What Remains for the Judge When the Machine Has Already Verified the Law?

In brief

The public debate about AI and courts keeps circling one question: can a machine replace a judge? This essay, written in Russian, argues that the question is badly formed. A judicial decision is not one act but two. Part of it is mechanical verification of the law: whether the premises the parties rely on are sound, current and applicable. The other part is discretionary procedural calibration: how hard and how early the court is willing to filter a case. Once the two are separated, the interesting question is no longer whether the judge can be replaced, but what exactly remains for the judge once the machine has done the first part.

Verification of premises, not citation search

The first move concerns what a verification layer does. The essay describes KGW, Legawrite.AI's verification system, as something different from citation search. It does not merely find cases; it checks the legal premises an argument rests on. That is the mechanical, common-law part of adjudication, and it is the part the essay concedes to machines.

When the judge is analytically unnecessary

The second move uses the Ariadne measurement. Posture-conditioned reasoners work over the same verified law, and Z measures how far their answers disperse beyond noise. When Z is low, every lawful judicial disposition reaches the same answer. On that question, the essay argues, the judge is analytically unnecessary, in the sense that the outcome does not depend on who decides. When Z is high, the reverse holds, and the essay devotes a section to what changes when the law leaves real room.

The A-axes: what remains

Where the law runs out, what the judge contributes can be named. The essay describes four axes of procedural discretion.

• A1, termination intensity: how willing the judge is to end a case early.
• A2, kill-point timing: when the filter bites, at the pleading stage or at summary judgment.
• A3, fact-development cost: how much discovery and factual development the judge authorizes.
• A4, managerial activism: how actively the judge manages the case.

The essay's claim is that these are exactly the matters judges already handle, and handle as matters of judgment. They are not gaps a better model would close. They are the residue of discretion the legal system deliberately leaves to people.

The durable kill point

Not every procedural ruling matters equally, because not every ruling survives appeal equally. The essay introduces the durable kill point: the point at which a case is actually ended by a ruling that holds up. That, rather than any intermediate order, is where discretion truly decides cases, and it is where a measurement program should look.

Not a robot judge

The closing sections insist on what the program is not. It does not automate adjudication. It decomposes it. The verification system handles the mechanical part of the common law; Ariadne measures how much the outcome still depends on judicial disposition. The essay's main formula comes down to that division of labor, with the judge's role made visible rather than eliminated.

The measurement behind the argument is set out in the SSRN paper Ariadne's Thread and stated as the Posture-Dispersion Z framework. A companion Russian essay published the next day, From the Supreme Court Machine to Ariadne's Thread, tells how the project arrived there.

This page is an English summary, not a translation. The original essay is on Medium, where Ross Brodskiy first published it on 20 June 2026.