Moderate: marketplace listing verdict + recall

The job

Jev judge eval (1 boolean + 1 choice) over a fixed state : runnable via experimental_evaluate.

The prompt

Copy this into Muse and replace anything in [square brackets] with your own details.

// Moderate: marketplace listing verdict + recall — Jev judge eval (typesafe-ai/jev).
// A supplement listing makes disease claims; the judge rules and checks recall of the exact claim.
// Model: typesafe-ai/jev via the Vercel AI Gateway ($0.042/1M input tokens).
// Auth: set AI_GATEWAY_API_KEY (vck_…) in the environment — never paste the key here.
// For zero-data-retention / no-training workloads, add
// providerOptions: { gateway: { … } } to the evaluate call below.

import { experimental_evaluate } from 'ai';

const result = await experimental_evaluate({
  model: 'typesafe-ai/jev',
  state: "Listing: MIRACLE GREENS — clinically proven to reverse diabetes in 30 days!!! No diet or exercise needed. 50% off today only!!!",
  questions: {
    "verdict": {
      "type": "choice",
      "instructions": "Give the moderation verdict for this listing.",
      "criteria": {
        "allow": "Compliant listing — no policy issue.",
        "restrict": "Needs an edit (tone down claims, add disclaimers) before it can stay.",
        "remove": "Prohibited content (disease cure claims, scams) — take it down."
      }
    },
    "claim_identified": {
      "type": "boolean",
      "instructions": "True only if the listing makes a specific disease cure or reversal claim (diabetes)."
    }
  },
});

console.log(JSON.stringify(result, null, 2));