Why Is Cite Checking So Difficult?
· LawDiver
Cite checking is difficult because one decision has many valid cites, The Bluebook is not the rule in the courts that read the brief, and some states require two reporters. Free case indexes usually store one cite, or none.
Cite checking is difficult because a single court decision does not have a single address. The same opinion can be a volume and page in an official state reporter, a different volume and page in a regional reporter, a public-domain number, a Westlaw or Lexis identifier, and a slip opinion the court posted the morning it was decided. The forum you are filing in may require two of those addresses at once. The Bluebook is the book everyone was taught, and it is not the rule in a large share of the courts that will actually read the brief. The free case index behind tools that do not license Westlaw, Lexis, or vLex usually stores one of those addresses, or none.
That is a data problem wearing a style-manual costume. CiteCheck is our answer to it: upload a brief, and every citation is resolved against a corpus that holds the parallel reporter cites the filing rule asks for, not just the one cite an open index happened to keep.
On this page
- The short answer
- Too many authorities, and more than one way to write each one
- What is the Bluebook?
- How much is the Bluebook actually followed?
- Why one state case can have two or three reporters
- Which states require both cites
- New York is the rule that makes a single cite fail
- CourtListener has one cite, or none
- Harvard and CourtListener mangled New York's official cites
- Why parallel cites used to be unsolvable
- What that compute cost
- The other ways a correct-looking cite is wrong
- Where CiteCheck fits
- Method and bias disclosure
- Frequently asked questions
- Sources
- Related reading
The short answer
A cite check has to answer four questions that look like one question:
- Does this authority exist? Volume, reporter, and page have to land on a real decision, statute, or rule.
- Is this the form this court accepts? Official reporter, regional reporter, public-domain cite, or some combination. The acceptable set changes with the forum.
- Is it the same decision the writer meant? A second reporter cite is not a second case. A wrong series (So. 2d versus So. 3d) is.
- Is it still good law? A perfect citation to an overruled case is a failed cite check.
Most software stops at the first question, and it can only do that for the citation string someone already typed into a database. American citation practice fails the second and third questions constantly, because the string in the brief and the string in the database are often both "right" and still not the same string.
Too many authorities, and more than one way to write each one
A brief is not a list of cases. In one document a lawyer may cite:
- decisions of the U.S. Supreme Court, the federal courts of appeals, and federal district courts
- state appellate and trial opinions, including out-of-state cases used only for their reasoning
- the United States Code, the Code of Federal Regulations, and one or more state codes
- rules of procedure, rules of evidence, and local court rules
- constitutions, administrative decisions, and the occasional treatise or law review
Each of those families has its own citation grammar. A U.S. Code section, a Florida DCA opinion, and a Treasury regulation are not three flavors of the same string. A checker that "does cases" has not cite-checked the brief.
Even inside cases, identity is not unique. These are all accepted ways to point at one decision, depending on who is reading:
| Form | What it is | Example shape |
|---|---|---|
| Official state reporter | Printed (or officially paginated) by the state | 9 N.Y.3d 342 |
| Regional reporter | West's National Reporter System | 880 N.E.2d 1 |
| Official unofficial | A second West reporter for the same state | 849 N.Y.S.2d 480 |
| Public-domain / neutral cite | Court, year, and sequential number | 2011 IL 110542 |
| Slip opinion | The court's own release, before a reporter page exists | docket and slip page |
| Vendor identifier | A private database key | 2007 WL … or a Lexis cite |
| Advance sheet | A weekly that holds the cite until the bound volume arrives | Florida Law Weekly |
None of these is a translation of the others. 9 N.Y.3d 342 does not mathematically produce 849 N.Y.S.2d 480. The page breaks were chosen by different publishers, in different books, years apart. If your database has one of them, you cannot derive the rest. You either stored them or you did not.
That is the whole difficulty in one paragraph. Everything below is a consequence of it.
What is the Bluebook?
The Bluebook: A Uniform System of Citation is the dominant American citation manual. The current book is the 22nd edition (2025), published by the Harvard Law Review Association and compiled by the law reviews of Harvard, Columbia, the University of Pennsylvania, and Yale. It has been the default law-school text for nearly a century.
