invalid_decision_questions

invalid_decision_questions means OpenType refused the question set of a decision run. Read this page when a decision run is refused with this code and you need to know which question, dependency or setting to change.

HTTP	`code`	Retryable
`400`	`invalid_decision_questions`	No. Fix the questions first.

What happened

Route: POST /v1/runs with "kind": "decision". The questions, their dependencies, or the draws and think_tokens settings break one of the bounds below. The run was refused before it existed. Nothing was stored and nothing was charged. Every message starts with the decision questions are not acceptable: and ends with one of these reasons. The limits in the right-hand column are not part of the message text.

Questions and answers

Reason	Bound
`a decision needs at least one question`	`questions` holds at least 1 question
`the decision asks more questions than permitted`	at most 64 questions
`every question id must be non-empty and contain no ':' or newline`	an id is not empty and has no `:`, `\n` or `\r`
`two questions share an id`	ids are unique
`every question needs at least two and at most twenty alternatives`	2 to 20 answers per question; a `noul` question always has 2 (`yes` and `no`)
`two alternatives of one question share a name`	option or level names are unique within a question

Dependencies and conditions

Reason	Bound
`a question depends on a question that is not another question in this set`	every `depends_on` entry and every `ask_if` key names another question in the set; a question cannot name itself
`the questions depend on each other in a cycle`	no dependency cycles; forward references are fine
`an ask_if names an answer the question it depends on cannot produce`	each `ask_if` answer is a legal answer of the target: `yes` or `no` for `noul`, an option name for `choice`, a level name for `score`
`an ask_if names no answers, so it can never be satisfied`	each `ask_if` list is not empty

Settings and size

Reason	Bound
`draws must be at least one and at most eight`	`draws` is 1 to 8; the default is 1
`think_tokens must be at most four thousand and ninety-six`	`think_tokens` is 0 to 4096; the default is 0
`the question set is larger than the permitted size`	`instructions`, `questions`, `draws` and `think_tokens` together fit in 32 KiB of canonical JSON; `state` does not count
`the question set could not be encoded`	the set could not be turned into canonical JSON

Two related refusals use other codes. capability_hint problems are reported as invalid_verdict_schema, even on a decision run. A missing questions field, or a question_order that does not name exactly the keys of questions, is invalid_body.

How to fix

Find the reason after the prefix in the tables above.
Check every depends_on and ask_if against the ids in questions.
For ask_if, use answer names, not indexes: yes or no for a noul question, the option name for choice, the level name for score.
Split a set that is over 64 questions or 32 KiB into several runs.
Send the corrected request. It was refused before a run existed, so you may keep the same Idempotency-Key.

Some problems are not caught here. Whether each answer label is a single token, and whether the set fits the answer template, are checked only when the run executes. A failure there ends the run with decision_unavailable. Keep option and level names short, simple words.

Example

{"error":{"code":"invalid_decision_questions","message":"the decision questions are not acceptable: two questions share an id","request_id":"req_7d3f0c1a9b2e4f6a8c0d1e2f3a4b5c6d"}}

A valid set with a condition: bucket is asked only when urgent is answered yes.

curl https://api.opentype.dev/v1/runs \
  -H "Authorization: Bearer $OPENTYPE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ticket-4822-triage" \
  -d '{
    "kind": "decision",
    "instructions": "You triage customer support tickets.",
    "state": {"ticket": "I was charged twice this month and nobody answers my emails.", "plan": "pro"},
    "questions": {
      "urgent": {"type": "noul", "instructions": "reply within the hour?"},
      "bucket": {"type": "choice", "instructions": "which queue?",
                 "criteria": {"billing": "payment problems", "other": null},
                 "ask_if": {"urgent": ["yes"]}}
    },
    "max_output_tokens": 16
  }'

const res = await fetch("https://api.opentype.dev/v1/runs", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.OPENTYPE_API_KEY}`,
    "Content-Type": "application/json",
    "Idempotency-Key": "ticket-4822-triage",
  },
  body: JSON.stringify({
    kind: "decision",
    instructions: "You triage customer support tickets.",
    state: { ticket: "I was charged twice this month and nobody answers my emails.", plan: "pro" },
    questions: {
      urgent: { type: "noul", instructions: "reply within the hour?" },
      bucket: {
        type: "choice",
        instructions: "which queue?",
        criteria: { billing: "payment problems", other: null },
        ask_if: { urgent: ["yes"] },
      },
    },
    max_output_tokens: 16,
  }),
});

const body = await res.json();
if (!res.ok) throw new Error(`${body.error.code}: ${body.error.message}`);
console.log(body.decision.answers);

import os
import requests

resp = requests.post(
    "https://api.opentype.dev/v1/runs",
    headers={
        "Authorization": f"Bearer {os.environ['OPENTYPE_API_KEY']}",
        "Content-Type": "application/json",
        "Idempotency-Key": "ticket-4822-triage",
    },
    json={
        "kind": "decision",
        "instructions": "You triage customer support tickets.",
        "state": {"ticket": "I was charged twice this month and nobody answers my emails.", "plan": "pro"},
        "questions": {
            "urgent": {"type": "noul", "instructions": "reply within the hour?"},
            "bucket": {
                "type": "choice",
                "instructions": "which queue?",
                "criteria": {"billing": "payment problems", "other": None},
                "ask_if": {"urgent": ["yes"]},
            },
        },
        "max_output_tokens": 16,
    },
    timeout=60,
)

body = resp.json()
if not resp.ok:
    raise RuntimeError(f"{body['error']['code']}: {body['error']['message']}")
print(body["decision"]["answers"])

Decision questions - question types, answers and stages.
Conditional questions - depends_on and ask_if in practice.
Limits - question, alternative and size limits in one table.
Problem codes - every code, its status, and whether a retry can help.

Overview

Errors

Limits

Release notes

invalid_decision_questions

What happened

Questions and answers

Dependencies and conditions

Settings and size

How to fix

Example

​What happened

​Questions and answers

​Dependencies and conditions

​Settings and size

​How to fix

​Example

​Related

What happened

Questions and answers

Dependencies and conditions

Settings and size

How to fix

Example

Related