Questions

Noul

A yes-or-no question. The answer is one number: how likely the answer is yes.

Shape

Question

{  "type": "noul",  "instructions": "…",              // optional: the question  "criteria": {                     // optional    "true":  "what counts as yes",  // optional    "false": "what counts as no"    // optional  }}

Answer

{ "type": "noul", "noul": 0.04, "stats": {} }

noul is the probability of yes, between 0 and 1. The probability of no is 1 - noul. A noul answer has no separate confidence: a value near 0 or 1 is a confident answer, a value near 0.5 is not.

Example

The customer wants a replacement, not their money back. The criteria spell out where store credit and chargebacks belong, so the model does not have to guess.

Request

curl -s "https://api.vakyn.com/v1/systemone" \  -H "Authorization: Bearer $TYPESAFE_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "model": "jev-latest",    "state": "Hi, the blender I bought last week stopped working. Can you send me a new one? I do not want my money back.",    "questions": {      "wants_refund": {        "type": "noul",        "instructions": "Is the customer asking for their money back?",        "criteria": {          "true": "They ask for a refund, a chargeback or store credit.",          "false": "They ask for a repair, a replacement or something else."        }      }    }  }'

Try it in the playground

Response · 200

{  "model": "vakyn-fake",  "answers": {    "wants_refund": {      "type": "noul",      "noul": 0.04,      "stats": {}    }  },  "usage": {    "input_tokens": 58,    "output_tokens": 0  },  "assets_used": null}

Describing yes and no

Without criteria, the model weighs a plain "yes" against a plain "no". With them, it weighs your descriptions. Descriptions help most when the boundary is a matter of policy rather than language:

Question	Useful criteria
Is this message spam?	`true`: unsolicited promotion or phishing; `false`: anything from a real customer, even if rude
Does the invoice need approval?	`true`: total over 5,000 or a new supplier; `false`: everything else
Is the reviewer recommending the paper?	`true`: accept or minor revisions; `false`: major revisions or reject

You can describe only one side; the other stays a plain yes or no.

Using the answer

Pick the threshold from what a mistake costs, not 0.5 by default. Flagging an urgent message at noul > 0.3 trades a few false alarms for fewer misses.
Keep a middle band for people: act automatically above 0.9 or below 0.1, ask a human in between. See routing by confidence.
One fact per question. "Is it urgent and about billing?" hides which half was true; ask two nouls in the same request instead.