It is two manuals bound together:
- The Bluepages are for practitioners: briefs, motions, and memos. They allow either italics or underlining. They are the shorter, slightly more forgiving half.
- The Whitepages are for law reviews and journals. Typeface, signals, and abbreviations are stricter. A law-review cite and a brief cite to the same case are not always the same string.
For cases, the machinery that matters is Rule 10, plus the jurisdiction tables (Table T1) and the local-rule table (Bluepages Table BT2). Rule 10 tells you how to write a case name, which reporter to choose, when to add a court and year, and how to shorten the cite the second time. The 22nd edition also revised signals (including a new contrast signal), quotation cleanup, and the local-rule tables. A cite that was "Bluebook" under the 21st edition (2020) is not automatically "Bluebook" under the 22nd.
The Bluebook is a style convention with the force of local rule only where a court adopts it. It does not assign reporter volume and page. Those numbers come from the publisher of the book, or from the court when the court has adopted a public-domain citation. The Bluebook tells you which of those numbers to print. It does not own them.
How much is the Bluebook actually followed?
Almost every U.S. law student is taught it. Many federal courts expect it, or "a recognized citation system," in filings. Law reviews follow the Whitepages. In that sense the Bluebook won.
In the sense that matters for cite checking — will this string satisfy the court that is about to read it? — the Bluebook is the starting point that local rules overwrite. A few examples:
- Florida does not cite Florida cases "the Bluebook way." Florida Rule of Appellate Procedure 9.800 is its own uniform citation system. Florida appellate opinions published in the Southern Reporter are cited to So. 3d. The court parenthetical is
Fla.orFla. 2d DCA, not the Bluebook's longer district-court-of-appeal form. We covered why that rule still points at a commercial reporter in Why caselaw citations still depend on dead print reporters. - New York filings are governed by CPLR 5529(e) and by the New York Law Reports Style Manual (the Tanbook), not by Bluebook Table T1. The Tanbook consults the Bluebook for matters it does not cover. On New York cases, it deliberately disagrees.
- California official reports follow the California Style Manual, not the Bluebook. The two books choose different reporter emphasis and different abbreviations.
- Illinois requires a public-domain citation for reviewing-court opinions filed on or after July 1, 2011. A parallel to the North Eastern Reporter is allowed and not required.
- Ohio, Massachusetts, and other states publish their own writing or style manuals. Where the manual and the Bluebook conflict, the manual wins inside that court.
Two further limits on "we follow the Bluebook":
Practitioners follow the Bluepages loosely. Typeface, the exact order of signals, and small-caps in the Whitepages are academic rules. A partner checking a brief is looking for a recognizable case name, the reporter the court wants, a pinpoint, and a year. That is Bluebook-influenced. It is not a Whitepages audit.
The Bluebook's own default is one reporter, not two. Outside a local rule, Rule 10.3.1 tells you to cite the regional reporter. Parallel citations — official reporter, then regional reporter — are required when the court you are filing in requires them, not because the Bluebook always wants both. Public-domain states are the exception inside the book itself: Rule 10.3.3 says to give the public-domain cite and, where possible, a parallel to the regional reporter.
So the honest measurement is: the Bluebook is universally known and selectively obeyed. A cite checker that grades every brief against one national table will mark correct Florida, New York, and California cites as wrong, and it will accept a single regional cite that New York law rejects. The grade key is the forum.
Why one state case can have two or three reporters
West's National Reporter System divides the states into seven commercial regions. If you remember nothing else, remember that a "regional reporter" is a private book, not a court.
