The Antitrust Case Over Westlaw's Purchase of Casetext
· LawDiver
A California firm says Thomson Reuters paid $650 million for Casetext, kept CoCounsel, and deleted the cheap research product. The complaint is a map of how legal tech actually competes.
On October 1, 2026, Rubin Law Firm, P.C., a solo practice in Carlsbad, California, filed a putative class complaint against West Publishing Corporation and Thomson Reuters Enterprise Centre GmbH. The pleading, Rubin Law Firm, P.C. v. West Publishing Corp., says the $650 million purchase of Casetext was a killer acquisition: Thomson Reuters bought the only low-priced horizontal alternative to Westlaw, moved CoCounsel into the premium stack, and on April 1, 2025 turned the standalone research product off. The plaintiff says it went from about $32.50 a month to $111.48 a month for the same job, case-law research with a way to tell whether a case is still good law.
Read as an industry document, the complaint is less a story about one invoice than a theory of how this market is built. Legal research sells three different things that happen to share a login: a corpus of public law, an editorial apparatus on top of that corpus, and, since 2023, a generative layer that drafts and reviews against the corpus. Casetext sold all three at a small-firm price. Thomson Reuters, on the complaint's account, kept the third, folded it into the expensive version of the first two, and deleted the price. That is the move the rest of legal tech has to understand, whether or not a court eventually trebles the damages.
The complaint is a pleading. None of it has been tested by an answer or a class-certification record. What follows is what it alleges, what the public deal record says alongside it, and what the theory implies for anyone who builds, buys, or replaces a research product.
On this page
- The product that was actually discontinued
- The price, and what it is evidence of
- What $650 million bought
- Killer acquisition is a term of art, and this deal is a hybrid
- The market the complaint draws, and the products it leaves out
- Four statutes, one industry structure
- The class is Westlaw customers, not Casetext refugees
- Why Casetext's dataset was smaller
- The other consolidation
- What it means if you build or buy legal research
- Disclosure
The product that was actually discontinued
Casetext was founded in 2013. For a decade it sold a federal-and-state case-law library, statutes and regulations, a treatment checker (SmartCite), and workflow tools that sat on that library: CARA, which read a brief and looked for missing authority, and Parallel Search. By the company's own March 1, 2023 launch release, more than 10,000 law firms used it, including more than 40 Am Law 200 firms.
The product that changed the strategic picture was CoCounsel. Casetext had pre-release access to GPT-4. On March 1, 2023, Jake Heller and Pablo Arredondo demonstrated the assistant on MSNBC's Morning Joe, two weeks before OpenAI published GPT-4. The launch release described a trust process of nearly 4,000 hours, more than 30,000 legal questions, and more than 400 beta lawyers who had already run it more than 50,000 times. The skills were research memos, document review, deposition preparation, and contract analysis. The complaint treats that launch as the moment a stagnant duopoly faced a product that could do the work the incumbents' search boxes only retrieved.
On June 26, 2023, Thomson Reuters Corporation agreed to buy Casetext for $650 million in cash. The deal closed on August 17, 2023. The closing release counted more than 10,000 law firms and corporate legal departments as customers.
The complaint's central factual allegation is what happened next. CoCounsel's code was absorbed and cross-licensed into premium tiers. The standalone Casetext database was retired effective April 1, 2025. The customer email that went out in September 2024 said the quieter version of the same plan: on March 31, 2025 the subscription would end, and with it access to the product and to work product saved inside it, unless the customer moved onto a Thomson Reuters product. A lawyer who came back to that thread on April 1, 2025 said the site no longer let them in.
That work-product sentence is the switching-cost mechanism in plain language. A research subscription is a login and a file. The file is the searches, folders, highlights, and briefs the lawyer has already paid to organize. When the file dies with the product, the rational response is to sign the replacement contract before the date certain, on the seller's form. The plaintiff's three-year Westlaw Classic agreement, with a built-in annual increase, is what that pressure looks like on one set of facts.
The thing that was killed, if the complaint is right about the killing, is the cheap all-in research seat. The thing that was kept is the generative assistant, repriced as an upgrade on the incumbent platform. Those are different products, and the antitrust theory only works if a court is willing to treat the first as the market that matters.
