What is a decision model

Short definition

• In: a context (a message, a record, a JSON object) and one or more questions, each with its possible answers.
• Out: one answer per question, always in the requested format, with the probability of each option.
• Does not: write, summarize or chat. It never invents an option outside the list.
• For whom: for software. The system reads the answer and acts, without parsing text.

Fast judgment, not long reasoning

The difference echoes the two modes of thought described in psychology and popularized by the book Thinking, Fast and Slow (2011): one fast and automatic, the other slow and deliberate. A decision model aims for fast judgment: a short answer, in milliseconds, with the uncertainty in plain sight.

The three kinds of question

Names vary from vendor to vendor; the idea is the same.

Example

A POST /v1/decide request to Iryx with three questions about the same message:

{
  "context": {"message": "I was charged twice on my card for the same order. I need this fixed today."},
  "decisions": {
    "topic":  {"kind": "pick", "prompt": "What is the message about?",
               "options": {"billing": "charges, refunds, cards",
                           "shipping": "delays, tracking, address",
                           "returns": "exchanges or returns",
                           "other": "anything else"}},
    "urgent": {"kind": "check", "prompt": "The customer needs a reply today"},
    "mood":   {"kind": "scale", "prompt": "Customer mood",
               "levels": ["calm", "worried", "annoyed", "furious"]}
  }
}

For each question, the response carries the choice and the probabilities (illustrative values, not measurements):

"results": {
  "topic":  {"kind": "pick", "value": "billing", "confidence": 0.94,
             "spectrum": {"billing": 0.94, "shipping": 0.02, "returns": 0.02, "other": 0.02}},
  "urgent": {"kind": "check", "value": 0.88},
  "mood":   {"kind": "scale", "value": 1.6, "level": "annoyed",
             "spectrum": {"calm": 0.05, "worried": 0.35, "annoyed": 0.55, "furious": 0.05}}
}

Every field and rule is in the API documentation.

When to use it and when not to

Good fit

• Classifying and routing: tickets, e-mails, comments, system events.
• Filtering and moderation: "does this break rule X?", with a probability.
• Scoring on a scale: priority, risk, tone, severity.
• Real-time, high-volume decisions.
• A second opinion for a software agent before it acts.

Not a fit

• Writing, summarizing, translating or chatting: Iryx does not generate text; it decides.
• Open questions with no list of answers.
• Facts that are not in the context: the model decides about what you send.

Probabilities and the confidence threshold

The practical difference is the probability. With it you pick a threshold: above it, the system acts on its own; below it, a person checks. The spectrum also shows the runner-up, which helps whoever checks.

A probability only helps if it is calibrated: among answers at 0.9, about 90% should be right. Iryx publishes the accuracy of Iryx 1.0 on its own answer key (see the numbers). A calibration measurement has not been published yet.

Questions

Is a decision model the same as a classifier?

It is a relative. A typical classifier is trained for a fixed set of labels. In a decision model, the options come with each question, on every request: you can change the options whenever you like.

Can't a large language model (LLM) do this?

It can, with a good prompt and structured output. But it generates text token by token, bills per token and may break the format. When it answers, a decision model returns the requested format, with probabilities.

How is Iryx 1.0 different from other decision models?

Its API is its own (POST /v1/decide, with pick, check and scale), its published accuracy numbers are in Portuguese and the company is Brazilian. In the 2026-10-19 round, among 11 decision models, it had the highest accuracy on Iryx's own test: 83.1% on 225 decisions, in Portuguese.

Can I use Iryx today?

The API is not open to the public yet. Access starts with a reservation: you create an account. The Iryx 1.0 page has a demo with prepared answers.

Sources

• Iryx, API documentation
• Book: Thinking, Fast and Slow, 2011