Guest Post: The Caselaw Access Project — Then, Now, Tomorrow

On my left: the sting of the off-ramp, a modest guardrail, and a fifty-foot drop. On my proper, inching nearer: a tractor-trailer decided to occupy my lane. I hit the brakes. The truck saved rolling. Its wheels pressed into my automobile because it wedged me towards the curb and carved a tail-to-nose dent in my poor Toyota.

This was early 2015, on my commute to Cambridge, Mass., the morning of a essential assembly at Harvard Legislation College, the place I labored. Harvard professor Jonathan Zittrain and l have been sitting down with Daniel Lewis and Nik Reed, the founders of a authorized analysis startup named Ravel Legislation, together with attorneys from Harvard’s Workplace of Basic Counsel, Debevoise & Plimpton and Gundersen Dettmer. We’d all been working for over a 12 months on a contract that may make it doable, sometime sooner or later, for everybody to have free and open entry to all of the official courtroom choices ever revealed in the USA. After an exhausting 12 months of negotiations, it was time to lock ourselves in a room and work out if we had a deal.

Quick ahead 9 years, and that “sometime sooner or later” lastly is right here. On March 1, 2024, our collective efforts on this undertaking — the Caselaw Access Project — culminated within the full, unrestricted launch of practically 7 million U.S. state and federal courtroom choices representing the majority of our nation’s frequent legislation. I had the privilege to steer this work at Harvard for nearly eight years. Wrecked Toyota apart, it was a career-defining expertise, and I’m immensely grateful to everybody at Harvard and Ravel who labored laborious to make it doable.

To mark the event, I wished to share a few of the undertaking’s inside story, replicate on its affect and stay up for what I hope this information will make doable sooner or later.

Why Even Do This Challenge?

Courtroom choices are public data — they’re authored by judges and issued publicly to inform us what the legislation is, and why. All of us ought to have free, quick access to the legislation, and nobody ought to achieve aggressive benefit from having privileged entry to the legislation itself.

However traditionally we’ve not handled the legislation this manner. As a substitute, we’ve acted like our legislation is created and owned by the businesses that publish it. Our courts, with few exceptions, have allowed publishers to regulate entry to the legislation and to dictate how we learn, examine, cite and use the legislation. Naturally, publishers have prioritized their business pursuits. They’ve made the legislation scarce and costly. The impact has been to stifle innovation and competitors within the subject of authorized data and, I might argue, to impede justice and the rule of legislation.

Harvard scanned 38.6 million pages from 39,796 books and transformed all of it into machine-readable textual content recordsdata.

That is why the Caselaw Entry Challenge wanted to occur and why it was price doing, even with all of the obstacles, frustrations and compromises alongside the best way.

Let’s Make a Deal

I interviewed to hitch the Harvard Legislation Library and handle the undertaking in late 2013, a few 12 months after Nik, Daniel and Prof. Zittrain (or “JZ” as he’s affectionately recognized) had hatched the concept for the undertaking and began figuring out a skeletal framework for a possible deal.

I’ll confess: once I first discovered the undertaking wouldn’t be paid for instantly by Harvard, however as an alternative could be funded by a venture-backed Silicon Valley startup that may get a number of years of particular entry in return, I nearly bailed. I assumed it was absurd. Why would Harvard depend on a fledgling startup for this, particularly at the price of limiting entry?

By the point we’d organized ourselves round a convention desk in early 2015, I had a unique perspective. I’d spent the final 12 months negotiating the cope with Daniel and Nik but in addition with Harvard’s many inside stakeholders. I’d come to grasp that whereas Harvard’s librarians and assets made the undertaking uniquely doable, _Harvard’s forms and wealth additionally made the undertaking just about inconceivable. It was solely by a succesful companion like Ravel that the undertaking had an actual likelihood.