The price, and what it is evidence of
The plaintiff says it held baseline Casetext from late September 2018 through March 2025 at an annualized rate of about $32.50 a month, then signed Westlaw Classic at $111.48 a month in year one, $115.94 in year two, and $120.58 in year three. Each step is a contractual 4 percent increase. The complaint calls that a 243 percent premium, then 257 percent and 271 percent.
| Monthly | Versus $32.50 | Twelve-month outlay | |
|---|---|---|---|
| Casetext, as pleaded | $32.50 | — | $390 |
| Westlaw Classic, year 1 | $111.48 | +243% | $1,338 |
| Year 2 | $115.94 | +257% | $1,391 |
| Year 3 | $120.58 | +271% | $1,447 |
The percentages are the increase, not the ratio. $111.48 is 3.43 times $32.50. Over three years the pleaded Westlaw outlay is $4,176. Three years of the old plan is $1,170. The difference is $3,006 before anyone asks whether Westlaw Classic contains something the 2018 Casetext plan did not.
Inside the complaint the old price is not stable. One paragraph uses $32.50. Another says the comprehensive tier typically ran $65 to $110 a month, and as low as about $45 for early annual subscribers "like Plaintiff." A third says $65 to $99 or even less. A grandfathered contract can be all of those things at once. It is a weak benchmark for the 2025 street price of the product.
The shutdown discussion among practitioners showed the same spread. One subscriber described a 2021 "30 percent off for life" all-jurisdictions plan at $62.30 a month, and a later migration offer of $70 a month for a one-year Westlaw Classic term, with a representation at the time that declining the offer would leave Casetext in place. Others expected a full Westlaw seat in the $750 to $1,000 range, or remembered a CoCounsel pitch near $500 a month on top of Westlaw. One solo was told Precision plus CoCounsel was sold only to firms of at least two lawyers. The plaintiff's $111.48 sits at the low end of that range. The industry fact is not a single percentage. It is that the floor moved, and that the floor was the product small firms were actually using.
The complaint also recites older transactional rates as proof that Westlaw has long been able to raise price without losing the customer: $113 a search, or $15.33 a minute, to search all federal cases in 2008; $19.05 a minute in 2010; about $20 a minute to view an opinion or a federal statute by 2014; and flat-fee deals with annual escalators of 3 to 7 percent. Those per-minute numbers describe a retail tariff West largely abandoned as it pushed firms onto flat rates. They show what the incumbent could charge when the alternative was a library. They are thin evidence of the elasticity of a flat-rate seat in 2026, which is the market the damages have to be computed in.
A 4 percent escalator, by itself, is not a merger effect. The complaint says escalators in the 3-to-7 percent band were already normal. The merger effect, if it exists, is in the base the percentage applies to, or in the loss of the bargaining line "we will move this work to Casetext." Large firms used that line. Solos used the product instead. The complaint needs both stories, and they are not the same story.
What $650 million bought
The complaint treats the multiple as the motive. It estimates Casetext's annual recurring revenue at signing at about $10 million to $12 million, and calls $650 million more than 50 times revenue, paid to remove a threat rather than to buy a cash-flow stream. Thomson Reuters has not published that revenue figure. On the investor call the morning after signing, CEO Steve Hasker said Casetext brought "limited initial revenue" and that the company expected it to grow quickly and become accretive. Executives declined to give a revenue or pricing number that day. "Limited initial revenue" is consistent with a high multiple. It does not establish $10 million or $12 million. At $12 million the multiple is about 54 times. At $10 million it is 65 times. Both sit inside the complaint's "over 50x." Both are pleaded estimates.
The buyer's explanation was more operational than the complaint's, and it cuts both ways. Investors asked why Thomson Reuters did not build generative AI itself, and why Casetext rather than Harvey, which Allen & Overy had already deployed firmwide. Chief Product Officer David Wong's answer was time-on-model plus data. Casetext was one of a few companies given GPT-4 in the first half of 2022. It had written a set of legal prompts, several of which Thomson Reuters had not started: review a document, summarize a document, search a database. Wong also described the Casetext corpus as real and limited in the same sentence: access to a research data set, "albeit again, a smaller, less sophisticated one than we have."
That clause is the whole market-definition fight in one quote. It supports the complaint's claim that in early 2023 Westlaw did not have a shipped generative assistant comparable to CoCounsel. Westlaw Precision, as the complaint describes it, was extractive search and a taxonomy. It also supports the defense that Casetext was not yet a full substitute for the Westlaw database. The acquirer paid $650 million and, in the same breath, called the underlying library smaller and less sophisticated than its own.