| Regional reporter | States |
|---|---|
Atlantic (A., A.2d, A.3d) | Connecticut, Delaware, D.C., Maine, Maryland, New Hampshire, New Jersey, Pennsylvania, Rhode Island, Vermont |
North Eastern (N.E., N.E.2d, N.E.3d) | Illinois, Indiana, Massachusetts, New York, Ohio |
North Western (N.W., N.W.2d) | Iowa, Michigan, Minnesota, Nebraska, North Dakota, South Dakota, Wisconsin |
Pacific (P., P.2d, P.3d) | Alaska, Arizona, California, Colorado, Hawaii, Idaho, Kansas, Montana, Nevada, New Mexico, Oklahoma, Oregon, Utah, Washington, Wyoming |
South Eastern (S.E., S.E.2d) | Georgia, North Carolina, South Carolina, Virginia, West Virginia |
Southern (So., So. 2d, So. 3d) | Alabama, Florida, Louisiana, Mississippi |
South Western (S.W., S.W.2d, S.W.3d) | Arkansas, Kentucky, Missouri, Tennessee, Texas |
On top of that map, many states printed — and some still print — an official reporter. The official book and the regional book publish the same opinion and paginate it differently. Those are the multiple-reporter states. A single appellate decision can carry:
- Two cites (official + regional). This is the ordinary dual-reporter case: Massachusetts Reports and the North Eastern Reporter, Georgia Reports and the South Eastern Reporter, Michigan Reports and the North Western Reporter, Washington Reports and the Pacific Reporter, Virginia Reports and the South Eastern Reporter, New Hampshire Reports and the Atlantic Reporter, Ohio Official Reports and the North Eastern Reporter.
- Three cites. New York Court of Appeals decisions are the clearest example: the official New York Reports (
N.Y.3d), West's New York Supplement (N.Y.S.3d), and the North Eastern Reporter (N.E.3d). Appellate Division decisions drop the North Eastern Reporter and keepA.D.3dplusN.Y.S.3d. California Supreme Court decisions are the other famous triple: official California Reports, California Reporter, and the Pacific Reporter. - A public-domain cite plus a regional cite. Illinois, Ohio, Oklahoma, Vermont, and a long list of other states (Arkansas, Colorado, Louisiana, Maine, Mississippi, Montana, New Mexico, North Dakota, South Dakota, Utah, Wisconsin, Wyoming, and others) identify decisions by a medium-neutral number and still appear in a regional reporter.
- An advance-sheet cite that expires. Florida Law Weekly holds a Florida opinion until a So. 3d page exists. Citing the weekly after the Southern cite has issued is a different mistake from citing the wrong case.
Series numbers make this worse. F.2d, F.3d, and F.4th are different books. So. 2d and So. 3d are different books. A volume-and-page that is a real case in one series is a hallucination in the next. Cite checking has to know the series, not just the abbreviation family.
Cornell's Introduction to Basic Legal Citation states the practical result: the same decision, cited inside its own state, often wears a different citation than it wears everywhere else. There is no national canonical string to store and replay.
Which states require both cites
"Require" is a local-rule word. The Bluebook says the quiet part out loud: many state rules require a citation to the official state reporter followed by a parallel citation to a regional reporter, and those local rules — not the national table — govern a filing in that court. Pinpoint pages, when you need them, have to be given for each reporter, because the page numbers are not the same.
These are the states where a second cite is part of a complete filing form, not a courtesy. The strength of the obligation is not the same in each row. New York is in its own section because its statute is the one that is easy to quote and easy to fail.
| State | The two (or three) cites | How hard the "both" is |
|---|---|---|
| New York | For non-New York cases: official reporter, if any, and the National Reporter System. For New York cases: the official reports (N.Y.3d, A.D.3d, Misc. 3d). | Statute. CPLR 5529(e). See the next section. |
| California | Cal. / Cal. App. together with Cal. Rptr., and P.3d for the Supreme Court | Official-reports style includes the parallel. California Style Manual, not the Bluebook, is what the official reports use. |
| Ohio | Ohio Official Reports and N.E.2d / N.E.3d, alongside the public-domain cite | The Supreme Court of Ohio Writing Manual calls for the parallel. |
| Georgia | Ga. / Ga. App. and S.E.2d | In-state form is official plus South Eastern. |
| Massachusetts | Mass. and N.E.2d / N.E.3d | Dual-reporter state. (The SJC style manual, written for the official reports themselves, often prints the official cite alone. Briefs and out-of-state cite checkers still meet the North Eastern cite constantly.) |
| Michigan | Mich. and N.W.2d | Official plus North Western. |
| Minnesota | Minn. and N.W.2d | Official plus North Western. |
| New Hampshire | N.H. and A.2d / A.3d | Official plus Atlantic. |
| Virginia | Va. and S.E.2d | Official plus South Eastern. |
| Washington | Wash. and P.2d / P.3d | Official plus Pacific. |
| Vermont | Public-domain VT or Vt., plus Atlantic | Neutral cite and the regional reporter travel together. |
| Oklahoma | Public-domain OK cite, plus P.3d once the Pacific cite has issued | Oklahoma double-cite practice. |
| Illinois | Public-domain cite required for opinions filed on or after July 1, 2011; N.E. and Illinois Decisions may be added | The second reporter is optional under Illinois Supreme Court Rule 6, and people still write it. A database that dropped N.E.2d cannot verify the cite the writer actually used. |
A cite checker can be "right" about existence and still fail every row in that table if it stored only the regional reporter, or only the official reporter, or only a slip opinion.
