---
title: "What a decision model returns"
lesson_id: "01"
---

# What a decision model returns

- **Lesson ID:** 01
- **Goal:** Call a local decision model and read its typed answers.
- **Human lesson:** [01-what-a-decision-model-returns.html](01-what-a-decision-model-returns.html)

## Prerequisites

- llama.cpp installed (a 2026-10 build or newer).
- uv and Python 3.11+.

## Inputs, outputs, and artifacts

- **Inputs:** A state text and typed questions.
- **Outputs:** Answers with probabilities and confidence.
- **Artifacts:** A saved example request and response.

## Agent build steps

1. Do the environment setup in the human lesson.
2. Start `llama-server -hf ggml-org/Kev-0.8B-GGUF:Q8_0 --port 8080 -np 1 -c 4096`.
3. Confirm `output_modalities` contains `decisions`.
4. Send the three-question request from the lesson.
5. Note the confidence of each answer, not only the pick.

## Constraints

Keep the language simple, friendly, and literal. No quizzes, no em dashes. Brand Edge FDE only. Use small local models. Do not invent flags, endpoints, or issue numbers. 

## Key concepts

- `/v1/systemone` takes a `state` and named `questions`.
- Types are `choice`, `score`, and `noul`.
- Answers include probabilities and a 0 to 1 confidence.
- Probabilities are not calibrated for your data.

## Takeaways

- Check `output_modalities` for `decisions` before calling.
- Read confidence with every answer.
- Plan to test on your own labelled data.

## Acceptance checks

- The request shape matches the llama-server README.
- The low-confidence score example is explained.
- The calibration warning from the README is quoted.
- [ ] Proceed to [lesson 02 brief](02-labelled-set-and-order-bias.llms.md).