Clef question types: noul, choice and score
Every Clef question has a type, and the type decides what comes back. Here is what each returns, a real Clef-Flash output for each, and how to choose.
| Type | Use it when | You get back | Example |
|---|---|---|---|
noul | the answer is yes or no | one number: the probability of yes | Is this message spam? |
choice | exactly one of several labels fits | the top option, a probability per option, a confidence | Which team should handle this? |
score | the answers are ordered levels | a weighted score, a probability per level, a confidence | How severe is the impact? |
noul: yes or no
Ask one clear yes/no question in instructions. No criteria needed. The answer looks like {"type":"noul","noul":0.97}: the probability that the answer is yes.
{
"urgent": {
"type": "noul",
"instructions": "Is this support request urgent?"
}
}Real result for “Checkout has been failing for every customer for the last hour.”:
Probability of yes: 0.97
| Answer | Probability |
|---|---|
| Yes | 96.8% |
| No | 3.2% |
clef-flash · 169 input tokens
choice: pick one label
criteria is an object of option id → description. Describe each option in a few concrete words; the description matters more than the id.
{
"team": {
"type": "choice",
"instructions": "Which team should handle this request?",
"criteria": {
"billing": "Payments, invoices and refunds",
"technical": "Outages, errors and configuration",
"sales": "Plans and upgrades"
}
}
}Real result for the same ticket:
Most likely: Technical: outages, errors and configuration (confidence 92.7%)
| Answer | Probability |
|---|---|
| Billing: payments, invoices and refunds | 1.7% |
| Technical: outages, errors and configuration | 97.5% |
| Sales: plans and upgrades | 0.8% |
clef-flash · 186 input tokens
score: rate on ordered levels
criteria is an array of levels from lowest to highest. The answer’s score is weighted by probability, so you can sort by it or set a threshold between levels.
{
"severity": {
"type": "score",
"instructions": "How severe is the customer impact?",
"criteria": [
"No impact",
"Minor",
"Major",
"Critical"
]
}
}Real result:
Weighted score: 2.68 (confidence 44.9%)
| Answer | Probability |
|---|---|
| 0 · No impact | 1.8% |
| 1 · Minor | 1.5% |
| 2 · Major | 24.1% |
| 3 · Critical | 72.7% |
clef-flash · 186 input tokens
How to choose a type
- One property, true or false → noul.
- Several categories and exactly one applies → choice.
- Levels with an order (none < minor < major) → score. A choice question would throw the order away.
- Several labels that can all apply (spam and abusive) → several noul questions in one request.
Acting on the answer
Don’t just take the top option. Decide what happens at each probability: act automatically above a line you trust, send the middle band to a person, and ignore the rest. Run 20–50 of your own examples first to see where right and wrong answers fall.
More patterns with real outputs: 10 Clef examples. Full request and response reference: Clef API guide. Try a question now in the playground.
FAQ
Can a choice question return more than one option?
No. The probabilities across options sum to 1 and the answer names one top option. When several labels can be true at once, ask one noul question per label in the same request.
Is the score always a whole number?
No. It is weighted by the probability of each level, so it can land between levels. In the severity example the score was 2.68 on a 0–3 scale: mostly “Critical”, with some weight on “Major”.
What is confidence?
Choice and score answers include a confidence value between 0 and 1. Use it together with the probabilities: a low value means the model is unsure and a person should look.
How many questions can one request have?
Up to 64. They all judge the same state, so asking them together is cheaper than separate requests.
Outputs above were produced by Clef-Flash on 2026-10-02.