I’d additionally seen that Daniel, Nik and the Ravel workforce weren’t in it for purely business causes. Though our workforce knew we needed to give Ravel a number of privileged years to take advantage of the undertaking’s information, we drove a tough cut price to make sure the undertaking would serve the pursuits of students and researchers and the broader public. Most significantly, we had to make sure that if (or when) one of many huge publishers purchased Ravel, their acquisition wouldn’t undermine the undertaking’s targets. We had to verify a purchaser could be locked into persevering with to help the undertaking and would haven’t any energy to cease it. We have been coping with Ravel, however we have been additionally negotiating towards Ravel’s future purchaser.

This led us to push for a battery of onerous protections and commitments. Ravel’s acceptance of those phrases made clear to me that even inside the context of their business targets, they shared the general public curiosity motivations of the undertaking. Most authorized tech startups make daring declarations about public curiosity, entry to justice and democratizing the legislation when it fits them. Only a few make company-defining commitments that put these priorities entrance and heart.

Case legislation books ready to be scanned.

In the end, by mid-2015, the deal had taken form. Harvard would contribute the legislation books and run the scanning course of contained in the legislation library. Ravel would pay for the scanning and subsequent information processing, together with redaction of any extraneous materials that didn’t originate from the courts. Each Harvard and Ravel would get entry to the processed information. Harvard would have the suitable to share the info on a restricted foundation instantly. Ravel could be obligated to supply public entry from day one and would put its supply code in escrow to safe this obligation. In trade, Ravel would get an unique proper to take advantage of the info commercially for roughly six years after we completed digitization – till March 2024. If Ravel or its successor ever stopped offering public entry to the info, they might lose their business benefit and all the info would go free.

The contract nonetheless took a pair extra months to finalize. There have been different phrases that have been essential to the buyers and college directors we would have liked to approve the deal. There have been a number of dicey moments the place it regarded like every little thing may crumble over trivial issues. However lastly, we closed the deal and the signature pages hit my inbox. A short time later, we publicly announced the project and the key terms of the deal.

Then got here the true work.

Making Mass Digitization Work Contained in the Legislation Library

Contained in the library, we’d been eagerly gearing up for the digitization effort. In parallel with the negotiations, we’d run a proof of idea that allowed us to determine the method, tools, methods and staffing we’d want to fulfill our high quality requirements. We’d rigorously modeled out the prices and timing. We knew precisely what number of pages per day we might scan, how a lot it might price and what dials we might flip to change price or throughput if wanted.

When the deal closed, we have been able to go. We’d already tackled lots of the hardest challenges:

• We didn’t know exactly which books to scan. There was no definitive listing of “all of the books containing official courtroom choices.” So we did analysis and made one.
• We didn’t have all of the books we would have liked. Like many legislation libraries, we had stopped shopping for a few of the books that contained official courtroom choices. They have been too costly, and nearly nobody ever used them. So we went out and purchased books to fill the few gaps.
• A lot of the books weren’t bodily within the library constructing. They have been 30 minutes away in the Harvard Depository, the place they have been blended in with about 10 million different books. So we found out get the 40,000 books we would have liked and transfer them over to the library effectively.
• We had nearly no book-level metadata, however we would have liked to report key details about each guide, resembling when it was revealed and what jurisdiction(s) it lined. We additionally wanted to verify there have been no lacking or broken pages. So we created a course of to visually examine each guide and to manually report the mandatory metadata.
• To scan the books at excessive pace, we would have liked first to free the pages from their binding. So we purchased a machine we referred to as the “Guillotine,” which sliced by the backbone of the books with a crashing thud. (Sure, there have been bodily security concerns). The Guillotine was so heavy we needed to put it on a strengthened ground. It was so loud we needed to droop work round examination time.
• The high-speed scanner was an incredible machine, nevertheless it wasn’t good, and so we needed to do high quality management on the scans to verify they met our requirements. Over the course of the scanning effort, we visually inspected roughly 20 million scanned pages.
• After scanning, we needed to protect the books, simply in case we would have liked to scan them once more or somebody wanted to reboot democracy. So we used a vacuum-sealing machine designed for meat-packing to individually seal each guide right into a moisture-resistant bag earlier than transport all of them to an underground limestone mine in Kentucky.
• We needed to discover area for all this within the library, the place college students studied, school labored and librarians served. We needed to put the metadata stations, the Guillotine, the scanner, and the vacuum-sealer in separate areas, on numerous flooring, which meant our workforce needed to bodily transport small carts filled with books between stations on an elevator.
• And eventually we needed to preserve observe of all 40,000 books each step of the best way, so we might account for every one, constantly monitor our progress and confirm that we had processed each guide we would have liked to. So we constructed customized software program and tailored a hand-scanner system so we might verify in each guide at every station.

