Editorial Standards

The Institute argues that legal AI should show its sources, admit its uncertainty and keep a history of its changes. These standards apply the same rules to the Institute's own writing.

Review

Every piece is reviewed by the founder and director before publication. Review checks, at a minimum, that:

• every authority is traced to its source and cited for a proposition it actually supports;
• every number is traced to its source, and a figure that changes over time carries an as-of date;
• no pinpoint citation is given unless it has been confirmed against the source;
• declared constructs of the Institute are labeled as constructs and never presented as settled law;
• load-bearing empirical claims in the Institute's own research are stated as conjectures, with the result that would refute them;
• pieces that re-tell a working paper (variations, persona pieces) stay faithful to every factual claim in the source;
• a disclosure line is present wherever a piece discusses Legawrite.AI products (see Independence and Disclosure).

Provenance tiers

The Institute's research papers label their sources by provenance tier, following the convention used in the Ariadne's Thread papers.

Two rules follow. First, nothing in Tier 3 is ever presented in the language of Tier 1: a construct can be useful, tested and well defended, but it is not law. Second, where a Tier 3 source is doing real work in an argument, the piece says so, and says what would change if the source did not hold up. The same discipline appears inside the research itself: the Ariadne method flags any point where a discretionary setting is dressed as a command of the law.

Corrections

• Substantive errors are corrected in place. The piece's revision history records the date, what changed and why.
• Errors that change a conclusion are corrected in place and also noted at the top of the piece, so that a reader who relied on the earlier version can see what moved.
• Retracted pieces stay at their address, marked as retracted, with the reason. Links and citations to them do not break.
• Superseded frameworks and pieces are marked as superseded and point to their successor.
• Typographical fixes that change no meaning are recorded with a one-line note.

To report an error, use the Contact page and name the piece, the version you read, and the claim at issue.

Versioning and revision notes

The Institute versions its writing the way software teams version code, because much of its argument is that law should be governed that way too.

• Original dates are kept. A piece first published elsewhere keeps its original publication date, and carries an archival note naming where and when it first appeared.
• Every piece has a revision history. Each entry is a date and a one-line note, like a commit message. The first entry records publication or import.
• Frameworks carry version numbers and changelogs. A change to what a framework requires, or how it is applied, increments the major version. A clarification that leaves its requirements unchanged increments the minor version. Earlier versions remain described in the changelog, so a citation to version 1.0 stays meaningful after version 2.0 appears.
• Theses carry the date first stated and a revision history. A thesis number is never reused for a different claim.
• Status is explicit. Pieces are current, superseded or retracted; frameworks are stable, draft or superseded.

See How to Cite for citing a specific version.

Editorial personas

Much of the Library is published under ten named house voices. They are editorial personas, not external contributors and not real people. Each is a defined writing voice with a characteristic way of arguing, used to test the Institute's ideas from different vantage points and to publish reviews and variations of the Institute's own work.

How the personas work:

• Purpose. A doctrinal argument reads differently to a trial judge, an in-house lawyer, a legal theorist and a critic of institutional power. Writing the same idea in several voices exposes weaknesses that one voice would hide. Persona reviews of the Institute's papers are deliberately critical, and variations re-tell a paper for a different reader.
• How persona pieces are made. Persona pieces are drafted with AI assistance from the Institute's working papers and reviewed by the founder before publication. They add no facts that the source material does not contain.
• No biographies. The personas have no careers, employers, degrees or years of experience, and they never claim lived experience as fact. A persona may write from a vantage point ("from the client's chair") but not from a résumé.
• Disclosure. Every persona piece carries a line identifying its byline as an editorial persona.
• What persona reviews are not. A review written in a persona's voice is internal critique. It is not independent peer review, and the Institute does not present it as such.

AI assistance

The Institute uses AI tools in its editorial work and discloses that use here as a general policy.

• Where AI assistance is used. AI tools may be used to draft, restructure, summarize, translate and check text, and to build the software and data tooling behind the site. All persona pieces are drafted with AI assistance. Pieces published under the founder's name may also have been drafted or edited with AI assistance.
• What AI output is never used for. No citation, quotation, number or factual claim is published on the strength of a model's output. Each is checked against its source before publication.
• Who is responsible. A human reviewer, the founder and director, is responsible for every piece the Institute publishes, whatever tools were used to produce it.
• Why this policy. The Institute argues that legal AI output should be grounded in verified authority and traceable to its sources. Its own publications are held to the same rule.