About the Computational Law Institute

The Computational Law Institute is a research publisher on law and machines, founded in 2026. It studies what machine-assisted legal work has to do before a lawyer can file it with a court, and it publishes that research in the open: working papers, essays, field guides, versioned frameworks, numbered theses, a lexicon of defined terms, a list of open problems and a public record of AI citation failures in court.

Its subject is applied computational law: legal reasoning by machines over the real body of case law, under the conditions of actual litigation, judged by whether a court would accept the result.

Mission

The Institute's charter, in full:

The Computational Law Institute exists to turn fifty years of legal reasoning theory into standards the profession can test, rerun, and rely on. Its mission is to define what "filing-grade" means for machine-assisted legal work and to measure it in the open: not merely whether an AI system fabricates citations, but whether it finds the controlling authority, surfaces the adverse precedent, confirms that a holding is still good law, and shows the exact reasoning path from precedent to conclusion. The Institute pursues this through four commitments. It publishes open, rerunnable benchmarks with public rubrics that grade completeness and omission alongside accuracy. It develops shared formal vocabularies for holdings, treatment, and the recognized modes of legal inference, so that reasoning can be audited rather than trusted. It advances the principle that reliability in legal AI must be designed into a system's architecture, not inspected in after the fact. And it maintains a public record of AI citation failures in court, so that bar associations, judges, and malpractice carriers can set policy on evidence rather than anecdote. The Institute builds on open law, publishes its methods, and invites replication by any vendor, court, or scholar, because in a profession that answers to judges, a standard only matters if anyone can check it.

Research and publications

The Institute's research is organized into six research programs, each framed as a standing question:

• Determinacy and Gray Areas. How do we measure where the law is genuinely unsettled?
• Verification and Usable Law. What does it take for a legal proposition to be safe to rely on?
• Adversarial Reasoning. How should machines model the other side's best case and allocate burdens?
• Legal Knowledge Engineering. How do we represent law so that it can be versioned, queried and audited?
• Legal AI System Design. What are the non-negotiable design rules for legal AI?
• Machine Jurisprudence. What remains for judges and lawyers once machines verify the law?

The work is published in these forms:

• Papers. Working papers posted to SSRN, each with a companion page on this site.
• Frameworks. Named models, laws and protocols, each versioned with a changelog.
• Theses. Numbered positions, each citable on its own.
• The Library. Essays, field guides, thought experiments, case studies and reviews.
• Lexicon. The terms the Institute uses or coined, each with a definition.
• Open Problems. Unsolved research questions, with the Institute's partial answers.
• The Citation Failure Record. Court decisions on AI citation failures in filings, classified by failure type.

Leadership

The Computational Law Institute is led by its director, Ross Brodskiy, who is an author of each of its working papers. Nathan Pokov is a co-author of two of them. Many of the Library's essays, reviews and variations are written by Institute staff. Everyone who writes here is listed under People.

Where the Institute came from

Building a system that lawyers could file from kept raising questions that were not engineering questions at all. When is the law genuinely unsettled, and how would anyone know? What makes a real, accurately quoted case unusable in a particular motion? Which inferences from precedent would a court accept, and which only sound as if it would? Who carries the burden, and what happens to a model when the Supreme Court moves it?

The answers accumulated as working papers on SSRN, essays on Medium and a larger body of working drafts. The Institute publishes that program openly and in one place, with stable addresses, version histories and citations, so that the ideas can be read, tested and argued with on their merits. Legawrite.AI remains the founding sponsor and the applied testbed. What that relationship means, and how it is disclosed, is set out in Independence and Disclosure.

What "applied computational law" means here

Computational law is not new, and the Institute does not pretend otherwise. In 1957 Layman Allen proposed normalizing legal text into logical form, so that one could retrieve the structure of a rule rather than every document that mentioned it. In 1977 L. Thorne McCarty's TAXMAN reasoned over the conceptual structure of tax law. Kevin Ashley's HYPO (1990) modeled the way lawyers cite a case, distinguish it and rebut the distinction, and the CATO work organized the facts of cases into side-relative factor hierarchies. Henry Prakken and Giovanni Sartor formalized defeasible argument, Trevor Bench-Capon developed argument schemes with the critical questions that attack them, Branting treated the ratio decidendi as a structured justification, and the Carneades model took on burdens of proof in legal dialogue.

That tradition was rigorous, and it was largely right about what legal reasoning requires. Its limits were practical. Systems reached production where the law was rule-like, as in statutory domains, and stayed in the laboratory where it was not. Case-based systems were deep but narrow, each resting on a hand-built model of a single domain. Argumentation frameworks worked beautifully once their input had been structured, and assumed that someone else would do the structuring. The hardest step, extracting holdings, treatment and inferences from unstructured opinions reliably and at scale, was the one left undone. The longer history is told in a Library essay on the road to applied computational law.

"Applied computational law", as the Institute uses the term, starts at that missing step.

The last row matters most. When the Institute's director ran a blind eDiscovery review with no humans in the loop, the write-up reported the gates the run failed alongside the ones it passed, and the engine was released under an open-source license so that others could rerun it. That is the standard the Institute sets for its own work and for everyone else's.

The four commitments

1. Open, rerunnable benchmarks. Public rubrics that grade completeness and omission alongside accuracy, because a motion is lost on the controlling case nobody pulled as often as on a bad citation. See what a filing-grade benchmark must measure and The Docket Test.
2. Shared formal vocabularies. Named, versioned definitions for holdings, treatment and the recognized modes of legal inference, so that reasoning can be audited rather than trusted. See the Lexicon, the Case Treatment Taxonomy and the Twelve Bridges.
3. Reliability by design. Guarantees that live in a system's architecture rather than in a checker bolted on afterwards. See the Four Laws and The Verification Layer Is a Smoke Alarm, Not a Building Code.
4. A public record of AI citation failures in court. An evidence base for the bar associations, judges and malpractice carriers who have to set policy, maintained so that policy can rest on evidence rather than anecdote.

What the Institute does not do

The Institute is a research publisher and nothing else. It runs no events and takes no speaking engagements. It offers no consulting and no other services. It has no membership, no summits and no sales funnel.

There are three things a visitor can do here:

• Read. Everything the Institute publishes is in the Library, organized by research program.
• Cite. Every piece has a stable address and a citation block. See How to Cite.
• Subscribe. New research by email or by feed, with no marketing. See Subscribe.