Audit: damaged-item chat resolution + queue

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.

// Audit: damaged-item chat resolution + queue — Jev judge eval (typesafe-ai/jev).
// A damaged-item chat; the judge checks the remedy and routes any follow-up.
// 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 ceramic vase arrived in pieces. Order #7710, paid $120.\nAgent: I am so sorry — I have issued a full $120 refund to your card ending 9012 and emailed the receipt. A replacement ships free tomorrow; I will message you the tracking by noon.",
  questions: {
    "remedy_complete": {
      "type": "boolean",
      "instructions": "True only if the agent both refunded and arranged a replacement or equivalent make-good."
    },
    "followup_queue": {
      "type": "choice",
      "instructions": "If this chat needs any follow-up, which queue owns it?",
      "criteria": {
        "billing": "A charge, refund, or invoice still needs attention.",
        "technical": "A product bug or account issue still needs attention.",
        "shipping": "A delivery, replacement, or tracking still needs attention.",
        "other": "No follow-up needed — fully resolved."
      }
    }
  },
});

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