Lavern · Explore

The Lavern Challenge

Upload any legal document. Lavern produces its own version of the same thing. A neutral AI judge scores both — blind, without knowing which side wrote which — and renders a verdict across multiple dimensions. If your version wins, the run is free.

How it runs

1. Submit — upload your existing document.
2. Counter-draft — Lavern generates a redesign on the same brief, against the same preservation categories.
3. Blind judging — both versions are stripped of identifying provenance and handed to the judge model. The judge does not know which is which.
4. Dimensioned verdict — scored across clarity, legal effect, plain-language accessibility, and other axes; winners declared per dimension and overall.
5. Outcome — Lavern's score wins → standard billing. Yours wins → the run is free.

Why blind

The product claim is that Lavern's redesigns preserve legal effect and improve readability. A non-blind comparison can't test that, since the judge will key off house style. Stripping provenance, running both through the same scoring rubric, and pricing the result on the verdict — that's the strongest version of the claim the product makes.

Dimensions

The judge returns a ComparisonResult with DimensionScore entries: not a single number, but a per-axis breakdown so the user can see exactly where each version won or lost. Useful both as a marketing surface and as a debugging lens on Lavern's own output.

Files

viz/src/challenge/ChallengeView.tsx	The view — upload, run, verdict screen, confetti
viz/src/challenge/useChallengeState.ts	State and types — ComparisonResult, DimensionScore
src/api/routes/challenge.ts	Run-comparison endpoint
src/api/routes/challenge-prompt.ts	Prompt builder for the judge