What we believe, and what we are trying to solve.

The Institute's agenda is not a list of topics. It is a set of claims, each derived from work already published in our library, each stated precisely enough to be tested, and each assigned to a standing research program that is responsible for testing it.

The claims share a starting point. The first generation of legal AI was judged almost entirely on whether it fabricated citations. That question matters, and grounded systems can now answer it. What they cannot yet answer, and what decides whether a brief can be filed, is whether a real, accurately quoted, good-law authority is the right one: controlling in this forum, pitched at this stage, cutting in this direction, and still standing. The agenda is about that harder question, and about what changes for judges and lawyers once machines can answer part of it.

Each claim below links to the program that pursues it and to one numbered thesis that states a position within it. The Manifesto explains why we hold these commitments; the full list of Theses states them one at a time, with defenses, supporting work and revision histories; the Open Problems record what we have not solved. Programs, theses and frameworks are versioned, and when the evidence moves, the agenda moves with it.

The six claims

1. Legal AI must surface uncertainty, not smooth it over. Program: Determinacy and Gray Areas. Thesis: Hallucinated certainty is more dangerous than hallucinated citations.
2. "Good law" is not enough; a proposition has to be usable in a forum, a posture and for a client. Program: Verification and Usable Law. Thesis: Is this proposition usable here, now, for this client?
3. Every legal output must be explainable, grounded in validated authority, testable and versioned. Program: Legal AI System Design. Thesis: The Four Laws as a design floor.
4. Law behaves like a versioned codebase and can be governed like one. Program: Legal Knowledge Engineering. Thesis: Law is a versioned codebase.
5. Adversarial reasoning, building the counter-model, is a first-class system function. Program: Adversarial Reasoning. Thesis: Build the counter-model, not a rebuttal.
6. Automation changes what judges and lawyers are for, and that question deserves rigorous treatment. Program: Machine Jurisprudence. Thesis: Verification and discretion are different parts of judging.