Hasker described a "build, partner, and buy" strategy. The cash, he said, came from selling down the stake in London Stock Exchange Group. Money is fungible, so the funding source does not decide the antitrust question. It does locate the check outside the Westlaw subscription P&L. The complaint's phrase for the same dollars is monopoly capital deployed to preempt a price war. Both descriptions can be true of one wire transfer. The documents that would separate them are the synergy model and any analysis of what would happen to Westlaw renewals if Casetext stayed independent. Those are why complaints like this get filed. They are not attached to this one.
Heller, in the signing release, said joining Thomson Reuters would advance Casetext's mission. The September 2024 customer email said the company had been excited to offer a more robust set of research tools on Westlaw and was now moving forward with that plan. A reader can hear integration. A reader can also hear a two-year glide path: keep the model access, the prompts, and the name CoCounsel, then delete the cheap SKU once the customers have nowhere equally cheap to go. The complaint alleges that the acquisition agreement itself included an understanding that Casetext would be eliminated. The public deal papers do not say that. The shutdown, nineteen months after closing, is the evidence offered for it.
One later fact bears on the "they bought the team" account. Pablo Arredondo left Thomson Reuters at the end of 2025 and in 2026 joined Clio to work on technology for courts. The human capital that justified part of the 2023 multiple was not still inside the building when this complaint was filed. What remains of "Casetext," as an asset someone could be ordered to spin out, is code, a brand, a customer list, and prompts that have since been pointed at Westlaw's corpus. That is a different company from the one that existed on June 26, 2023.
Killer acquisition is a term of art, and this deal is a hybrid
"Killer acquisition" comes from pharmaceutical economics. Cunningham, Ederer, and Ma, writing in the Journal of Political Economy in 2021, studied overlapping drug projects and found incumbents buying pipeline drugs and then discontinuing them, particularly when the project threatened an existing franchise and particularly when the deal was small enough to avoid a hard look. The mechanism is specific. The acquired product is shut down so it cannot cannibalize the incumbent's rent, and the technology is not shipped as a new product.
Apply that mechanism here and it fits one SKU.
The standalone research subscription was discontinued. Customers were pointed at Westlaw. Saved work on the old platform did not come with them unless they migrated on Thomson Reuters' terms. That is discontinuation, and it is the half of the theory a solo can describe without an economist.
CoCounsel was not discontinued. It is still the assistant Thomson Reuters sells, attached to the premium tier the complaint's own class definition excludes. Absorbing a feature and charging for it as an upgrade is ordinary product strategy. It is anticompetitive only if the point of the absorption was to stop that feature from being sold at a price that disciplined the platform. It is not the Cunningham pattern, in which the project disappears and the old product faces no new version of it.
The accurate description is a hybrid. The low-priced research brand was retired. The model-integration layer was kept and repriced. A maverick-firm merger case can be built on that hybrid. The 2010 Horizontal Merger Guidelines, which were the guidelines in force when this deal was signed, discuss maverick firms directly: a firm that constrains prices or innovation, whose removal matters even when the shares look modest. The December 2023 Merger Guidelines, issued four months after closing, say in Guideline 6 that a merger can entrench a dominant position by eliminating a nascent threat. Guidelines are not law, and the later ones did not govern the 2023 review. They are the vocabulary the complaint is using.
A pharmaceutical killer-acquisition case cannot be built on this deal without ignoring the product that is still on the price list. The complaint uses the label for the whole transaction. The proof, if it comes, will be about the research seat.
The multiple does less work than the complaint assigns it. In 2023, buyers paid high revenue multiples for generative-AI distribution and for scarce experience running a particular model on a particular corpus. A high multiple is consistent with buying a rival's disappearance. It is also consistent with buying eighteen months of GPT-4 practice, a handful of prompts the product team had not written, and a brand that had just been on national television. Wong's emphasis on prompts and time-on-model is the second story in operational detail. Motive is a document question. Effect is a price question. The complaint pleads both as if they were the same fact.