Overcoming these sensible challenges was the toughest work we did, and the success of this section was due totally to the professionalism, dedication and flexibility of the library team within the face of fairly a little bit of strain and skepticism, together with from inside Harvard. There have been no high-paid consultants, distinguished thought-leaders or pompous muckety-mucks telling us how to do that. Principally it was only a bunch of library professionals, a programmer and a token overbearing lawyer rolling up our sleeves within the basement and striving collectively to determine it out as a result of we cared. Actual innovation.

How Imperfect Legislation Turns into Imperfect Knowledge

Scanning was the toughest factor, nevertheless it wasn’t the one factor. We additionally needed to remodel 40 million scanned web page photos into structured information representing the entire particular person instances, which might be displayed for individuals on the net, downloaded in bulk and served machine-to-machine by APIs.

We had plenty of assist right here, each from Ravel and from the seller we relied on to deal with the processing. What stands out particularly from this section are two, associated issues: redaction and imperfection.

The Unlucky Have to Redact

Within the undertaking’s early years, the distant risk {that a} authorized writer may attempt to cease our work loomed massive. It consumed plenty of time, power and assets, and it pressured us to make compromises.

The ‘guillotine’ the undertaking used for slicing books’ bindings.

The issue was this:

• Lots of the books that comprise our official case legislation have been revealed by corporations that had a historical past of performing aggressively by litigation to forestall others from copying the legislation or from competing within the realm of authorized data.
• Whereas nobody would declare in good religion that courtroom choices authored by judges will be copyrighted by publishers, many publishers had adopted a apply of injecting into the textual content of judge-authored choices quite a lot of editorial gadgets (resembling headnotes and different annotations). In these, publishers did declare copyright.
• This intermingling of editorial content material with official statements of legislation has a contaminating impact. You can not get your palms on the official frequent legislation with out additionally touching editorial content material, which is innocent to learn however considerably radioactive to repeat and share.

To realize our targets on the undertaking, we needed to take care of this gnarly downside. The one resolution accessible to us was redaction.

Redaction means the removing or obfuscation of undesirable data. “Undesirable” is precisely how we felt in regards to the headnotes and different editorial supplies embedded inside the pages of the books we had scanned. We might have gladly labored with a “clear” model of the official legislation, nevertheless it didn’t exist. The one official model of the legislation was the contaminated one. And so we needed to prioritize, above nearly every little thing else, the correct identification and removing of those undesirable supplies from each web page and each courtroom choice that got here out of each guide that was not but within the public area. This was not simple.

The brief model of this story is: we had to determine what editorial content material to anticipate within the scanned pages; we needed to be able to alert on any sudden content material; we needed to establish the place this content material lived inside a case and on a web page; we needed to excise this materials from the textual information; and we needed to paint stable black containers over the content material on the scanned photos. We needed to do all of this with excessive precision to make sure that everybody might see the legislation and nobody might see the editorial litter.

Now let me inform you what I actually suppose. Headnotes, key numbers, annotations and the like will be helpful. Seen in their very own proper, they’re not rubbish in any respect. They’re the product of main funding and severe effort by skilled professionals. There was a time once they have been wanted to help the invention and understanding of the legislation. They do deserve safety, however solely as an unbiased enhancement layer that’s distinct from the legislation itself. After they’re mixed with the official legislation in a means that interferes with propagation and entry, they’re finest seen as air pollution. It’s an amazing failure of our judges, courts and legislatures that they’ve allowed — and proceed to permit to this present day — business entities to mingle their owned commentary with our official legislation.