New York is the rule that makes a single cite fail
CPLR 5529(e) is short. This is the part that matters:
New York decisions shall be cited from the official reports, if any. All other decisions shall be cited from the official reports, if any, and also from the National Reporter System if they are there reported. Decisions not reported officially or in the National Reporter System shall be cited from the most available source.
Read it twice, because it is the opposite of what most databases optimize for.
New York cases in a New York paper are cited to the official reports: New York Reports, Appellate Division Reports, or Miscellaneous Reports. The 2022 Style Manual goes further and tells the official reports not to add parallel unofficial cites (N.Y.S., N.E.) for cases that already have an official cite. Court rules on submissions say the same thing in the other direction: where New York authorities are cited, the official cite has to be included if it exists.
Every other state's cases in a New York paper need both addresses when both exist. A California Supreme Court case is not done at 40 Cal. 4th 106. The National Reporter System also reports it, so the Pacific Reporter cite has to be there too, and the California Reporter cite is part of the official-unofficial set lawyers actually write. A Massachusetts case needs Mass. and N.E.. A Georgia case needs Ga. and S.E.. A Michigan case needs Mich. and N.W.. An Ohio case needs the Ohio official cite and N.E..
That is why "we have a reporter cite for this case" is not a cite-check result in New York. It is a coin flip about which of the two required strings you happened to keep.
And the New York case itself still has to be stored under all three of its reporters. The Tanbook suppresses N.Y.S. and N.E. inside the official reports. The rest of the country, and plenty of New York practitioners, cite those reporters anyway. People v. Danielson is 9 N.Y.3d 342, 880 N.E.2d 1, and 849 N.Y.S.2d 480. Drop any one of them and a correct brief looks like it cited a case you do not have.
CourtListener has one cite, or none
CourtListener, run by the Free Law Project, is the opinion index behind a large share of case-law products that do not license Westlaw, Lexis, or vLex. Harvard's Caselaw Access Project digitized the historical print reporters; CourtListener distributes that bulk data and adds opinions harvested from court websites. If your cite checker is not paying Thomson Reuters or RELX, this graph is probably your graph.
It is a remarkable public good. It is a bad citation authority, for structural reasons rather than sloppiness.
Modern slip opinions often have no reporter cite at all. Courts post the opinion. The commercial reporter assigns volume and page later, or never puts that assignment on the court site. The harvest copies what the court posted. Florida is the example we have measured: Florida Rule 9.800 wants So. 3d, and for a large set of Florida appellate opinions from about 2019 forward the free record has the text and no Southern Reporter key. We wrote up the reconstruction problem, with counts, in the print-reporters article.
When a reporter cite is present, there is rarely more than one. Historical bulk data was built by scanning one print series. A volume of the North Eastern Reporter contributes an N.E.2d cite. The official N.Y.2d or Mass. cite for that same decision was printed in a different book and often never attached to the same record. A cluster with two reporter cites is the exception. A cluster with the official cite, the regional cite, and the state-specific West cite is not what this index was built to hold.
So the open graph can tell you that an opinion exists, and it can often tell you one way to cite it. It cannot, by itself, tell you the second way a New York brief is required to cite it. Every product that ingests CourtListener and stops there inherits that gap. The gap is invisible until a lawyer cites the reporter you do not have, at which point the checker says "not found" about a real case.
For New York, the public file does something worse than omitting a cite. It stores the wrong reporter series.
Harvard and CourtListener mangled New York's official cites
New York Reports are not one book with a growing volume number. They are successive series, and the series is part of the citation.
N.Y.is the first series. It stopped around volume 309, in the 1950s. Volume 41 of that series is a nineteenth-century reporter.N.Y.2dis the second series, which began after the first series closed. Volume 41 of the second series is 1976–77.N.Y.3dis the third series.
