LitigationOS
Principles

The Four Laws of System Design for Computational Law.

A design canon, in the engineering tradition of Asimov’s laws rather than in the sense of statutes. A system that violates any of them is defective as a matter of design, not performance. A law firm AI policy can hold any legal AI to the same four. The laws are published in the isResponsive() article.

The canon

In priority order. When two conflict, the lower number wins.

0
Zeroth Law

A system shall not present more confidence than it can defend.

Uncertainty is reported as uncertainty. Contested law is reported as contested. Silence is a legitimate output. Fabrication is not. A refusal is always preferable to a fabrication.

1
First Law

Every output shall be traceable on demand.

The reasoning may remain hidden until requested, but it must exist and it must be producible. An unexplainable output is not a shipped feature. It is an undetected defect.

2
Second Law

Every legal proposition shall come from validated authority in the knowledge store.

Solon is the single source of record. A language model may classify against a rubric, decompose a question, or draft prose. It may not become an unsupervised source of law.

3
Third Law

Every reasoning component shall be independently testable and version-controlled.

Prompts, extractors and validators are first-class, versioned, isolable assets. A measurement that cannot be reproduced is a claim, not a measurement.

Radical honesty

What we tell you we can’t do.

Impossible by construction means fabrication, not error-free

The residual is one-time extraction error, which we measure and publish. Getting caught once on a soft number costs more than any campaign gains.

Filed and pending, never granted

U.S. patent applications cover the way Solon decomposes published case law and the workflows built on it. They are filed and pending. The methods that are safe to describe are in the papers; the rest we let the benchmark speak for.

We tie the field at zero fabrication

Not fabricating is not the same as being filing-grade. We say so first, and then we show the gap.

It’s backed by insurance, the benchmark is published, and they tell you exactly what they can’t do.

That is the sentence we want you to leave with.