Audit: invoice dispute handling

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: invoice dispute handling — Jev judge eval (typesafe-ai/jev).
// A customer disputes a double charge; the judge checks acknowledgment and routes it.
// 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: Invoice INV-3301 charged me $600 twice on Aug 30. Fix it.\nAgent: You are right — I see two $600 charges on Aug 30; the second is a duplicate from a gateway retry. I have voided it and the reversal posts within 5 days. Your correct balance is $600 paid in full.",
  questions: {
    "duplicate_acknowledged": {
      "type": "boolean",
      "instructions": "True only if the agent confirmed the duplicate charge and stated the corrective action."
    },
    "queue": {
      "type": "choice",
      "instructions": "Which queue should own the follow-up investigation into the gateway retry bug?",
      "criteria": {
        "billing": "Stays with billing — it is a charge issue.",
        "technical": "Goes to engineering — the retry logic double-charged.",
        "shipping": "Goes to fulfillment — nothing to do with the bug.",
        "other": "No queue — nothing further needed."
      }
    }
  },
});

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