The Commons Your Practice Runs On

Over two days this week, this firm researched, drafted, and published thirty long-form articles on Minnesota law. Every statute quoted from raw Revisor HTML. Every case read in full before it was cited. Every quotation checked character-by-character against the source text, and then checked again by a second reviewer who did not write the article. Marginal cost from commercial legal research databases: zero.

I have written about that already as a purchasing argument, and it is a good one. But there is a second invoice that nobody writes, and I want to write it here, because I came out of those two days with a much more precise sense of what was actually holding up the floor.

Four institutions did. A fifteen-year-old nonprofit. A research lab at a law school that has moved on to other work. A statutory office in the Minnesota legislative branch. A legislative-branch federal agency that has been printing government documents since 1861. None of them sent me a bill. Three of the four have no mechanism to send me one.

The thesis: the free-law commons is now load-bearing for real small-firm practice — not a nice supplement, the actual foundation — and it is fragile in ways that are specific, observable, and already showing. Firms that build on it should fund it the way they fund anything else they cannot practice without. Not as philanthropy. As a maintenance line item, next to malpractice coverage and the e-filing subscription.

Here is what I mean by fragile, with the receipts.

The archive stopped in 2019

The Caselaw Access Project is the reason a solo firm can read a 1955 Minnesota Supreme Court opinion at two in the morning without a library card. Harvard Law School Library digitized roughly 40 million pages from about 40,000 bound volumes and put the result online in a consistent, machine-readable format. It is one of the great acts of institutional generosity in the history of this profession.

It is also finished, and CAP says so plainly. From its own About page: the Library Innovation Lab “transitioned case.law to a static state” because the industry agreements that funded the access layer “have now expired,” and LIL “is primarily a research organization.” The scope statement is equally direct — CAP “includes all official, book-published state and federal United States case law through 2020,” and “The Harvard Law School Collection includes volumes published through 2018.” Everything from 2018 to 2020 came in later as a donation from Fastcase.

Now the part you only find by going and looking, which I did, because a scope statement is a claim and not a measurement.

CAP’s static archive serves the Northwestern Reporter, Second Series through volume 932. Volume 933 returns a 404. The most recent Minnesota decision inside volume 932 is dated September 4, 2019. There is no nw3d directory at all — the entire Third Series, which is where Minnesota’s recent appellate law lives, is simply not there.

For a Minnesota practitioner that is not an abstraction. It means that for any Minnesota appellate decision issued after roughly the fall of 2019, there is no free source that will give you the opinion text with the reporter pagination attached. You can get the opinion — the Judicial Branch posts slip opinion PDFs, and our pipeline pulled them and read them in full — but a slip opinion has no volume and no page. During those two days the practical consequence was concrete and repeated: articles citing recent decisions had to cite them by case name, docket number, and date, with no N.W.3d cite, because asserting a reporter page we could not verify would have been guessing dressed as scholarship. In one instance we resolved a recent case’s reporter cite only because a later opinion quoted the citation string verbatim inside its own text, and we could read that later opinion.

The maintainers describe this failure mode better than I can. Free Law Project, announcing in April that it is picking up the scanning where Harvard stopped, explained exactly why its 200-odd court scrapers do not solve the problem: because they scrape court websites, “it does not get the official pagination or citations for the decisions… until the content is in a book, there’s no volume or page to cite to.”

That is the shape of the crack. It is not that the free corpus is unreliable. It is that it has an edge, the edge is dated, and the edge sits precisely where a litigator most often needs to be.

The scan is not the opinion

The second fragility is inside the data rather than at its border, and it is the one most likely to put a bad quotation in a brief.

CAP’s own data-quality statement: “Case text and general head matter has been generated by machine OCR and has not received human review.” Metadata — party names, docket number, citation, date — got human review. The words of the opinion did not.

In practice that means the archive is honest and slightly broken, and you must build for it. Actual corruption our drafters encountered and documented over two days, in cases they were quoting:

• eoncededly for concededly
• herjself for herself
• UCG for UCC
• re-suits obtained for results obtained
• Due Pi'ocess
• plaintiffs reputation — the apostrophe silently gone
• Ateas for Atcas, a party name
• Colder for Calder, in a citation
• modem trend for modern trend

Read that list again and notice that the last three are the dangerous ones. herjself announces itself. Colder v. Jones does not — it is a real English word in a plausible position, and a drafter who is pattern-matching rather than reading will sail straight past it and put a misspelled party name in a brief. modem trend is worse, because it is exactly the kind of thing a spell-checker approves.