The market the complaint draws, and the products it leaves out
The pleaded product market is "Comprehensive Digital Legal Research Databases and Information Services" sold to legal professionals in the United States. The contents are a cross-referenced collection of federal and state case law, statutes, constitutions, regulations, and court rules, plus a citation-validation tool. The geographic market is the United States, because a lawyer practicing under U.S. law cannot substitute a foreign research service.
The complaint applies the Hypothetical Monopolist Test. A monopolist of that market, it says, could impose a price increase on the order of 5 to 10 percent without losing customers to Google or to a general-purpose chatbot. The reason is professional duty. A lawyer's obligation of competence is not met by an unverified web search or by a model that does not sit on a curated, updated, citator-checked archive. That argument is serious. Malpractice exposure is a real switching cost, and it is why "just use the free web" fails for a lawyer filing a brief tomorrow.
The test still asks what customers substitute, not what a slogan says they need. The complaint draws the boundary against Google and against consumer ChatGPT. It does not plead the products lawyers actually named when the shutdown email arrived.
Fastcase, including the bar-association plans that put a federal-and-state library in front of members at a price of zero or close to it. Fastcase says it partners with more than 80 associations. vLex, which acquired Fastcase in 2023. Bloomberg Law, with its own citator and a flat fee aimed at firms that do not want a transactional tariff. Lexis, which the complaint treats as a duopoly partner rather than as the nearest substitute. For a Westlaw customer shopping a renewal, Lexis is the obvious outside option. And the public-domain corpus itself. Judicial opinions are public. Courts and public archives distribute the text. What a century of publishing adds on top of that text — headnotes, a taxonomy, pagination, a citator — is a body of work. It is not, in our view, a copyright. The practical gap between a startup library and Westlaw's library is coverage and maintenance, which is a different problem, taken up below.
A market limited to malpractice-grade research with a full citator can be coherent. If that is the market, Casetext has to have been inside it. SmartCite was a treatment checker. It was not KeyCite. Wong's "smaller, less sophisticated" line is the acquirer saying the corpora were not the same product. The substitution fact the complaint needs is that a Westlaw price increase sent a material number of seats to Casetext, or that the credible threat of that switch changed renewal quotes. The complaint alleges that bargaining-chip use, including at firms that also kept Westlaw. It does not attach a single renewal memo.
There is a second market the complaint describes and then declines to plead. CoCounsel, in March 2023, was sold on top of research seats, including to firms that already paid for Westlaw. The complaint says firms bought it "on top of or in lieu of" existing platforms. "On top of" is a complement. "In lieu of" is a substitute. The class definition then excludes anyone who bought the AI tier, and the relevant market is defined as the database, not the assistant. The acquisition's stated rationale was the assistant. The damages class is buyers of the database. A complaint can challenge a merger in the market where the effects landed. It has to keep the markets straight when it argues that a 50-times-revenue price for an AI workflow is evidence of monopoly power over case-law seats.
The power allegation has the same split. The complaint says Westlaw and Lexis together have long held more than 80 percent of digital legal-research revenue, assigns Westlaw a nominal share of about 40 percent, and then says 40 percent understates power because large firms and law schools buy both. A win for Lexis is often not a loss for West. The industry calls that structure Wexis. Dual-homing is a real fact about Am Law procurement. It does not convert a duopolist into a monopolist. Section 2 of the Sherman Act reaches monopoly and attempted monopoly. A durable oligopoly with parallel pricing is a different problem, and it has a poor record as a single-firm monopolization theory.
The complaint's better path, which it also pleads, is direct evidence: a firm that can raise price, impose multi-year terms, and add annual escalators without losing the sale has power, whatever its share. The evidence offered is the old transactional tariff and the flat-fee escalators. The customers who dual-home and the solo whose only option was Casetext at $32.50 have different demand elasticities. Averaging them into one finding of monopoly power is the work an expert has to do.
Potential-competition cases fail when the threat is speculative. Casetext was not speculative. It had been selling research since 2013 and had a shipped assistant by March 2023. That distinguishes it from a firm that might someday have entered. It does not distinguish a research adjunct from a research substitute, which is the factual fight.
Four statutes, one industry structure
The complaint pleads four claims, all on behalf of baseline subscribers, all against both defendants. "Baseline" means a Westlaw plan that includes the research databases and does not include a separate generative tier such as CoCounsel. Westlaw Classic and Westlaw Edge are named. Anyone who bought a CoCounsel or Westlaw Precision AI bundle is out of the class.