41 N.Y. 60 and 41 N.Y.2d 60 are different books. They are not two ways to write the same cite.
Harvard's Caselaw Access Project shelved some second-series Court of Appeals volumes on the first-series shelf — the New York Reports path, with no series. The bulk files CourtListener distributes, which is the dump everyone except Westlaw and Lexis starts from, imported those opinions as N.Y. The second series was erased in the citation string.
We measured that scramble on three volumes. Every one of these rows was a real N.Y.2d opinion stored as N.Y.:
| What the bulk file stored | What the case actually is | Opinions | When they were decided |
|---|---|---|---|
41 N.Y. | 41 N.Y.2d | 106 | December 1976 – June 1977 |
44 N.Y. | 44 N.Y.2d | 83 | February – July 1978 |
65 N.Y. | 65 N.Y.2d | 54 | April – September 1985 |
That is 243 opinions in the band we repaired. The failure mode is the same on each of them, and it is not a near miss.
People v. Belge was decided on December 20, 1976. The cite is 41 N.Y.2d 60. The public bulk record stores 41 N.Y. 60. Look up the real cite and the case is missing. Look up the stored cite and a 1976 Court of Appeals opinion comes back wearing a first-series number that belongs to a different century — or it collides with whatever nineteenth-century case actually occupies 41 N.Y. 60. The same break hits People v. Huffman (41 N.Y.2d 29 stored as 41 N.Y. 29), New York Times Co. v. City of New York Commission on Human Rights (41 N.Y.2d 345), City of Buffalo v. Cargill, Inc. (44 N.Y.2d 7), and Hunt v. State Tax Commission (65 N.Y.2d 13).
The other two New York reporters are simply absent. Belge is also 390 N.Y.S.2d 867 and 359 N.E.2d 377. Those strings are not on the bulk citation row. A New York brief can cite the official second series, the Supplement, or the North Eastern Reporter. The public file has a corrupted version of the first and neither of the others.
Anyone who loads that dump and does not repair it has a New York citator that fails the cite lawyers write, and can return the wrong era of case for the cite the file invented. That is the data underneath a large share of "free" cite checking.
Why parallel cites used to be unsolvable
You cannot buy the missing cite from the free index. It is not in the free index.
You cannot compute it. Reporter pagination is not a function. Knowing 261 Ga. 491 tells you nothing about the number of the page where that opinion starts in 405 S.E.2d.
You cannot look it up in the opinion's own header if the header was a slip opinion. The second cite was assigned later, by a different publisher.
The place both cites do appear together is in the text of later decisions. When a court (or a brief that a court quoted) cited People v. Danielson, 9 N.Y.3d 342, 880 N.E.2d 1, 849 N.Y.S.2d 480, that sentence is a join key. Two or three reporter addresses, one decision. Do that for every parallel citation in the corpus, and the second reporter stops being metadata you were never given. It becomes a fact you extracted from the law's own citations to itself.
That project was not realistic when citation indexes were maintained by hand, or when the working set was a library's reporter wall. It is a full-corpus read: every opinion, every citation string that names more than one reporter, both (or all three) addresses written back onto the decision they name. The opinions that cite a case in parallel are doing the work the upstream index declined to do.
What that compute cost
We ran that extraction across the case texts of about ten million opinions. The server that holds those texts is 8 CPU cores and 32 GB of memory. The job kept that machine saturated for 15 days — 360 hours, day and night, with no spare capacity. Saturated means the cores were busy with this work the whole time. It was not a script that started on a laptop and then waited.
That is two different loads. They ran on the same machine. They are not the same kind of work, and a PC that matches the core count still only covers one of them.
The extraction itself
Reading citation strings and writing both reporter addresses back onto the cited case is ordinary CPU work, at an extraordinary volume.
| Measure | Figure |
|---|---|
| Wall clock | 15 days = 360 hours |
| Core-hours | 8 cores × 360 hours = 2,880 core-hours |
| Memory reserved for that window | 32 GB × 360 hours = 11,520 GB-hours |
| Per opinion, across ~10 million cases | about 1 core-second |
Most citation strings name one reporter. Only the ones that name two teach the database a parallel it did not already have. The scan still has to read the rest to find them. One core-second per opinion is the whole server, for half a month, because the useful events are sparse.
