Independence and Disclosure

The relationship

Legawrite.AI is the Institute's founding sponsor and its applied testbed. The Institute's founder and director, Ross Brodskiy, is the founder of Legawrite.AI. The research published here grew out of the work of building Legawrite.AI's systems, and ideas developed here are tested in those systems. Problems met in building them become research questions here.

The Institute does not claim a distance it does not have. Its independence rests on something else: disclosure of the interest, publication of the methods, and standards written so that the sponsor's own products can fail them. Readers should weigh the work with that in mind, and the most direct way to weigh it is to rerun it.

The founder's interest

The founder has a financial and proprietary interest in Legawrite.AI. The Institute discloses that interest in the form first used in the paper Good Law for What?, whose disclosure note reads:

the author has a financial and proprietary interest in commercial software that performs proposition-level legal knowledge extraction and citation validation, and related patent applications are pending. That interest bears directly on the argument advanced here, which is why this article proposes an evaluation standard that the author's own products can fail, states it in terms any librarian can apply without vendor cooperation, and describes no proprietary system.

That note is the model for everything on this site. The interest is stated plainly, its relevance to the argument is admitted, and the response to it is structural rather than rhetorical.

Editorial rules

1. A disclosure line on every piece that discusses Legawrite.AI products. Any piece that names, describes, evaluates or reports results from a Legawrite.AI product carries a disclosure line stating that Legawrite.AI is the Institute's founding sponsor and that the founder has a financial and proprietary interest in it. The line appears on the piece itself, not only on this page.
2. Methods are published so anyone can check them. Frameworks, rubrics, protocols and benchmark designs are published in full, with version numbers and changelogs. Claims are stated in terms a reader can test without the cooperation of any vendor, Legawrite.AI included.
3. Standards are written so that Legawrite.AI's own products can fail them. The Institute's standards start from the legal problem (what a court will accept, what a brief must contain, what a lawyer must be able to trace) and not from any product's feature list. A standard that the sponsor could not fail would not be worth publishing.
4. No proprietary internals. The Institute does not describe the proprietary internals of Legawrite.AI systems. Where a piece refers to such a system, it treats it as a black box, as the source papers do, and its arguments do not depend on taking that box on trust.
5. Failures are reported with results. Where the Institute reports results from work done on Legawrite.AI systems, it reports what failed as well as what passed. The blind eDiscovery review published on Medium in July 2026 is the pattern: the failed elusion and calibration gates were reported next to the recall figures, and the review engine was open-sourced under the Apache 2.0 license so that others could rerun it.
6. Other vendors on the same terms. When a piece discusses other research products, it holds them to the same standards it applies to Legawrite.AI and describes them only from public information.

Raising a concern

If you believe a piece falls short of these rules, whether through a missing disclosure line, a standard tilted toward the sponsor, or a claim that cannot be checked, write to the research desk through the Contact page. Where a concern leads to a change, the change is recorded in the piece's revision history under the Institute's Editorial Standards.