Sherman Act § 1 reaches a contract that unreasonably restrains trade. Before the merger, Casetext and Thomson Reuters were separate firms and, on this market, horizontal competitors. A merger agreement between them is a contract. After closing, a parent and its wholly owned subsidiary are one enterprise for Section 1 purposes. Copperweld Corp. v. Independence Tube Corp., 467 U.S. 752 (1984). A decision in 2024 to turn the product off in 2025 is unilateral conduct. It is a Section 1 violation only if the decision was a term of the 2023 deal between firms that were still separate. The complaint says it was. The shutdown, standing alone, does not prove the term.
Horizontal mergers are judged under a structured rule of reason. They are not per se illegal. The Cartwright Act count floats "per se or rule of reason" as alternative labels for a combination that limits output or holds a premium. A merger is the second of those, if it is either.
Sherman Act § 2 reaches monopolization: monopoly power in a relevant market, plus willful maintenance of that power by anticompetitive conduct rather than by a superior product, business acumen, or historic accident. United States v. Grinnell Corp., 384 U.S. 563 (1966). Buying a nascent threat can qualify. United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001), treated conduct aimed at middleware that had not yet replaced the monopoly product, but that reasonably appeared capable of threatening it, as reachable under Section 2 when the conduct harmed the competitive process. That is the doctrine this complaint is reaching for when it calls Casetext nascent and technologically ahead, and calls the response a purchase instead of a build.
The complaint also calls the acquisition predatory. Predatory pricing has a specific test: prices below cost, plus a dangerous probability of recoupment. Brooke Group Ltd. v. Brown & Williamson Tobacco Corp., 509 U.S. 209 (1993). Paying a high price to buy a rival is the opposite pattern. The multiple may show what the buyer thought the threat was worth. It is not a below-cost pricing case, and the word does not make it one.
Clayton Act § 7 bars an acquisition whose effect may be substantially to lessen competition, or to tend to create a monopoly. The standard is incipiency. The plaintiff does not have to wait until the monopoly has arrived. Brown Shoe Co. v. United States, 370 U.S. 294 (1962). Private damages still require antitrust injury that has actually happened. Brunswick Corp. v. Pueblo Bowl-O-Mat, Inc., 429 U.S. 477 (1977). Paying more because a rival was removed is the core of that injury. Paying more because the remaining product is a different product is not.
The count alleges that the deal eliminated direct competition, removed a nascent rival that was gaining share, increased concentration in a duopoly, and prevented a price-and-innovation rivalry in legal software. It does not plead Herfindahl numbers. Under United States v. Philadelphia National Bank, 374 U.S. 321 (1963), a merger that produces an undue share in a highly concentrated market is presumptively unlawful. A complaint that skips the shares is asking the court to skip the presumption and decide a nascent-threat narrative. That narrative can win. It has to be proved with a market and with documents.
The damages limitations period is four years. The agreement was June 26, 2023, closing was August 17, 2023, and the complaint was filed October 1, 2026. The shutdown is the later act the overcharge theory depends on. Agency clearance in 2023, or the quiet expiry of a Hart-Scott-Rodino waiting period, is not a defense. The signing release said closing depended on regulatory approvals. A $650 million deal sat well above the 2023 size-of-transaction threshold of $111.4 million. It closed fifty-two days after announcement, which is what a file looks like when it does not draw a Second Request. The agencies were reviewing an acquisition of a going concern whose press release promised faster AI. They were not reviewing a customer email, nineteen months later, that set a date certain for deleting the cheap seat. Post-merger evidence of what the buyer did with the product is the reason this complaint exists in 2026.
The Cartwright Act, California Business and Professions Code § 16720 and following, is the state claim for a California subclass, with treble damages under § 16750. Casetext was a California company. Federal courts usually track the Sherman Act analysis. The complaint does not develop a California-only theory beyond the headquarters.
The prayer asks for class certification, treble overcharge damages, and an injunction ordering Thomson Reuters to divest, spin out, or re-establish Casetext and CoCounsel as an independent competitor, or else to stop enforcing mandatory annual escalators against baseline subscribers. Private plaintiffs may seek divestiture. California v. American Stores Co., 495 U.S. 271 (1990). A court that granted it would have to decide what "Casetext" still is: a 2023 code snapshot, a brand, a corpus the acquirer did not consider equivalent to its own, or a prompt library retrained on Westlaw content. The escalator injunction is the narrower request, and it has the causation problem already noted. A percentage that was normal before the deal is not, without more, a merger effect.