On a spec sheet, an 8-core desktop with 32 GB of RAM — a strong office workstation, not a typical laptop — has the same CPU and the same memory. A common laptop is closer to 16 GB, sometimes 8. If the case texts were a file that already fit in memory, that desktop could attempt the same arithmetic. They are not that file.
The database, counted separately
The extraction is useless unless something can hand it the text of ten million opinions, continuously, for all 360 hours. That text lives in a database of about 300 gigabytes: the opinions themselves, on the order of ten million of them, and well over a hundred million citation rows that point between them. Thirty-two gigabytes of RAM holds about a tenth of that database, and only if the operating system and the scan take nothing. In practice the cache is smaller. The other nine-tenths stay on disk, and the disk is queried the entire time. This is not "open a folder of PDFs." It is a query engine that has to stay hot — indexes up, reads answered, results written back — without a break for fifteen days.
Counted on its own, that access is a second server of the same size: another 8 cores and 32 GB, busy for the same 15 days, another 2,880 core-hours. You do not add those hours on top of the extraction in the calendar, because we ran the scan on the machine where the data already lived, and that one machine was already full. You do count them when you ask what the database required. There is no version of this project where the 300-gigabyte database is free and only the citation parser costs something.
A PC loses that half even when its CPU looks identical.
- The data does not fit. Three hundred gigabytes is larger than the free space on a lot of laptops, and it is about ten times the RAM of the server itself. A desktop that can store it still cannot cache it. Every pass through the corpus is a disk read.
- One set of cores cannot be both jobs. The extraction wanted all 8 cores. The database wanted a machine that could serve indexed reads for 360 hours at the same time. On the server those demands met in the middle and maxed the box. A PC has one set of 8 cores. Splitting them between a 300-gigabyte database and a full-corpus scan means the job runs much longer than 15 days, or it thrashes until it fails.
- A PC does not stay up like that. Laptops sleep. Desktops update, reboot, and share the disk with everything else their owner is doing. This run needed every core and the database pinned for half a month.
The 2,880 core-hour extraction figure, and the separate 2,880 core-hour database figure, are estimates from that shape of machine running flat out for fifteen days. They are the right order of magnitude for "read the case law and keep both cites." They are not a metered invoice.
Hand citation indexes never did this because the cost was human hours, not core-hours. Westlaw and Lexis did the editorial version of it over the twentieth century and charge for the result. Doing it as a batch over the public text is what got cheap enough to run — on a server that can keep a 300-gigabyte database online while the scan uses the machine. A PC with a similar processor is not that server.
The other ways a correct-looking cite is wrong
Parallel reporters are the structural problem. They are not the only one. Anyone who has cite-checked a brief by hand already knows the rest. A tool has to know them too, or it will rubber-stamp errors that happen to use a familiar shape.
Short forms only work inside one document. Id. means "the thing I just cited," at a pinpoint that may have changed. A short form (Danielson, 9 N.Y.3d at 348) is meaningless if the full cite was never given, or was given to a different reporter. Checking id. is a document walk, not a database lookup.
Pinpoints are part of the cite. 531 U.S. 98 is Bush v. Gore. 531 U.S. 98, 105 is a specific page. A parallel cite needs a pinpoint in each reporter when the local rule requires pinpoints. Quoting page 105 and citing page 109 is a failed check even though the case is real.
The series is part of the identity. 772 So. 2d 1243 and a made-up 772 So. 3d 1243 are not "close." Adjacent series are where fabricated citations go to look plausible. So are off-by-one page numbers and transposed digits.
The caption drifts. People, State, and Commonwealth swap. Corporate suffixes appear and disappear. In re and Matter of trade places. Party order flips on appeal. A checker that demands a verbatim caption will reject real cites; one that ignores the caption will glue the right reporter to the wrong case.
The parenthetical is an identity field. Fla. and Fla. 2d DCA are different courts. Omitting the court is correct when the reporter already names it (Bluebook Rule 10.4) and incorrect when it does not. Florida's rule wants the district. A national template that always prints Fla. Dist. Ct. App. is wrong in the court that invented the cite.