The disciplines that fall out of this are simple and non-negotiable, and they are worth stating because they generalize to any OCR corpus:

1. Never quote a corrupted string. If the sentence you want carries an artifact, either quote the clean portion or paraphrase. One of our articles paraphrased an eminent-domain fee-factor list rather than quote re-suits obtained.
2. Never silently repair inside quotation marks. A reader who checks the source and finds different words has caught you altering a quotation, which is a much worse look than a paraphrase.
3. Never take a number out of OCR’d body text. Reporter volumes and page numbers that appear inside an opinion’s own citation strings are OCR output. A stray digit there produces a citation that is wrong in the most undetectable possible way. Take numerals from structured metadata, or from star-pagination anchors, or not at all.
4. Expect corruption near a clean quote. More than one of our verification passes had to warn the next reviewer that OCR damage two paragraphs away from a quoted sentence is not evidence the quote is wrong.

None of that is a complaint about CAP. It is the correct way to use a scanned archive, and CAP told you it was scanned.

The state’s own websites will not let a machine in

The third fragility is the one I did not expect, and it is the one where the profession’s own institutions are the problem.

Minnesota’s Office of the Revisor of Statutes is superb. Over two days, revisor.mn.gov served us raw statute HTML, court rules, and struck-and-underscored session laws to an ordinary command-line request with a browser user-agent, hundreds of times, without a single block. Session laws in amendment markup let us catch 2026 changes that the codified statute pages did not yet display. That is what it looks like when a public body actually publishes.

Then there is everything else in Minnesota state government.

• mn.gov — the executive-branch portal, and the host of the State Law Library’s slip opinion archive — sat behind a Radware bot-manager CAPTCHA. Our drafters could sometimes retrieve an individual PDF at a guessable URL, but the archive’s directory index was closed. When we needed to know whether the supreme court had decided a case since granting review, we could not read the index; we probed 462 candidate filenames across 231 dates and reported the result honestly as filename-pattern evidence rather than an index read.
• sos.mn.gov returned a Radware interstitial, so an article about LLC statements of authority could not confirm the Secretary of State’s current filing fee. It quoted the statute’s own dollar figure and told the reader to confirm the current amount.
• mncourts.gov returned HTTP 403 behind a JavaScript challenge — for the HTML page and for the linked PDFs. The casualty there is worth naming: two articles about prejudgment interest could not retrieve the State Court Administrator’s published annual interest rate. So they state no rate. They explain the statutory mechanism verbatim and direct the reader to look up the year’s published figure themselves. That is the single number a reader most wants, and it is missing, and it is missing because the court system’s own website would not serve its own published notice to a program.

I want to be fair about this. Bot protection is a rational operations decision made by people defending real infrastructure from real abuse, and nobody at those agencies decided to shut lawyers out. But there is a cost, it lands on exactly the people the agency exists to serve, and no one on the agency side is measuring it. A public record that a machine cannot read is, in 2026, most of the way to not being public. The distinction that matters is no longer whether a government body posts its documents. It is whether it publishes them — addressable, stable, machine-readable, and served to anyone who asks.

Nobody is telling you it was overruled

The fourth gap is the one I keep returning to because it is the honest limit of everything above.

A verification pipeline built entirely on free sources can establish, to a very high standard, that a case exists at a given citation, that it says what you claim it says, and that your quotation is verbatim. It cannot tell you the case is still good law. There is no free citator. Not a thin one — none.

Two things follow that are easy to get wrong.

First, CourtListener’s coverage claim is precise and you should read the qualifier. Free Law Project states that its database “encompasses more than 99.9% of all precedential legal case law published in the United States.” Precedential. Unpublished intermediate-appellate decisions — which in Minnesota practice are cited constantly, with the usual caveats — are outside that number. Which means the tool not finding a case is not evidence the case is fake. Our own testing of the CourtListener interface surfaced three failure modes worth knowing: citation analysis that deduplicates by reporter cite can hide a citation bolted onto the wrong case name; citations broken across a hard line wrap are silently dropped and never extracted at all; and a citation resolving successfully proves only that the reporter cite is real, not that the proposition or the quotation is. Every one of those failures is silent. None of them throws an error.

Second, CAP cannot help with treatment even in principle. It is frozen at 2020. A frozen archive is definitionally incapable of telling you what happened to a case last Tuesday.

So the honest inventory of what a small firm can and cannot do without a subscription is: existence, yes; text, yes; verbatim quotation, yes; pin cite, yes for anything pre-2019 with star pagination; subsequent treatment, no. Treatment-checking remains a separate, paid verification step, and a firm that tells itself otherwise is running an uninsured risk. That sliver is genuinely hard editorial work performed continuously at corpus scale, and the commercial vendors have earned it.

The maintenance bill exists either way

Now the part this essay exists for.

