Computational Law Institute
Modern Applied Computational Law Lab
Law is rarely settled. Legal machines should say so.
We turn fifty years of legal reasoning theory into standards the profession can test, rerun and rely on, and we define what filing-grade means for machine-assisted legal work, in the open.
Filing-grade, defined and measured in the open.
Computational law needs its own discipline because legal reliability is not a property of fluent text. It is a property of authority, posture, burden and omission: whether a system 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 publishes openly because, in a profession that answers to judges, a standard only matters if anyone can check it. We publish rerunnable benchmarks with public rubrics, shared vocabularies for holdings, treatment and the modes of legal inference, and a public record of AI citation failures in court.
Six claims we are prepared to defend.
Legal AI must surface uncertainty, not smooth it over.
Determinacy and Gray Areas02 · Thesis 6Good law is not enough. A proposition has to be usable in a forum, a posture and for a client.
Verification and Usable Law03 · Thesis 21Every legal output must be explainable, grounded in validated authority, testable and versioned.
Legal AI System Design04 · Thesis 16Law behaves like a versioned codebase and can be governed like one.
Legal Knowledge Engineering05 · Thesis 11Adversarial reasoning, building the counter-model, is a first-class system function.
Adversarial Reasoning06 · Thesis 27Automation changes what judges and lawyers are for, and that question deserves rigorous treatment.
Machine JurisprudenceFrom the Library
Right Law, Wrong Stage
Procedural Posture as the Failure No Citator Flags
A case can be real, accurately quoted, good law and on point, and still be unusable in the motion that cites it, because every holding is announced under a standard, on a record, with a burden, in a forum. This essay sets out four kinds of posture mismatch (standard, record, burden and forum), a test for which propositions carry across procedural stages, and a four-question pre-filing check.
By Eleanor Voss
Six standing questions.
Each program is a line of inquiry, not a committee. It keeps its own theses, frameworks, reading list and open problems.
Named models, versioned like code.
The Four Laws of System Design for Computational Law
A legal AI system must never overstate its confidence, must explain every output, must ground every legal proposition in validated authority, and must keep its reasoning components testable and versioned, with each law yielding to the laws above it.
Read the Four LawsObserve, theorize, formalize, test.
Every framework on this site went around this loop at least once, and carries a version number and a changelog because it will go around again.
Observe
A failure in legal practice or legal AI: the wrong-stage citation, the missing controlling case, the confident answer that cannot be defended.
Theorize
Turn the failure into a claim precise enough to be wrong. It becomes a numbered thesis.
Formalize
Express the claim as a framework, protocol or metric with named parts and a version number.
Test
Run it in code, data or experiment. Publish the method and the failures, then version the result.
Worth rereading
What a Best-in-Class Opposition System Must Actually Do
A motion is a draft of the court's order, so an opposition that only objects leaves the judge holding one coherent account and a list of complaints. Ross Brodskiy argues that a best-in-class opposition system builds a...
The Promise Fulfilled
For thirty-five years, the computational jurisprudence tradition has specified what genuine legal reasoning requires: adversarial directionality (Ashley, 1990), typed rule systems (Prakken and Sartor, 1996), and...
Three reading paths.
For lawyers
Start with the errors that pass every citation check and still lose motions. Finish with protocols you can run, with or without AI, before your next filing.
Start reading8 piecesFor legal technologists
Start with the design rules and the argument against treating verification as architecture. Finish with systems that were built, run and measured in the open, failures included.
Start reading8 piecesFor academics
Start with the history of computational law and the gap between what the field specified and what shipped. Continue to the formal work on openness, inference and interpretation, where replication and criticism are most needed.
Start reading