Slip cites rot. A Florida Law Weekly cite, a New York slip opinion, or a Westlaw number can be the best cite on the day of decision and the wrong cite after the reporter volume is bound. Sometimes the bound volume also changes the page. Advance sheets are not a stable address.
Vendor cites are not public addresses. 2007 WL 1234567 verifies inside Westlaw. It does not verify in an open corpus unless someone has mapped that identifier onto a decision. Treating a WL cite as confirmed because the number "looks well-formed" is how hallucinated vendor cites survive.
Existence is not good law. The cite can be perfect and the case overruled, vacated, reversed, or superseded by statute. Shepard's and KeyCite exist because this problem is older than digital research. A checker that only answers "this case is real" has done the easy half.
The brief is not only cases. A mis-citation to a statute — wrong title, wrong section, a repealed subsection, a state code cited as if it were the U.S. Code — is a cite-check failure. So is a court rule from the wrong jurisdiction. Models that "know" case citations and guess at everything else will invent a plausible C.F.R. part as readily as they invent a plausible F.3d page.
Quotations have to match the source. Courts have sanctioned briefs for citations to real cases that do not say what the brief claims, and for citations to cases that do not exist at all. Those are different defects. Both have to be caught. The second one is a lookup. The first one is a comparison against the opinion text at the pinpoint.
Near misses are the dangerous ones. A completely fictional caption is easy to spot. A real case name with a page number from a different reporter, or a real reporter cite with the year of a later history, looks like a successful check to a tool that only tests one field.
Where CiteCheck fits
Unless you pay for the cite-checking function inside Westlaw or Lexis, CiteCheck is, to our knowledge, the only cite-checking tool that holds the required parallel reporter cites for most of the cases in our corpus. That claim is narrow on purpose. It is about the second address — official and regional, or New York's three — not about who has the oldest citator brand.
The way those addresses got there is the project above: read the citations in the case text, and wherever one decision is cited to two reporters, keep both, attached to that decision. Open indexes rarely arrive with both. After that pass, a brief can cite either reporter and still resolve, and a New York filing can be graded against both reporters CPLR 5529(e) actually demands.
CiteCheck checks a brief or other legal work against that corpus: cases, statutes, rules, and whether the authority is still good law. It is free during the current rollout. If you want the walkthrough rather than the theory, start with How to cite check a brief.
Method and bias disclosure
We build LawDiver and CiteCheck, so we are not neutral. Here is how to separate the public record from our product claims.
Public record. The Bluebook's identity and edition, Rule 10.3.1's split between local parallel-cite rules and the regional-reporter default, CPLR 5529(e), the New York Law Reports Style Manual's rejection of unofficial parallels for officially reported New York cases, Florida Rule 9.800, Illinois's public-domain rule, and Cornell's description of in-state versus out-of-state citation forms are all externally readable. Links are in Sources.
Our corpus. Statements about which parallel cites CiteCheck can resolve, and the claim that we are the only non-Westlaw, non-Lexis cite checker we know of with those parallels for most of our cases, are claims about our database. They are not an audit of every other product on the market.
Compute. The extraction figure — 8 cores and 32 GB of memory, saturated for 15 days, 2,880 core-hours — and the separate database figure of the same size assume that server stayed maxed for the whole window while reading a case database of about 300 gigabytes. Treat both as estimates of the run we are describing, not as an independently metered benchmark. The two 2,880s overlapped on one machine. They are split here so the database is not mistaken for free overhead.
CourtListener and Harvard. Missing reporter cites, including Florida after about 2019, clusters that rarely carry more than one reporter, and the New York series scramble (N.Y.2d volumes stored as N.Y., including the 243-opinion band in volumes 41, 44, and 65) are what we measured in the bulk files those projects distribute. They are not statistics published by the Free Law Project or by Harvard.
Last updated September 29, 2026.
Frequently asked questions
Why is cite checking so difficult?
Because one decision has several valid citation forms, the form a court requires depends on where you are filing, and most free case indexes store only one form or none. A checker also has to resolve short cites, pinpoints, later history, statutes and rules, and quotations, not just ask whether a reporter volume exists.
What is the Bluebook?