Should you’re all for studying extra on this matter, I like to recommend studying the Supreme Courtroom’s 2020 choice in Georgia v. Public.Resource.Org and the numerous briefs submitted supporting entry to legislation, together with the amicus brief that we filed. Should you’re a redaction nut, please get pleasure from an instance of our work on Vol. 323 of the Federal Reporter 2d.

Getting Snug with Imperfection

As a result of we invested a lot in redaction, we needed to make sacrifices elsewhere. The 2 greatest sacrifices have been within the transcription of opinion textual content and within the scope of the undertaking. We used a know-how referred to as optical character recognition (“OCR”) to extract all of the case textual content from the scanned photos. OCR output will not be good. It sometimes requires some extent of machine and/or human correction. Whereas we corrected a few of the OCR output – textual content that recognized events and courts, for instance – we didn’t right the OCR output of the particular opinion textual content. In actual fact, the uncooked OCR high quality is extraordinarily good, and greater than ample for many functions. But it surely’s not good, and our legislation deserves perfection.

We additionally couldn’t preserve digitizing the legislation eternally. We needed to restrict the scope of the undertaking, and we would have liked to show our consideration to the work of constructing the info accessible on-line. And so we needed to finish scanning in early 2017, though finally we have been in a position to prolong it into 2018.

I’ve heard individuals query these compromises, as in the event that they made the undertaking pointless. That’s bunk. We calculated that if we made certain to create and share top quality scanned photos and metadata for the total historic report — the work that may be hardest to breed — know-how would proceed to enhance and others (ideally the courts) would step as much as contribute going ahead. Certainly, that is what’s taking place. OCR know-how is way improved, and it’s not too laborious to redo the OCR to get higher outcomes. With all the photographs and metadata now freely, publicly accessible for anybody to entry, we will all go to work making the textual content constancy even higher.

As for the undertaking’s scope, Ed Walters and the great people at Fastcase (now vLex) generously agreed to share their transcriptions of some newer courtroom choices. On the identical time, the non-profit Free Law Project, led by Mike Lissner, continues to set the usual and do a much better job than the federal government itself in offering widespread public entry to newly issued courtroom choices and case dockets. The courts haven’t accomplished their half but, however I’m nonetheless hopeful.

So the info isn’t good. It’s a little bit bit stale. However these gaps are closing, and sometime they’ll be gone.

Entry, Exploration and Experimentation

All the pieces I’ve shared thus far was a precursor to the final word finish objective: free public entry on-line. Satirically, after we began, we had no concept what public entry would appear to be or if our workforce within the library would ship it. That is why we made certain the contract required Ravel to ship public entry.

An Awkward Dance

Then in June 2017,  LexisNexis announced that it had bought Ravel. Their public statements expressed an intention to proceed supporting the undertaking and to comply with by on Ravel’s commitments. Privately they mentioned the identical factor. That they had little alternative; they inherited the contract, and it was hermetic. Both comply with by and achieve the advantage of the remaining business exclusivity and a pleasant relationship with Harvard, or renege and see all the info — which by this time was practically full — go free instantly.

However phrases are simple. In sensible actuality, we have been caught in an ungainly dance by which Lexis did the minimal required beneath the inherited settlement, and provided that we held them to it. Their follow-through on public entry was perfunctory at finest. I might’ve been joyful to see Lexis lean into the chance and grow to be a daring standard-bearer for true public entry to legislation. I additionally would’ve been joyful to see Lexis wholly abandon the commitments Ravel had made. However now that the laborious digitization work was principally accomplished, I had little curiosity in frantically waving across the contract and chasing Lexis to do one thing it had no intrinsic motivation to do. I additionally knew it might be tough and irritating to get Harvard to throw any actual institutional weight behind persuading Lexis to do way more.

So as an alternative of focusing our power on pushing Lexis, we began working earnestly inside the Library Innovation Lab to make the most of the rights the contract gave us to supply public entry instantly ourselves.

Delivering on Public Entry

This was my favourite a part of the undertaking. That is what our Library Innovation Lab beloved to do and did finest: design and code high-performing, open supply software program that may fulfill the elemental library mission of enabling entry to information.

