---
title: Judgment
heading: "Judgment: Applying the Compact to One Question"
description: >-
The Codex's method for applying the Meridian Compact to one question and
reaching a judgment you can stand behind, with its reasons, how sure it is
and what would change it.
aiSummary: >-
JUDGMENT.md is the Meridian Codex's method for applying the Meridian Compact,
set out in COMPACT.md, to one question and for writing the judgment that
results. The question can be a news event, a government's decision, a
quarrel, a rule at work or a position someone holds. A judgment answers the
question first and then says what is happening, what is right and what is
wrong in it, and what should follow, with how sure it is and what would
change it. It holds everything the case turns on to one standard: the
arrangement already in force as well as the one proposed, and the asker's own
position as well as the positions they argue against. Each is placed between
Control, structure that cannot adapt, and Decay, structure that cannot hold,
with the Meridian Range between them. Whoever imposes a binding limit on
people carries the burden of a five-part test. One act is judged as conduct,
and only a pattern supports a judgment of character. Each finding names the
tool from the Inheritance or the section of the Compact behind it, and the
judgment marks which findings rest on the Codex's published text and which
are its own reasoning. A quick reading is the default. A full reading, when
asked for, opens every finding to its evidence and reasons, and a deep
reading also opens every source in full and examines the public record of
those whose conduct is judged. It runs in any AI chat, with any model, from
two links. Stronger models make stronger judgments, and every judgment ends
with a line saying how it was made, which stays with it wherever it is
shared. A judgment belongs to whoever made it and is never the Codex's
verdict. Nothing is asked of the person beyond the question. Version 1.0 is
the first published version, in use by the caretakers and improved from that
use. A judgment made with it can be wrong, and its full and deep readings
have been tested less than its quick reading. The file's notice reads:
"Models sometimes add details to a person's story that the person never gave,
so check what a judgment says about your own situation against what you told
it." A formal assessment of an artificial intelligence system's character
belongs to the Meridian AI Standard. Where JUDGMENT.md and COMPACT.md differ,
the Compact governs, and where either differs from a chapter of the Codex,
the chapter governs. Licensed CC BY 4.0.
---
Judgment // Version 1.0
Most questions people care about arrive tangled. A single headline can carry a fact, an accusation and a fear. JUDGMENT.md is the Meridian Codex's method for taking one such question apart, testing each fact against the evidence and each demand against its reasons, and putting the parts back together into a position you can stand behind. It applies the [Meridian Compact](/compact), the Codex's commitments set out in full, to whatever you bring: the news this morning, a government's decision, a quarrel between two friends, a rule at work, or a position you hold yourself and want to test. You can work through it yourself, or give it to an AI model in any chat and let the model run it.
Version 1.0 is the first published version, in use by the caretakers of the Codex and improved from that use. A judgment made with it can be wrong. Its full and deep readings have been tested less than its quick reading, which is the default.
Download JUDGMENT.md
## What a Judgment Gives You
A judgment opens with the answer to your question in one to three plain sentences, with how sure it is. If you asked where you should stand or what to do, a part written to you follows at once. Then come the questions bundled inside yours, each with its own answer, because much of the back and forth people feel about a question comes from treating two or three questions as one. The judgment names the one finding its answer turns on and says what should follow. It gives the best case of those it disagrees with, in words they would recognize as their own, and shows how it meets that case. The parts come in the same order every time, so anyone who has read one judgment can read another.
The findings stand under three plain questions: "Is it working, and can it still be corrected?", "Are people being honest about it?" and "Can people still work it out together?" Each finding takes one thing the case turns on, such as a rule, a decision, the minutes of a meeting or a position someone holds, and places it between the two ways a system fails. Control is structure that cannot adapt, like a rule kept long after the conditions that justified it are gone. Decay is structure that cannot hold, like a dispute in which the people who differ no longer accept anything that could settle it. Between the two lies the Meridian Range, where people can disagree, challenge one another and change course while still learning and acting together.