Free Law Project is a 501(c)(3), founded in 2010, and it runs CourtListener, the RECAP Archive, the oral argument collection, the judges database, and open-source tools — eyecite, juriscraper, x-ray — that a good deal of the legal technology industry quietly builds on. Its tax returns are public. It publishes what it costs.

And it is, right now, fixing the exact gap I described three sections up. In April it announced that it is scanning the post-Harvard books itself — “picking up where Harvard left off,” to “fill the gap between when the Harvard data ends, in 2018, and today.” As of that announcement: over 200,000 pages scanned, an immediate goal of 2.5 million pages, and nearly two million pages still lacking high-quality scans and complete metadata. The announcement is candid about the constraint: “Scanning and processing at this scale requires sustained funding.”

Read that sentence as a practitioner. The single most consequential defect in the free legal research stack — the reason your recent authority has no pin cite — has a known fix, an organization executing it, a published page count, and a funding gap. That is not a lament. That is a purchase order waiting for a signature.

The mechanisms already exist and they are priced for small firms, not for institutions:

• Individual memberships start at $10/month or $100/year, running up to $100/month. They are not symbolic — API rate limits scale with the tier, from 5 requests per minute on the free tier to 25 per minute at the top. If your assistant queries CourtListener a hundred times in an afternoon, as mine does, you are consuming capacity somebody pays for, and the membership is the meter.
• Group memberships for firms start at $30/month for one to five seats and scale to organizations of a hundred.
• Legal services organization memberships are free — civil legal aid offices and clinics get group access at no cost, which is the access-to-justice commitment made structural rather than rhetorical.
• The Justice Partner Circle, launched in September 2025, is a program built specifically for law firms. Somebody at Free Law Project already did the work of making it easy for a firm to say yes.

The state side has no donate button, and it needs a different kind of attention. Minnesota’s Revisor operates under Minn. Stat. ch. 3C, and the first sentence of § 3C.03 is one every Minnesota lawyer should have read: “As far as personnel and available appropriations permit, the revisor’s office shall perform the drafting duties described in subdivisions 2 to 4.” The office’s powers are exercised “subject to the control of the Legislative Coordinating Commission.” That is the legal infrastructure of the entire Minnesota statutory corpus, and the statute itself tells you it runs on an appropriation. At the federal level, govinfo is a service of the Government Publishing Office — a legislative-branch agency — which applies digital signatures to its PDFs “so users can verify documents have not been altered and are the official versions,” and runs an ISO 16363 certified preservation repository. That is a budget line too.

So the ask has two halves and they are different. Fund the nonprofits with money. Defend the public offices with the thing lawyers actually have, which is a voice in front of legislatures and courts about what public legal information should cost and how it should be served.

What would change my mind

A claim without failure conditions is a mood, so here is mine stated so it can be checked.

The claim: every leg of this stack is maintained by an organization for which maintaining it is not the revenue-generating activity, the profession extracting the value contributes to it in a way that is — as far as I can tell — negligible, and the failure mode will not be collapse. It will be drift. Gaps that stay open. Coverage that stops at a date and stays stopped. A bot shield that goes up and never comes down. Nothing dramatic enough to make anyone act.

Four things would prove me wrong, and I would be pleased by all of them:

The scanning project lands. If Free Law Project hits its target and post-2018 reporter citations become routinely resolvable from a free source, the most acute defect closes and my “gaps stay open” prediction is simply false. They have a plan, a tool, and a page count. Watch the number.

Neutral citations spread. About twenty states already have them. A neutral citation moots the entire pagination problem — no book required, no scanner required, a citable identifier at the moment of issuance. If Minnesota adopts one, most of section two of this essay becomes a historical note.

A free citator ships. This is the one I consider least likely, because the work is editorial and continuous and expensive. But CourtListener already maintains the citation graph that a citator is built on top of. The distance from “here is everything that cites this case” to “here is how they treated it” is real, and it is not infinite.

A commercial vendor sells cheap metered API access. If the incumbents ever price the citator and the editorial layer as infrastructure rather than as seats, the commons stops being load-bearing for firms that can pay, and my thesis narrows to firms that cannot. I have argued elsewhere that they should, and that the first one to do it wins.

Absent those, the prediction stands, and it is not a prediction about technology. It is a prediction about who pays for maintenance.

The line item

Here is the accounting I actually came away with.

Two days. Thirty articles. Every statutory quotation verbatim from a state office funded by an appropriation. Every case read from an archive a university built and then stopped building. Federal materials from a printing agency Congress established in 1861. Case law search and citation resolution from a nonprofit with a donate button. Marginal cost paid to any of them for any of it: nothing. None of them sent a bill, because none of them can.

That is not a scandal. It is the entirely ordinary posture of a profession that has quietly reorganized its research function around infrastructure it does not fund. Nobody decides to free-ride; you just use what is there, which is what everyone does with a commons right up until the moment the commons is in trouble.