We had nearly free rein to construct something we wished that may make it simpler for individuals to learn and examine the legislation. The largest query we confronted was whether or not to attempt to construct a free authorized analysis device which may substitute for costly business merchandise. We determined to not. As a substitute, we centered on offering direct entry to the info. We wished to allow others to construct instruments and merchandise, and we wished to discover new methods of interacting with the info. We did construct a easy search and viewer interface for individuals who simply wished to learn a number of instances, however we selected primarily to prioritize issues that business distributors would by no means do.

The scanner belts.

It’s laborious in a put up like this to explain the know-how we constructed, so as an alternative I’ll invite you to make use of the Caselaw Access Project and, in case you’re so inclined, to repeat and remix the project’s code. Once you go to CAP at present, you’ll see that the legacy web site and instruments are nonetheless accessible at https://old.case.law, however they’re set to sundown in September 2024 now that there aren’t any restrictions on the info and everybody can do what solely the Lab might do earlier than. Try Trends, an incredible interface constructed by the Lab’s present director, Jack Cushman, to permit individuals to discover how authorized language and concepts developed. One other favourite of mine is Colors, constructed by Anastasia Aizman in 2019 as an early, whimsical exploration of the info utilizing pure language processing and neural networks.

The Virtues of Good Plumbing

These explorations mattered, however our greatest technical achievements weren’t the vivid demo purposes we constructed ourselves. The true contribution was setting up the strong “plumbing” by which we might ship the info to others.

The plumbing we constructed had two primary components: an API and a bulk information service. The technical particulars are wonderful, and if technical particulars are your factor, cease studying and go take a look at the code. Attain out to Jack Cushman and the Lab’s present workforce to study extra about what we did. Discover methods to contribute to the wonderful work the Lab is doing now within the areas of authorized AI and internet archiving.

Broadly talking, we designed the API for individuals writing pc packages that would want on-demand entry to details about explicit U.S. courtroom choices, or who wished to maximise what they might do with their each day allotment of full-text instances. We designed the majority information service for verified non-commercial researchers who wished to work with massive volumes of courtroom choices to achieve some new perception or to analyze huge concepts throughout the dataset.

One key emphasis was to slice and cube and repackage the info in as some ways as we might, to help the widest doable vary of customers and makes use of. In consequence, now you may get PDFs of the scanned photos, both as particular person instances or complete volumes. You may get instances as JSON or XML, with the textual content of opinions as plaintext, HTML or XML. You may get complete instances, or simply the metadata. You may get smaller datasets reflecting any of the time durations, jurisdictions, courts and titles within the assortment. You may get specialised datasets that replicate all of the citation-based connections (the “quotation graph”) amongst instances. You can also create your individual specialised datasets based mostly on any search time period and quite a lot of complicated filters. If you wish to shortly curate and download a dataset of all decisions issued between 1960 and 1990 by courts in Iowa, which mention “farm” and cite to the Indiana Supreme Court, go for it. If you wish to put the complete assortment of revealed U.S. courtroom choices on a thumb-drive, have at it.

Influence

Whereas immersed in fixing the sensible challenges of the contract, the scanning, the processing and the supply, we didn’t suppose a lot in regards to the affect the undertaking may need as soon as we made the info accessible. We took it on religion that somebody would discover it helpful.

We launched each the API and the majority information service publicly in late 2018 and bought a wave of favorable publicity. The one little bit of recognition that stands out for me was an editorial within the The Harvard Crimson titled “In Favor of the Caselaw Access Project.” For some cause, there’s one thing particular a few pupil publication expressing gratitude for our work.