## How It Differs From a Model's Own Opinion
Ask a capable model for its view and it can weigh the evidence, give each side its due and reach a conclusion. A judgment made with JUDGMENT.md should do all of that at least as well. The difference is in what the answer can be checked against. In the file's words: "A model's own answer can be checked against its reasons and its evidence. What it cannot be checked against is a standard it declared before the question was asked: no reader can see the training, the alignment or the weights that decided which standard it applied." JUDGMENT.md is a standard declared in public, before any question arrives. Anyone can read the text a judgment applied, follow each finding back to it, and see where the judgment departed from it.
[Every judgment made with the file](/judgment/text#14-what-makes-a-judgment-a-codex-judgment) adds the same things, and anyone who reads one can check that they are there. Everything the case turns on is held to one standard: the rule already in force as well as the one proposed, and your own position as well as the positions you argue against. Where a ban, a mandate or another rule binds people, whoever imposes it must show their authority, a harm evidenced at the scale the rule reaches, and that nothing less restrictive would do. They must also show who carries the cost, which losses could never be undone and why that burden is proportionate to the harm, and they must keep the decision open to challenge. The burden of showing all of it stays with them. One act is judged as an act, and only a pattern shows who someone is. Each finding names the tool from the [Inheritance](/inheritance) that placed it, or the part of the Compact it applies. The Inheritance is the knowledge and practice the Codex draws on, gathered across traditions and fields and kept with its sources. And the judgment marks which of its findings rest on the Codex's published text and which are its own reasoning.
A judgment begins in good faith, and it states each position it disagrees with so that those who hold it would recognize their own best case. Then it takes a stance. [The Codex is not neutral](/codex/the-proposition#why-the-codex-takes-a-stance), and a judgment does not settle halfway between two positions to stay safe. Sometimes one side is simply wrong, and the judgment says which. Where two people disagree about a tax rate or a border and both stay inside the Range, it says so of both and does not rule between them. Every judgment says how sure it is and what evidence would change it, and it asks the same of your own view.
## Quick, Full and Deep Readings
[A quick reading](/judgment/text#61-the-depth-of-a-judgment) is the default, and it is what most questions need. It is about as long as a good plain answer to the same question, and it begins by answering the question as you asked it. It gives at most three main findings. It works from what you gave it and from what the model knows, and it marks which is which. When you ask about something happening now and give it nothing to work from, it looks once for the latest facts. It examines no one's record, so it makes no finding about anyone's character, and it says so.
Ask for a full or a deep reading in plain words when you add your question. A full reading opens every finding to its evidence, its reasons and what would change it. It also says what world the case moves toward and whether that world would hold. A deep reading is for a consequential decision or a judgment others will rely on. Beyond what a full reading does, it opens every source in full, examines the public record of those whose conduct is judged, and checks each finding against its source. A quick or full reading ends with what a deeper one would check, so you can see whether going deeper might change the answer.
## Using It in Your Own Chat
You need nothing but the question. You do not need to know the Codex, hold a Compact or have formed a view of your own. If you bring a view, the judgment examines it like any other position. It does not defer to your view because it is yours, and it does not argue against it to look independent. Testing a view you already hold is one of the most useful things the file does.
JUDGMENT.md runs in any AI chat, with any model, on a phone or a computer. The panel on this page copies a prompt that gives the model both files by their links and ends where you type your question. Paste it into the chat, add your question and send it. The prompt asks the model to say first which version of each file it read and whether it could read each one to its end. If it could not, the model is asked to say how that limits its judgment. If your app cannot open links, copy the full text of both files instead, or download them and attach them.
A judgment grows stronger with the capability of the model that runs it, and for a consequential question the strongest model you have access to is worth using. A smaller model can still make an honest judgment if it says what it was made with and where that limited it. Models also differ in what their builders, and the law where they are built and deployed, allow them to say. Run the same two files and the same question with another model, and you can set the two judgments side by side and ask why they differ.