The class is Westlaw customers, not Casetext refugees
The nationwide class is everyone in the United States who held a baseline Westlaw plan, Classic or Edge or the like, at any time from June 26, 2023 through the present, and who did not add a premium generative tier. The California subclass is the same group with a principal place of business in California. The defendants' own people, anyone who bought CoCounsel or a Precision AI bundle, and the judges are excluded.
The plaintiff was a Casetext customer from 2018 until 2025, and a Westlaw customer only after the shutdown. The class is people who held baseline Westlaw from the signing date forward. The plaintiff is in the class because it holds Westlaw now. Its injury story is "I lost a $32.50 product and had to buy a $111.48 product." The class injury story is "Westlaw's own baseline price is higher than it would have been, because nobody can threaten to walk to Casetext anymore."
Those are different damages. Comcast Corp. v. Behrend, 569 U.S. 27 (2013), requires the model to measure the injury caused by the theory that is certified. "Casetext price versus this firm's Westlaw invoice" measures a switching loss for former Casetext customers. It does not measure an overcharge for a firm that was already on Westlaw and never held a Casetext quote. The classwide model has to be a before-and-after price on Westlaw's baseline SKUs, or a difference-in-differences against a control that faced the same cost shocks. Lexis prices over the same window are the obvious control and the obvious defense. If Lexis escalated in parallel, the Casetext event is harder to see. If Westlaw's residual widened after the shutdown announcement and Lexis's did not, the regression has a place to start.
A solo offered $111.48 for Classic is also not situated like an Am Law firm whose renewal is a negotiated stack of seats, content, credits, and concessions. The complaint's dual-homing discussion, offered to prove power, is a description of those large buyers. The class includes them, then excludes them if they added CoCounsel. Telling those two groups apart from billing records that bundle seats and workflow tools under customized names is the kind of individualized question that eats a damages class even when the theory is common.
Direct-purchaser standing is in better shape. A law firm that pays Westlaw is the purchaser the federal damages statute contemplates. Illinois Brick Co. v. Illinois, 431 U.S. 720 (1977); Apple Inc. v. Pepper, 139 S. Ct. 1514 (2019). The class is not pleading that the overcharge was passed on through a bar association or a reseller.
Numerosity will not be the obstacle. The complaint says tens of thousands of baseline subscribers. Ten thousand subscribers, fifty dollars a month of overcharge, for twelve months, is $6 million before trebling, which clears the Class Action Fairness Act's $5 million floor without any heroic assumption. The hard questions are whether the overcharge is common, whether this plaintiff is typical of Westlaw customers, and whether an economist can measure it without the model that died in the last antitrust case against this defendant.
Why Casetext's dataset was smaller
Wong told investors that Casetext had a research data set, and that it was "a smaller, less sophisticated one than we have." That sentence is the most important product fact in the deal. It is also easy to misread as a claim that Westlaw owns the law, or owns the notes it has layered on the law. It does not. Judicial opinions are public. Wheaton v. Peters, 33 U.S. (8 Pet.) 591 (1834), and Banks v. Manchester, 128 U.S. 244 (1888), are why a publisher cannot copyright the text of a decision. Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), is why the labor of gathering public material does not, by itself, create a copyright in the collection. LawDiver does not agree that headnotes, the Key Number system, star pagination, KeyCite, or anything else Westlaw adds on top of the opinions is the subject of a valid copyright. We are not going to litigate that question in this post. The size gap is real for a reason that has nothing to do with ownership.
Compiling a case-law file as large as Westlaw's is brutally hard. We know, because we are in the middle of doing a version of it.
The raw material is not a database someone hands you. It is every appellate opinion, a large share of trial opinions, the unpublished dispositions that never received a reporter page, the slip opinion posted the morning of decision, and the bound volume that replaces the slip six months later with different page breaks. The same decision arrives three or four times, under different captions, from a court site, a bulk archive, and a regional reporter. Parallel citations do not mathematically convert into each other. 370 U.S. 294 does not compute 82 S. Ct. 1502. Someone has to have stored both, or the citator and the brief will disagree about whether they are the same case. Deduping that pile, attaching the right court and the right date, and throwing away the copy that is a reprint of a reprint, is most of the job. Doing it once is a backfile project measured in years. Doing it every morning, for every court that released something overnight, is the product.
