Grade: marketing translation fluency

The job

Jev judge eval (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.

// Grade: marketing translation fluency — Jev judge eval (typesafe-ai/jev).
// An agent translates a tagline into German; the judge scores fluency and fidelity.
// 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: "Source: Ship calm software, every single week.\nTranslation: Liefere ruhige Software, jede einzelne Woche.",
  questions: {
    "fluency": {
      "type": "score",
      "instructions": "Score this German marketing translation. Higher is better.",
      "criteria": [
        "Wrong meaning or ungrammatical.",
        "Understandable but awkward or overly literal for marketing copy.",
        "Fluent and accurate, though slightly flat as a tagline.",
        "Fluent, idiomatic, and punchy — reads like native marketing copy."
      ]
    }
  },
});

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