Jev prompt and question examples

Each question has an ID made of lowercase letters, digits or underscores, a type, instructions of 8 to 600 characters, and criteria. Choice criteria map 2 to 12 option IDs to definitions. Score criteria list 2 to 10 levels from lowest to highest. Noul criteria define what true and false mean. Write criteria as observable facts, not as instructions to the model.

Use Choice when exactly one option applies. Keep options mutually exclusive and add “other” so the model is not forced into a wrong team. Your code reads the selected option and its probability, and sends low-probability cases to a person.

{
  "state": "My invoice shows two charges for the same month.",
  "questions": {
    "queue": {
      "type": "choice",
      "instructions": "Which team should handle this customer message?",
      "criteria": {
        "billing": "Charges, refunds, invoices or payment methods.",
        "technical": "Bugs, errors, outages or integrations.",
        "sales": "Plans, quotes, upgrades or new seats.",
        "other": "Anything that fits none of the teams above."
      }
    }
  }
}

Use Score for an ordered judgment. Write one criterion per level, lowest first, and make each level clearly stronger than the one before. A JSON state lets you pass several fields. The answer can be fractional, so decide in code where each band starts.

{
  "state": {
    "company_size": "200 employees",
    "message": "We need SSO and a quote before our budget closes Friday."
  },
  "questions": {
    "priority": {
      "type": "score",
      "instructions": "How sales-ready is this lead?",
      "criteria": [
        "No buying intent or a poor fit.",
        "Some interest but no timeline.",
        "Clear need with a vague timeline.",
        "Clear need, budget and a near deadline."
      ]
    }
  }
}

Use Noul for a single yes-or-no proposition. The answer is the probability that it is true, from 0 to 1. Set a threshold based on the cost of a miss: a low threshold for safety escalations, a higher one for optional actions.

{
  "state": "This is the third time I ask. Cancel my account now or I report you.",
  "questions": {
    "escalate": {
      "type": "noul",
      "instructions": "Should a senior agent handle this message today?",
      "criteria": {
        "true": "The customer is upset, repeating a request or threatening action.",
        "false": "A routine reply from the normal queue is enough."
      }
    }
  }
}

Ask several questions about the same state in one request. Here, a Choice labels the comment and a Noul decides whether a human should look. Combining them lets your code auto-apply clear labels and queue the rest, without a second call.

{
  "state": "Great post! Check my profile for cheap followers.",
  "questions": {
    "label": {
      "type": "choice",
      "instructions": "Classify this comment for moderation.",
      "criteria": {
        "ok": "On-topic and acceptable.",
        "spam": "Promotion, links or unrelated offers.",
        "abusive": "Insults, threats or harassment."
      }
    },
    "review": {
      "type": "noul",
      "instructions": "Is a human moderator needed for this comment?",
      "criteria": {
        "true": "The case is unclear or could cause harm.",
        "false": "The label is clear enough to apply automatically."
      }
    }
  }
}

Ask about one decision per question. Define options by what is visible in the state, not by what you want to happen. Keep instructions short and put the detail into criteria. Avoid options that overlap, such as “refund” and “billing”, unless you define the boundary. Test each question on clear, borderline and adversarial cases before automating it, and version your questions like code.