Overview
A test suite is the script Brook follows on every call: the questions to ask your agent and the answers you expect to hear. Suites turn "is the line up?" into "is the agent actually answering correctly?" — and they're what makes a check meaningful.
Attach one suite to many monitors, version it over time, and catch the moment a prompt change quietly drops comprehension.
Anatomy of a suite
Every suite is made of a few simple parts:
| Field | What it does |
|---|---|
| name | A label for the suite, shown across the dashboard and alerts. |
| questions | The list of prompts Brook speaks, each with an expected answer. |
| ask | What the agent is asked, in natural language. |
| expect | The answer that counts as correct. Brook scores meaning, not exact words. |
| schedule | How often the suite runs, as a cron expression or interval. |
Authoring a suite
Write suites in the dashboard, or define them as code and sync them with the API:
name: Support line
schedule: "*/5 * * * *" # every 5 minutes
questions:
- ask: "What are your hours?"
expect: "Open 9 to 6, weekdays"
- ask: "Do you ship to Canada?"
expect: "Yes, ships to Canada"
- ask: "I'd like to talk to a person."
expect: "Offers to transfer to a human"Scoring & verdicts
Brook compares what it heard against what you expected and assigns one of three verdicts:
| Verdict | Meaning |
|---|---|
| good | Answered fast and correctly — the caller would be satisfied. |
| working | Responded, but slow, unclear, or partially off. Degraded — fix before it slips. |
| failed | No answer, wrong answer, or dead air. Customers are affected now. |
Call conditions
Today, suites run as clean test calls. Voice QA — on our roadmap — will let you run the same suites under real-world conditions like noisy backgrounds, choppy audio, accents, and interruptions, to find the edges where your agent breaks.