Skip to main content

About

Why DraftGrade exists.

The most expensive paper in America is the one you file in court without knowing your cases are real.

The problem

In 2024, the Stanford RegLab found that large language models hallucinate 58–88% of legal citations they produce (Dahl et al.). That is not a rounding error. It means a pro se litigant who uses ChatGPT to draft a motion will, on the modal day, file a brief whose case law does not exist. The consequences range from embarrassment to Rule 11 sanctions to losing a meritorious case on a procedural ground that had nothing to do with the merits.

Meanwhile, the legal-tech market has answered this problem with subscription tools costing hundreds of dollars per month, priced for big-firm legal departments. The people most likely to lean on an LLM — pro se litigants, public defenders, legal-aid attorneys, solos — are the people who can least afford the verification layer.

What we built

DraftGrade is two things, deliberately:

  1. An open methodology — a seven-section rubric, a thirty-two-technique register, and a calibration set that proves the rubric discriminates between expert and amateur drafting. The rubric is CC-BY-SA. The system prompt that operationalizes it is AGPL. The calibration set is CC-BY-SA. Fork any of it. Audit any of it. Run it yourself. The full methodology lives at this URL, not behind a login wall.
  2. A proprietary SaaS that operationalizes it — paste a motion, get a citation-verification report against the canonical CourtListener / Free Law Project database, stored in your account so you can revisit it. Free during MVP, including the v1 API. No card, no plan, no metering.

The two halves serve different audiences. The open methodology serves the legal-tech ecosystem — vendors, researchers, court technology teams, integrators. The SaaS serves the end user who has a motion they need to file tomorrow.

What we believe

The model

The methodology lives under permissive open-source licenses so it can't be enclosed. The SaaS lives under a proprietary license so the company can sustain itself if a tier model ever makes sense. The two communicate one way: the SaaS reads the open methodology as text and data, but never imports its code. That separation is enforced by tests, not just convention.

We are looking for a single anchor sponsor to fund the free-for-pro-se-litigants tier indefinitely. If your organization shares the access-to-justice mission, see /sponsor.

How to help

Honest constraints

DraftGrade is early. The rubric scoring engine (LLM-backed) ships later; today only the citation verifier is in-product. We do not currently offer email verification, multi-factor authentication, OAuth login, or jurisdiction-specific motion templates. We will, in roughly that order. The roadmap is public via the GitHub issues and the methodology version pin at the repo root.