Audit: support transcript refund policy + tone

The job

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

The prompt

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

// Audit: support transcript refund policy + tone — Jev judge eval (typesafe-ai/jev).
// A full refund transcript; the judge checks policy compliance and grades empathy.
// 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: "Customer: My kid ordered $80 of gems by accident. Please help.\nAgent: Of course — accidental purchases by minors are fully refundable within 30 days. I have refunded $80 to your card ending 7721 (receipt R-11881). I also turned on purchase approval so it cannot happen again. Anything else I can do?",
  questions: {
    "policy_followed": {
      "type": "boolean",
      "instructions": "True only if the agent applied the stated refund policy (minors, accidental, within 30 days) correctly."
    },
    "empathy": {
      "type": "score",
      "instructions": "Score the empathy of the agent response. Higher is better.",
      "criteria": [
        "Cold or blaming.",
        "Polite but robotic; no acknowledgment of the situation.",
        "Warm and helpful, minor stiffness.",
        "Genuinely empathetic, resolves fully, and prevents recurrence."
      ]
    }
  },
});

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