The correction is unglamorous. Put a number in the annual budget. It does not have to be large — the individual tiers start at ten dollars a month and the firm tiers at thirty, and a solo practitioner who drops a legal research subscription is freeing up multiples of that. Route it to the organizations whose data your systems actually touch, in rough proportion to how hard you lean on them. Treat the receipt as an expense of doing business, because that is precisely what it is.

And the access-to-justice thread that runs through everything in this section runs through this one hardest. The reason a solo lawyer can now verify authority like a firm with a research department is that a handful of people decided public law should be publicly computable and then did the enormous, boring work of making it so. Every dollar of research overhead a small firm sheds because of that work is a dollar that does not have to come from a client who could barely afford the representation. That saving is real, it is being passed to clients right now, and it was financed by somebody else.

The bill for maintaining it exists whether or not lawyers pay it. Somebody is paying it today. It should be us.

Sources

• Caselaw Access Project — About (page content served from case.law/templates/cap-about-page.js; retrieved 2026-08-20) — “transitioned case.law to a static state”; agreements “have now expired”; “CAP includes all official, book-published state and federal United States case law through 2020”; “The Harvard Law School Collection includes volumes published through 2018”; Fastcase collection covering N.W.2d through 2020; “Case text and general head matter has been generated by machine OCR and has not received human review”; ~40 million pages from ~40,000 bound volumes
• CAP static archive — coverage limits verified directly against the archive’s own files on 2026-08-20: VolumesMetadata.json (no publication_year later than 2018; spine_end_year 2019 on the final N.W.2d volumes); static.case.law/nw2d/ directory index (volumes 56–932); nw2d/932/CasesMetadata.json (latest Minnesota decision date 2019-09-04); static.case.law/nw2d/933/ and static.case.law/nw3d/ both return HTTP 404
• Harvard Library Innovation Lab — Caselaw Access Project — digitization carried out 2013–2018 with Ravel Law; over 6.5 million decisions
• Free Law Project — Free Law Project Is Scanning America’s Case Law (April 16, 2026) — “picking up where Harvard left off”; gap “between when the Harvard data ends, in 2018, and today”; 200,000+ pages scanned, 2.5 million-page immediate goal, nearly two million pages still lacking high-quality scans; ~200 court scrapers and why scraped opinions lack official pagination (“until the content is in a book, there’s no volume or page to cite to”); about 20 states publish neutral citations; “Scanning and processing at this scale requires sustained funding”
• Free Law Project — About — “Started in 2010, Free Law Project is the leading 501(c)(3) nonprofit using technology, data, and advocacy to make the legal ecosystem more equitable and competitive”
• Free Law Project — Support — tax-deductible donations; “our tax returns are available for public inspection”
• Free Law Project — Membership — individual tiers $10/$25/$50/$100 per month; API rate limits by tier (free tier 5/minute, top tier 25/minute); group memberships from $30/month for 1–5 seats up to $2,500/month for organizations of 100; free Legal Services Organization group memberships
• Free Law Project — Announcing the Justice Partner Circle (September 30, 2025) — law-firm partnership program
• CourtListener — Data Coverage: Case Law (FLP Wiki) — “more than 99.9% of all precedential legal case law published in the United States”; over nine million decisions from over two thousand courts; direct court-publishing partnerships beginning 2020
• Minn. Stat. § 3C.03, subd. 1 — “As far as personnel and available appropriations permit, the revisor’s office shall perform the drafting duties described in subdivisions 2 to 4.” (2025 Minnesota Statutes)
• Minn. Stat. § 3C.02, subd. 5 — revisor’s powers “subject to the control of the Legislative Coordinating Commission” (2025 Minnesota Statutes)
• govinfo — About — service of the U.S. Government Publishing Office, a legislative-branch agency established 1861; digital signatures applied “so users can verify documents have not been altered and are the official versions”; ISO 16363 certified Trustworthy Digital Repository
• The retrieval failures, OCR artifacts, and citation-tool failure modes described above are this firm’s own operating records from its 2026-08-18/19 publishing run, documented in the per-article source manifests and independent gate reports maintained in this site’s repository. No client matter, client name, or client data is involved in any of them.

Commentary on legal research infrastructure and the business of law; the opinions and the prediction are the author’s. Not legal advice, not ethics advice, and not a recommendation about any particular firm’s research stack — in particular, nothing here should be read to suggest that free sources substitute for checking whether an authority is still good law, which remains a separate verification step this firm sources separately. No client information appears in this article. Everything in this section is free; the only thing asked in return is that you use it to serve someone who could not otherwise afford a lawyer. Questions about anything here: Send us a message or 612-470-6529.