response carries the full section (the same shape as /us/statutes/section/{actId}), plus subsection when the citation had a pinpoint like 42 U.S.C. § 1983(b)(2). When resolved is false, treat the citation as unverified, not current.

/us/statutes/resolve is charged whether or not the citation resolves, because a confident “this does not resolve” is the answer you want when checking an LLM’s output. Only server errors are refunded.

A guidance citation can be ambiguous, and we would rather answer nothing than guess. Some regulators reuse one citation across several documents: Delaware reissues DE Domestic/Foreign Bulletin No. 113 five times, Vermont twice, and Massachusetts has three separate documents citing MA Bulletin 1991-01. Where the publisher’s own citation does not identify a single document, resolved is false rather than one of the candidates picked arbitrarily. Reach those documents by actId, or with a corpusType=STATE_AGENCY_GUIDANCE search scoped to the state.

Resolve a whole document’s citations at once

A brief, a memo, or a model’s output typically carries dozens of citations, and checking them one at a time is dozens of round trips. POST /us/statutes/resolve takes up to 50 citations in one call and returns the same verdict per citation.

curl -X POST https://api.vaquill.ai/api/v1/us/statutes/resolve \
  -H "Authorization: Bearer vq_key_..." \
  -H "Content-Type: application/json" \
  -d '{"citations": ["42 U.S.C. 1983", "16 C.F.R. 444.1", "Cal. Civ. Code 1950.5"]}'

import requests

resp = requests.post(
    "https://api.vaquill.ai/api/v1/us/statutes/resolve",
    headers={
        "Authorization": "Bearer vq_key_...",
        "Content-Type": "application/json",
    },
    json={
        "citations": [
            "42 U.S.C. 1983",
            "16 C.F.R. 444.1",
            "Cal. Civ. Code 1950.5",
        ]
    },
)
resp.raise_for_status()
data = resp.json()

print(f"{data['resolvedCount']}/{data['count']} resolved, "
      f"{data['creditsConsumed']} credits")
for item in data["results"]:
    if item["resolved"]:
        print("OK  ", item["inputCitation"], "->", item["section"]["actId"])
    else:
        print("FLAG", item["inputCitation"])

const resp = await fetch(
  "https://api.vaquill.ai/api/v1/us/statutes/resolve",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer vq_key_...",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      citations: ["42 U.S.C. 1983", "16 C.F.R. 444.1", "Cal. Civ. Code 1950.5"],
    }),
  },
);
if (!resp.ok) throw new Error(`resolve failed: ${resp.status}`);
const data = await resp.json();
for (const item of data.results) {
  console.log(item.resolved ? "OK" : "FLAG", item.inputCitation);
}

A trimmed response:

{
  "results": [
    {
      "resolved": true,
      "inputCitation": "42 U.S.C. 1983",
      "section": {
        "actId": "USC_T42_C21_S1983",
        "citation": "42 U.S.C. § 1983",
        "title": "Civil action for deprivation of rights"
      },
      "subsection": null
    },
    {
      "resolved": false,
      "inputCitation": "42 U.S.C. 19833",
      "section": null,
      "subsection": null
    }
  ],
  "resolvedCount": 1,
  "count": 2,
  "creditsConsumed": 4
}

Four things about the batch route are worth knowing before you build on it.

Every input gets an entry. An unresolved citation comes back with resolved: false, never as an omission, so you can tell WHICH of thirty citations failed. The results array lines up with your de-duplicated input, in order.
Duplicates are collapsed before pricing and before the response is built, so the same citation sent twice costs once and returns one entry.
Cost is 2 credits per citation, the same as resolving them one at a time. Batching buys a round trip and latency, not a discount. Thirty citations is 60 credits either way. Read creditsConsumed rather than multiplying: citations we failed to process because of a backend error are refunded.
Scope the whole batch with state or corpusType when every citation belongs to one jurisdiction or corpus. 22 CCR 76227 is California and 8 CCR 1206-2 is Colorado, and the acronym alone cannot say which. Omit both for a batch that spans jurisdictions and let each citation name its own.

POST /us/statutes/resolve accepts at most 50 citations, and rejects an empty list or a longer one with 422. Chunk a longer document into batches of 50.

If you need your own match tolerance

Both resolve routes run the server-side citation parser, which knows every form in Citation Formats, including the ones a state’s own courts print and the agency-guidance strings that have no Bluebook form at all. You only need to build your own matcher if you want tolerance the parser does not offer, for example accepting a citation whose section number is off by a digit and reporting it as a near miss. In that case, send the citation as the query on POST /us/statutes/search, compare the top result’s citation and citationShort against your input under your own rules, and call GET /us/statutes/section/{actId}/body when you are satisfied. Do the arithmetic before you commit to it. That path is 4 credits for the search plus 6 for the body, so 10 credits against 2 for /resolve, and a substring comparison is a weaker matcher than the parser you just skipped. It is the right call only when the tolerance itself is the point.

Flagging what you cannot resolve

Return an unresolved flag, not a guess

resolved: false means unverified, not “not current”. Do not fall through to a loosely related section, and do not present a near match as the answer. If you are checking a citation an LLM produced, an unresolved flag is exactly the signal that the model may have fabricated it.

Scope by corpus to cut false matches

The corpus token is usually right there in the citation. ... CFR ... means corpusType: "CFR"; ... U.S.C. ... means "USC"; a state code section means "STATE" with the two-letter state. Scope is a constraint, not a hint: a citation naming a different jurisdiction returns resolved: false rather than being forced into the scope, so leave the scope off unless you are disambiguating. FAR/DFARS citations live in CFR Title 48, so they are corpusType: "CFR"; add source: "far" or source: "dfars" to exclude every other CFR title from the ranking.

Text unavailable on a resolved citation

A citation can resolve while its full text is not yet ingested. In that case GET /us/statutes/section/{actId}/body responds with available: false and creditsConsumed: 0. Treat it as resolved-without-text: you have a confirmed actId, citation and source link, but nothing you can quote yet.

Picking the official source link

A resolved section carries several optional source URL fields and any of them may be null. Use the canonical preference order in Text Formats and Source URLs rather than writing your own, so every surface in your product links the same reader to the same official page.

Ground an LLM answer (RAG)

Use verified citations as grounding sources for an LLM answer.

Map a requirement to the law

Go from a plain-language obligation to the governing sections.

Grounding LLMs

Why citation verification belongs in every legal AI pipeline.

Statute lookup chatbot

A conversational front end over the same lookups.

DATA COVERAGE

USE CASES

GET STARTED

CORE CONCEPTS

LAW CHANGE ALERTS

WORKING WITH THE API

GROUNDING AI

RECIPES

Resolve and verify a statute citation

Resolve in one call

Resolve a whole document’s citations at once

If you need your own match tolerance

Flagging what you cannot resolve

Ground an LLM answer (RAG)

Map a requirement to the law

Grounding LLMs

Statute lookup chatbot

​Resolve in one call

​Resolve a whole document’s citations at once

​If you need your own match tolerance

​Flagging what you cannot resolve

​Related

Ground an LLM answer (RAG)

Map a requirement to the law

Grounding LLMs

Statute lookup chatbot

Resolve in one call

Resolve a whole document’s citations at once

If you need your own match tolerance

Flagging what you cannot resolve

Related