First decision

One request, three typed answers, and what every field in them means.

At the end of this page you have asked Jev a yes/no question, a pick-one question and a rate-it question about the same sentence, in one request, and you can read the answer without guessing.

The call

jev_ask is the raw API as a tool. A Strands Agent exposes every tool for direct calls under agent.tool.<name>, no chat model involved, so the whole thing is one Python call:

from strands import Agent
from strands_jev import ALL_TOOLS

agent = Agent(tools=ALL_TOOLS)
result = agent.tool.jev_ask(
    state="Help! My payouts have been failing for 3 days.",
    questions={
        "urgent": {"type": "noul", "instructions": "Does this convey urgency?"},
        "team": {
            "type": "choice",
            "instructions": "Which team should take this?",
            "criteria": {"billing": "payments, charges, payouts", "technical": "bugs and outages", "sales": "plans and pricing"},
        },
        "frustration": {
            "type": "score",
            "instructions": "How frustrated is the writer?",
            "criteria": ["calm", "frustrated", "very angry"],
        },
    },
)
print(result["content"][1]["text"])
answers = result["content"][0]["json"]["answers"]

The second content block is a one-line summary for a model to read; the first is the JSON. One recorded run of this exact request on 2026-09-29 came back in 178 ms for 434 input tokens, which is $0.000018 at $0.042 per million.

Reading the answers

{
  "urgent": {"type": "noul", "noul": 0.95},
  "team": {"type": "choice", "choice": "billing", "confidence": 0.98,
           "probabilities": {"billing": 0.99, "technical": 0.01, "sales": 0.0}},
  "frustration": {"type": "score", "score": 1.04, "confidence": 0.94,
                  "legend": {"0": "calm", "1": "frustrated", "2": "very angry"},
                  "probabilities": {"0": 0.0, "1": 0.96, "2": 0.04}}
}

field	on which type	what it is
noul	noul	The probability that the answer is yes. Compare it to a threshold you choose; the package default in strands_jev/questions.py is 0.5.
choice	choice	The option with the highest probability, by the key you gave in criteria.
confidence	choice, score	How sure the model is of the winning option or level, 0 to 1. A three-way choice split 0.4/0.3/0.3 has a winner and a low confidence at once.
probabilities	choice, score	Every option or level with its probability. Read this when the second place matters, or when you want your own threshold.
score	score	The probability-weighted position on the scale, index 0 the first level: 1.04 sits just above "frustrated". Do not average scores across items; read the legend.
legend	score	Index to level name, so a score is never a bare number in a log.
model	the result	The versioned id that answered, jev-1.13.0 on 2026-09-29, even when the request said jev-latest.
latency_ms, input_tokens	the result	This request's wall clock and billed tokens.

What just happened

Three questions, one forward pass. Questions cannot see each other's answers, so a follow-up question is asked alongside the question that decides whether it matters; the fan-out pattern builds on that. Keys such as urgent are yours; the model never sees them. Everything the model does see is state, instructions and criteria, so that is where the care goes: State.

Refusals

The tools refuse before they spend. A criteria list of one level for a score, a type that is not one of the three, a questions value that is not a mapping, list or JSON string: each returns status: "error" with the fix in the text and no request sent. An oversized request is refused the same way by the budget check, described on What it costs.

Edit page