Choice: “Which team handles this?”
Use Choice when your code needs one category from a list. For support tickets, describe shipping, billing and returns, plus other when none fits. Distinct category descriptions matter more than clever label names.
Score: “How relevant is this passage?”
Use Score for ordered descriptions, such as irrelevant, partly relevant and directly answers the question. With those three levels, the scale is 0–2 and can include fractional values. Adding a fourth level changes the scale; an old cutoff will no longer mean the same thing.
Noul: “Does this ask for urgent help?”
Use Noul for a single yes/no proposition. The answer is a number from 0 to 1 under noul. It has no separate confidence field. Keep “Is this urgent?” separate from “Is the customer angry?” so you can inspect each judgment.
Several questions can share the same state
You can ask about department, urgency and frustration together. Each question sees the state independently. If your code merely needs to choose which answer to use, it can do that after one call. A later question that needs an earlier answer as new input requires another call.
Read the right field
Choice returns choice, Score returns score, and Noul returns noul. Choice and Score also include confidence. Do not write one parser that assumes all three answers have the same fields.
Request example
{
"questions": {
"urgent": {
"type": "noul",
"instructions": "Does the customer explicitly ask for immediate help?"
},
"frustration": {
"type": "score",
"instructions": "Rate the customer frustration.",
"criteria": [
"Calm",
"Concerned",
"Very angry"
]
}
}
}