Judge Jev

At a glance

Judge Jev is a short fictional courtroom game: read five exhibits, write a defense within three minutes and receive an AI verdict. It works as a compact evidence-reading puzzle for someone who enjoys finding contradictions and committing to an argument under a clock. The payoff is a verdict with model probabilities, rather than a conversation explaining which parts of your defense persuaded the judge.

Good fit

A brief logic-game detour where the supplied record matters more than inventing a dramatic backstory. The live rules describe fifty stories; repository documentation explains 150 authored evidence variants with mechanically varied dates, places and quantities. Case generation itself makes no model call. The accusation, exhibits and your response then go to Jev from TypeSafe AI for a typed guilty-or-not-guilty decision.

Consider something else when

You want legal instruction, a free-form role-playing conversation or detailed feedback on your writing. The actual result screen supplies a verdict and two probabilities without a case-specific explanation. Those percentages are model outputs, not validated probabilities of real-world guilt. The appeal is weighing the evidence yourself; a short round cannot establish the judge’s general accuracy or consistency.

Start here

Read “How to play” before pressing Play, because dealing the case starts the three-minute timer. Read all five exhibits and explain a concrete gap or connection in at most 1,000 characters, then press “Plead your case.” At zero, the app submits the current response automatically. One completed trial returned GUILTY with 67% guilty and 33% not-guilty probabilities; the available next step was another case.

Cost and requirements

A browser and network connection were enough for the hosted trial, with no account or payment step. Availability depends on the host’s model connection and rate limits. The reviewed backend sends the fictional record and defense to TypeSafe’s API. TypeSafe’s privacy policy says it does not train on submitted inputs, but retains personal data as reasonably necessary for stated purposes; that is not a fixed deletion deadline.

Limits to know

The rules ask the model to evaluate the whole record, so a weak defense alone is not meant to prove guilt and a supported accusation may still withstand a clever objection. The trial established one complete round, not fairness, calibrated probabilities or the quality of all scenarios. App-host retention is not established by the model provider’s policy. Keep the exercise within its supplied fictional facts instead of treating it as a place to submit a real case.

Related work

• TypeSafe AI (context)

Still unclear

• A fixed retention period for submitted defenses at the app host and model provider.