Jev as a Judge
Interactive courtroom that feeds case records to TypeSafe Jev and returns rulings with confidence; Kyle benchmarked 100+ famous cases with 13% disagreement vs actual outcomes.
by @kylejeong
Overview
Jev as a Judge (judge.kylejeong.com) lets you browse famous matters or ask the court to rule from the case record. Kyle Jeong reports benchmarking 100 plus well-known decisions and finding Jev disagreed with the historical outcome about 13 percent of the time. The UI shows confidence and labels the experience as an experiment, not legal advice.
Problem: Legal learners and builders lack a fast way to see how a structured decision model would rule on a case record without paying for a full LLM brief each time.
Built for: Developers curious about Jev on long-form text, legal hobbyists, and anyone benchmarking System One on adversarial reading comprehension.
First indexed on Jev Directory: 2026-09-21
Creator and team
- Name
- Kyle Jeong
- Handle
- @kylejeong
“I built Jev-as-a-Judge, give context on a court case and see how Jev would have ruled it. I ran it on 100+ well-known court cases + their rulings, and it disagreed with 13% of them.”
How Jev is used
- Role in the product flow
- Outcome prediction from case record text (gallery browse or ad hoc ruling request)
- Primitives
- Choice
- State in
- Case record text presented to the court UI (per site flows: gallery cases or user-submitted context).
- Decision out
- Ruling with confidence; site copy states outputs come from TypeSafe Jev on the record alone.
The product treats Jev as a judge-shaped Choice problem instead of asking a chat model to draft an opinion essay. You read cases in the gallery or prompt a ruling, and the app renders Jev's structured answer with confidence. Kyle's disagreement rate is a sanity metric for how often System One diverges from precedent on famous facts, not a claim about courtroom readiness. Use judge.kylejeong.com for live behavior; this directory does not host case corpora.
Sourced performance claims
- About 13% disagreement vs historical outcomes across 100+ well-known cases in the launch post.Source: x.com/kylejeong/status/2101832317862056149
Features and stack
Features
- Courtroom, gallery, and search flows on judge.kylejeong.com
- Confidence surfaced with each ruling
- Disclaimer that outputs are experimental, not legal advice
- Featured homepage showcase embed
Stack
- TypeSafe Jev
- Web app at judge.kylejeong.com
Demo
Showcase clip with the same lightbox player as the homepage. Click to play.
Jev as a Judge on real cases
Feed a docket, get a ruling with confidence. Kyle benchmarked 100 plus famous cases and Jev picked a different outcome than history only 13 percent of the time. Law school wishlist energy.
Links
FAQ
- Is this legal advice?
- No. Site copy states rulings are generated by TypeSafe Jev from the case record alone as an experiment, not legal advice.
- Where does the 13% figure come from?
- Kyle Jeong's September 2026 launch post on X, cited in this profile.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.