A citator is the same job a second time. Each opinion cites other opinions. Those citations have to be found in the text, resolved to the right decision rather than the decision with a similar volume and page, and marked for what the citing court did — followed, distinguished, overruled, or merely mentioned. A national file on the order of ten million opinions carries a citation graph on the order of a hundred million edges. Miss the edge and the treatment signal is wrong. Miss the opinion and the edge has nowhere to land. Westlaw's file is larger than a case-law corpus on top of that, because a research seat also expects statutes, regulations, court rules, and a century of secondary sources kept current. Casetext, in a decade, built a federal-and-state case library good enough that thousands of firms used it as a seat and that Am Law firms added it beside Westlaw. It did not build the rest. Wong was describing that gap. A startup with a few dozen million dollars of venture capital and ten years does not finish a file that a publishing house has been extending since the nineteenth century. The check for $650 million bought the assistant and the head start on the model. It did not buy a second Westlaw, because Casetext had not built one.
That is also why "smaller" and "not a substitute" are different claims. A smaller library can still be the library a solo uses every day, and the library a large firm threatens to move work to. The complaint's market definition depends on that second claim. The first claim, the one Wong made, is the engineering truth: comprehensive, in this industry, means you closed the gaps, and closing them is the work.
The other consolidation
A duopoly paragraph written as if the vendor list stopped at Westlaw and Lexis does not describe 2026. The complaint can still be right that, for a particular buyer with a brief due tomorrow, those two seats are the ones that feel safe. The supply side has been consolidating on a parallel track, and one of those deals is already the subject of a Section 7 pleading.
vLex acquired Fastcase in 2023, the same year as the Casetext deal, and with Fastcase came Docket Alarm. In June 2025 Clio agreed to buy vLex for $1 billion, practice management buying a research library and the Vincent assistant. That deal closed in November 2025. In January 2026 Alexi Technologies answered a Fastcase suit with counterclaims that treat the Clio combination as a Section 7 violation in a market for comprehensive legal databases and a market for AI legal research. Alexi's market, as described in the pleading coverage, counts three comprehensive databases: Westlaw, Lexis, and Fastcase/vLex. That is a different market from the two-firm market in Rubin. Both are pleadings. The useful point is that "the" legal-research market changes with the pleader. A small-firm plaintiff who lost Casetext sees a duopoly. An AI company that needs a licensable corpus sees three libraries and a foreclosure story about the one that will still license.
Bloomberg Law remains a full-price incumbent for firms that want dockets, a different publisher's analysis, and a citator that is neither KeyCite nor Shepard's. Harvey and the other assistant-only tools sit where Wong put them on the 2023 call: model access without an authoritative corpus. They compete with CoCounsel. They do not, without a corpus and a citator, compete with the seat the complaint says was destroyed.
"No viable, comprehensive independent alternative," which is the complaint's causation sentence, is an allegation about one firm's choices in March 2025. Fastcase through the bar, vLex, Bloomberg, Lexis, and the public archive are the set a cross-examination will walk, one product at a time, asking what was priced and what was rejected. For the industry, the sentence is still the right question. If those products were real substitutes at the small-firm price point, Casetext's disappearance is a product retirement. If they were not, because citator quality, coverage, or malpractice risk kept them outside the consideration set, then deleting the one product inside that set is a structural event, and the invoice is evidence rather than an anecdote.
What it means if you build or buy legal research
The complaint is a theory of three layers, and the theory is useful even where it is unproven.
The corpus is public, and it is still enormous to build. Primary law is public. A company can assemble a large share of federal and state opinions from courts and from public archives. Casetext did that, over ten years, well enough that thousands of firms used it as their research seat and well enough that Am Law firms added it beside Westlaw. "Someone would have to digitize millions of opinions" is a 1990s description of the barrier. The 2026 version is meaner: the texts exist, and turning them into one deduplicated, citator-checked, current file on the scale lawyers expect from Westlaw is a decade of pipeline work. Casetext got partway. That is the smaller dataset. Nothing in that gap is a copyright we recognize in Westlaw's editorial layer.