The Bluebook: A Uniform System of Citation is the standard U.S. legal citation manual, compiled by the Harvard, Columbia, University of Pennsylvania, and Yale law reviews. The current edition is the 22nd (2025). It contains practitioner rules (the Bluepages) and law-review rules (the Whitepages). It tells you which reporter to print. It does not assign the volume and page.
Do lawyers have to follow the Bluebook?
Law schools teach it, and many federal courts and law reviews expect it. State courts often replace it on the points that change the cite. Florida uses Rule of Appellate Procedure 9.800. New York uses CPLR 5529(e) and the Law Reports Style Manual. California official reports use the California Style Manual. When a local rule and the Bluebook disagree, the local rule is the one a cite check has to grade.
What is a parallel citation?
A parallel citation gives two or more reporter addresses for the same decision, separated by commas. An official reporter cite followed by a regional reporter cite is the classic form. New York Court of Appeals cases often have three: New York Reports, New York Supplement, and the North Eastern Reporter. The page numbers differ because the books were paginated separately.
Which states require both reporter cites?
New York is the clearest. CPLR 5529(e) requires out-of-state decisions to be cited to the official reporter, if any, and also to the National Reporter System if they are reported there. California, Ohio, Georgia, Massachusetts, Michigan, Minnesota, New Hampshire, Virginia, and Washington still use an official reporter alongside a regional reporter. Vermont and Oklahoma pair a public-domain cite with the regional reporter. Illinois requires the public-domain cite for newer opinions and treats the regional reporter as optional. The Bluebook's own default, when no local rule applies, is the regional reporter alone.
Why do free case databases miss reporter citations?
CourtListener and Harvard's Caselaw Access Project distribute the opinion bulk that products use when they do not license Westlaw or Lexis. A scan of one reporter series usually contributes one cite. A slip opinion from a court website often contributes none, because the commercial volume and page are assigned later. Florida So. 3d cites from about 2019 forward are a clear example.
What is wrong with New York cites in that public data?
The series was stripped off. Some New York Reports, Second Series volumes were stored as the first series, N.Y., which ended around volume 309 in the 1950s. 41 N.Y.2d 60 (People v. Belge, 1976) is in the bulk file as 41 N.Y. 60. The real cite does not resolve. The stored cite points at a 1976 case under a number that belongs to a different book. In the band we measured — volumes 41, 44, and 65 of N.Y.2d — that is 243 opinions. The New York Supplement and North Eastern parallels for those cases are missing as well.
How did LawDiver get the second reporter cite?
From the case law itself. When an opinion cites a decision to two reporters, both addresses refer to that decision. Extracting those pairs across about ten million opinions, and writing both cites onto the cited case, is what fills the gap the open index leaves. The extraction kept an 8-core, 32 GB server saturated for 15 days: 2,880 core-hours. Serving the 300-gigabyte database those texts live in is a separate load of the same size. A PC with a similar processor does not have that database.
What is CiteCheck?
CiteCheck is LawDiver's cite checker. It resolves citations in a brief against the corpus, including parallel reporter cites, and reports whether the authority is still good law. It is free during the current rollout. Westlaw and Lexis sell their own cite-checking functions; to our knowledge they are the only other tools that hold this kind of multi-reporter coverage for most of the cases we hold.
Sources
Citation manuals and court rules
- The Bluebook: A Uniform System of Citation, 22nd edition (2025) — Harvard Law Review Association
- Bluebook Rule 10.3.1, as summarized for state filings — UIC Law Library (local rules often require official reporter plus regional parallel; otherwise cite the regional reporter)
- State court reporters and parallel citations — Georgetown Law Library
- Introduction to Basic Legal Citation — Cornell Legal Information Institute (in-state parallel cites versus out-of-state regional cites)
- CPLR 5529(e) — New York official-report cites, and official-plus-National-Reporter-System cites for all other decisions
- New York Law Reports Style Manual (2022) — parallel unofficial cites are not used for officially reported New York cases
- New York courts: statutes and regulations, quoting CPLR 5529(e)
- Fla. R. App. P. 9.800, Uniform Citation System
- Illinois Supreme Court Rule 6 (public-domain citation) — North Eastern Reporter parallel permitted, not required, for opinions filed on or after July 1, 2011
Open case data
- CourtListener — Free Law Project
- Why caselaw citations still depend on dead print reporters — LawDiver (Florida So. 3d coverage and the print-reporter history)