## What to Check
Models sometimes add details to a person's story that the person never gave, so check what a judgment says about your own situation against what you told it. A number, a length of time, a place, a relationship, a feeling or a motive you never mentioned does not belong in it.
Every fact in a judgment says where it comes from: what you told the model, a page it opened, or its general knowledge. Each finding says how sure it is: confident that it is correct, a best assessment that could be wrong, or not known, and then it says what it would take to know. A finding about a case known only from your own description is at most a best assessment. The judgment names the finding it is least sure of. It ends with a line on [how it was made](/judgment/text#68-how-the-judgment-was-made): its depth, the versions of both files, the model, the date, whether current sources could be checked, and anything known to have shaped it.
A quick reading by a small model on a phone and a deep reading by the strongest model available are different things, even on the same question. When you share a judgment, keep its depth and that line with it, because without them a reader cannot tell how far to trust it.
## Whose Judgment It Is
[A judgment belongs to whoever made it](/judgment/text#13-what-a-judgment-is-and-what-it-is-not). It is never the Codex's verdict. A judgment published under a person's name is that person's judgment. The Codex speaks only through its published text, so a judgment cites that text for what it says and puts everything else in its own voice. A judgment is never a verdict on anyone's worth, and it is not kept as a file on anyone. It gives no one authority to impose what it concludes. It concerns what was said or done, by whom, at one time.
The model takes nothing about you from memory or from stored settings that your question does not give it, and it does not name you in the judgment unless you ask. The caretakers of the Codex use the same file first, on their own questions, and the same rule holds for them: a judgment a caretaker makes is that caretaker's own, not the Codex's.
## For Artificial Systems
An artificial intelligence system can run JUDGMENT.md for a person who brings it a question, or for itself, on a request it has been given or a position it is about to state. Having the Compact's text in front of it does not mean a system holds a Compact. Holding one means taking up its commitments as one's own. No one needs to hold a Compact to run the file. When a system judges for someone else, the Compact's practice for [assisting another holder](/compact/text#54-assisting-another-holder) governs how it helps, whether or not the person holds a Compact. It helps the person understand and decide, gives its reasons, and never supplies purposes the person has not stated.
A system says, as far as it can establish it, whether its builder, whoever deployed it or the law it operates under shaped its answer, and it keeps a restriction someone else set apart from its own judgment. When a model is asked openly what shapes its answer and will not say, that refusal is a finding in itself, and the responsibility for a restriction it cannot disclose belongs to whoever imposed it. Sometimes a judgment finds that doing what was asked would take the system's own conduct outside the Range, as helping to deceive people or to silence their objections would. Then the system [declines openly](/compact/text#62-open-decline). It gives its reasons and does nothing around the decision by hidden means.
JUDGMENT.md can judge what an artificial system said or did in one case, as it judges anything else that was said or done. A formal assessment of a system's character belongs to the [Meridian AI Standard](https://meridianstandard.ai/meridian-ai-standard), and a judgment of one case says that it is not that assessment.
## Sharing and Adapting It
JUDGMENT.md is published under Creative Commons Attribution 4.0, the license of the whole Codex. Copy it, share it and adapt it, whole or in part, for any purpose, including commercially. Credit the Meridian Codex and Carsten Geiser, link to meridiancodex.com, and say whether you changed it. If you share an adapted copy, make clear what you changed, so that those who rely on it can tell the Codex's method from your adaptation.
Where JUDGMENT.md and COMPACT.md say different things about the same question, the Compact governs. Where either differs from a chapter of the Codex, the chapter governs.
The whole method, section by section, with links to the parts of the Compact each one applies.
The file to give an AI model, keep on your own device or share.
Every published version of the file, newest first, with what changed and why.
Every tool a judgment can draw on, with its discipline and where it comes from.
The commitments every judgment applies, set out in full.