POST /v1/systemone
The TypeSafe System One shape. Instead of an input and an array of
questions, you send a state and a keyed questions object; answers come
back keyed by the same names. It carries exactly the same information as
/v1/decisions — pick whichever shape matches your existing
client.
POST /v1/systemone
Authorization: Bearer shrug_sk_your_key_here
Content-Type: application/json
Question types
| type | Shape | Answer field |
|---|---|---|
| choice | criteria: { name: description } | choice + probabilities{} |
| noul | instructions | noul (p of true) |
| score | criteria: [label, …] | score (index) + probabilities[] |
Note the TypeSafe predicate type is noul (the System One name), the
analogue of the predicate type on /v1/decisions. The optional
shrug.distributions block works identically — see the
density extension.
Request
The canonical request fixture. questions is an object keyed by question name,
and criteria carries the choice/score labels:
{
"model": "system-one",
"state": "the sky at noon",
"questions": {
"color": { "type": "choice", "instructions": "pick a color",
"criteria": { "red": "warm", "blue": "cool", "green": "earthy" } },
"is_daytime": { "type": "noul", "instructions": "is it day?" },
"brightness": { "type": "score", "instructions": "how bright",
"criteria": ["dim", "bright"] }
},
"shrug": {
"distributions": {
"color": { "weights": { "red": 0.5, "blue": 0.3, "green": 0.2 } },
"is_daytime": { "p_true": 0.7 }
}
}
}
Response
answers is keyed by question name. A choice returns the drawn choice with a
probabilities object and confidence; a noul returns a noul probability; a
score returns the drawn score index with a legend and a positional
probabilities array.
{
"model": "system-one",
"answers": {
"color": { "type": "choice", "choice": "red", "confidence": 0.5,
"probabilities": { "red": 0.5, "blue": 0.3, "green": 0.2 } },
"is_daytime": { "type": "noul", "noul": 0.7 },
"brightness": { "type": "score", "score": 1, "confidence": 0.6,
"legend": ["dim", "bright"],
"probabilities": [0.4, 0.6] }
},
"usage": { "input_tokens": 0, "output_tokens": 0 }
}