Publicity will not be the identical as affect. What actually mattered was whether or not individuals used the info. For some time it was laborious to know what individuals have been doing, however now we will begin to see the proof. Should you take a look at references to the undertaking on Google Scholar or SSRN, you’ll see tons of of articles throughout a dizzying array of subjects like antitrust legislation, linguistics, judicial partisanship, tax legislation, organ transplant litigation, machine studying and LLMs, authorized pedagogy, and the long-term frequent legislation affect of instances involving enslaved individuals, simply to call a number of. Should you search on the net, you’ll see over 50 library guides that spotlight the undertaking as a supply for authorized analysis or scholarly information and tons of of hundreds of hyperlinks into the undertaking’s web site. Should you take a look at Reddit, you’ll see an limitless scroll of posts mentioning the undertaking in all types of helpful (and a few wild) contexts. Should you take a look at Github or HuggingFace, you’ll see a rising variety of technical initiatives utilizing and crediting the undertaking. Should you speak to a number of authorized tech startups, like I do, you’ll hear how a lot simpler it’s to start out one thing new due to the undertaking.

That is solely what’s public and straightforward to search out in a couple of minutes on-line, or what individuals I occur to speak to are keen to share. That is simply within the comparatively brief time since we launched the info, and all of it got here throughout a interval by which we needed to artificially restrict and situation entry. Now the floodgates can open.

What Comes Subsequent?

Now I shift from private recollection and remark to hypothesis. What comes subsequent because of the Caselaw Entry Challenge? I don’t actually know.

I consider the undertaking will proceed to allow scholarly analysis that helps us higher see the dangerous patterns, prejudices and previous failures of our authorized system, in order that we will work collectively towards one thing a lot better. In legislation college, I discovered quite a bit about civil process and business transactions however completely nothing about how our courts dealt with slavery earlier than the Civil Battle. For a very long time, I used to be ignorant in regards to the lively efforts of our career in perpetuating this sin. However by the Caselaw Entry Challenge’s data and tools, I discovered that defending slavery was certainly one of our courts’ most outstanding early priorities. By important work by others using the data, I’ve discovered that this shameful legacy continues to affect our legislation at present.

I even have a robust hunch that generative AI will remodel the authorized business, and that the undertaking’s information will play a significant position. My hope is that the undertaking will make it simple for good, inventive individuals to discover new AI-enabled concepts that may have been inconceivable if the legislation remained locked away in books and proprietary databases. I’m joyful that it is going to be inside attain for anybody with technical abilities to construct their very own model of an AI authorized assistant, quite than it being reserved solely to corporations with particular entry to the legislation. I additionally suspect the undertaking’s information will probably be a part of the answer to quotation hallucination, and I hope courts will quickly notice that the basis causes of this downside are unhealthy lawyering and inaccessible legislation, not know-how.

There are constructive variations of the longer term by which the undertaking contributes to instruments and companies that assist decrease the entry to justice barrier, enhance the standard and worth of authorized work and permit individuals to raised perceive their rights and obligations. These are the longer term eventualities I’m dedicated to and can proceed working towards enthusiastically.

However there are additionally variations of the longer term by which technical specialists, with no consciousness of or regard for the character of the legislation, may use the undertaking’s information to inadvertently do harmful and dangerous issues. Right here I’ll share a phrase of warning in regards to the frequent legislation and the info we’ve helped make accessible: it’s at all times difficult, usually ugly and regularly simply plain evil. These instances are filled with horrific particulars of violence and struggling. The individuals talked about within the instances are actual. A lot of them, or their households and buddies, are nonetheless residing. And eventually, over the course of the 350-plus years represented within the dataset, the legislation has usually been horribly, disgustingly flawed. Don’t make the error of believing the legislation of Alabama or Massachusetts from 175 years in the past is match to tell a modern-day free authorized recommendation chatbot. Don’t assume judges are at all times neutral or by no means prejudiced. Don’t presume all legislation is sweet legislation.

These will not be causes to maintain the legislation closed or to proceed giving privileged entry to a couple massive corporations, however they’re compelling causes for all of us to be considerate about how we use and share the info. Maybe they’re causes, going ahead, for judges to suppose in a different way about how they write opinions and what particulars have to be made specific for a choice to hold its weight.

***

All in all, I’m extremely lucky that I might contribute to this undertaking and work intently with so many wonderful individuals to see it by from concept to affect. I’ve been fortunate to work on plenty of nice initiatives, however this one stands alone in each means. So price it.