The citator is the part that takes the longest. KeyCite, Shepard's, and a classification scheme are why a cheap library does not automatically clear the Hypothetical Monopolist Test. Lawyers do not pay for a login so they can read Brown Shoe. They pay to know what has happened to it, and to find the next case in the same line. Those signals are produced by reading opinions and recording how they treat one another. SmartCite was a thinner version of that graph. Building the thick version means walking the citation edges, not taking a license to someone else's notes. The complaint's market definition stands or falls on whether buyers treat that gap as too wide to switch.
The assistant was the complement that got repriced as a platform feature. CoCounsel's value, on the buyer's own account, was the combination of early model access and a legal corpus, not the corpus alone. Once that combination sits inside Westlaw, the assistant is no longer an independent constraint on the price of the seat. It is a reason the seat costs more, and a reason the class definition has to exclude the people who bought it. For buyers, the practical change is that the cheap way to get generative legal work was a separate subscription, and the current way is an attachment to the incumbent. For builders, an assistant with no citator and no corpus is what Wong called "LLMs kind of by themselves," and he did not think that was the asset worth $650 million.
Small-firm pricing was the competitive constraint, not small-firm market share. Casetext did not need 30 percent of industry revenue to matter. It needed to be the credible place a renewal conversation could go, and the actual place a solo could go. The complaint's 40 percent share figure is almost beside the point. The maverick fact, if it is a fact, is the existence of a published price far below the incumbent's floor. Removing a published price changes negotiations even when most customers never leave. That is the allegation in the complaint's paragraph about bargaining leverage, and it is the allegation that does not depend on proving that Westlaw is a monopolist rather than a duopolist.
Switching costs were enforced with the file, not only with the contract. Multi-year terms and 4 percent escalators are the contractual half. The September 2024 email's statement that saved work product would not survive the shutdown is the technical half. A research product that stores the lawyer's organization of the law can raise the price of leaving without changing the monthly rate at all. Anyone who holds a library of user work inside a research login is holding the same lever. The antitrust question is whether it was pulled in order to move a captive installed base onto a higher price, or as an ordinary incident of retiring a codebase. The complaint says the first. The customer email's tone says "more robust tools on Westlaw." The timing, nineteen months after the check cleared, is what a jury would be asked to interpret.
Divestiture is a software problem before it is a decree. Ordering CoCounsel and Casetext back into the market means identifying code, weights or prompts, a corpus, a brand, and a team. The team has already scattered. The corpus was, on the buyer's account, the lesser one, and it has been pointed at the greater one for three years. A spin-out that has to lean on Westlaw's file recreates the dependency the merger created. A spin-out that does not is a 2023 startup with a 2026 model bill, staring at the same assembly problem Casetext spent a decade on. Damages, if the overcharge can be measured, are the remedy that matches the pleading. Structural relief matches a market that still contains a separable Casetext. That market has to be proved to still exist.
The agencies' silence in 2023 is not the industry's answer. A reportable deal that closed in a summer is a deal the government did not sue over. It is not a finding that retiring the cheap seat in 2025 was the competitive effect of a lawful integration. Builders who assume "the DOJ cleared it" as a description of the market are reading a waiting-period expiry as a market study. The market study is the price list before and after April 1, 2025, firm by firm, against Lexis and against whatever bar-plan library the same lawyers could have used. That study is not in the complaint. It is the study the complaint is a request for.
If the six propositions close — a market that contains both products, power or a maverick constraint, documents showing the constraint was real, a price path that moves with the shutdown, a classwide model that measures Westlaw's price rather than one refugee's bill, and a remedy that still corresponds to an asset — then the hybrid shape of the deal supports a real Section 7 case and, with a real power finding, a Section 2 case. If they do not, the complaint remains an accurate description of a price a solo did not want to pay. The distance between those two readings is the distance between a product retirement and a change in the structure of legal information. The pleading argues they are the same event. The next two years of this case are about whether the price list agrees.
Disclosure
LawDiver sells legal research and a caselaw API. We compete for lawyers who might otherwise pay a legacy research vendor, and we build the case-law file this post describes, which is why the size comparison is not a theory to us. We do not agree that anything Westlaw adds to the public opinions is validly copyrighted. This is a reading of the complaint and of the public record of the deal and the shutdown. It is not a prediction and not legal advice. The complaint is here.