# Trust Lab

Define the conversations your agent must always get right, then run them on a schedule to catch regressions before users do.

<HeroTrustLab />

Trust Lab is Foldspace's test automation module. It lets you define conversations the agent must handle correctly, then run them automatically to catch regressions before they reach users. Access it by selecting **Trust Lab** from the top-left app switcher.

A [scenario](/user-guides/trust-lab/scenarios/) is a single test conversation and the evals that grade it, a [suite](/user-guides/trust-lab/test-suites/) groups related scenarios into the unit you run and track, [Runs](/user-guides/trust-lab/runs/) is the execution log, [Schedules](/user-guides/trust-lab/schedules/) automate it, and the [Dashboard](/user-guides/trust-lab/dashboard/) is the health view across all of it.

:::tip[Where Trust Lab fits]
Use the [Playground](/user-guides/playground/) to experiment, [Conversations](/user-guides/conversations/) to review what really happened, and Trust Lab to lock in the behavior that must never break.
:::

<Mermaid chart={`flowchart LR
    SC[Scenarios<br/>test conversations]
    TS[Test Suites<br/>grouped scenarios]
    RU[Runs<br/>execution log]
    SCH[Schedules<br/>automated runs]
    DA[Dashboard<br/>health view]
    SC --> TS
    TS --> RU
    SCH -- "triggers" --> RU
    RU -- "feeds" --> DA
    DA -- "failures surface" --> SC
`} />

## The pieces

- **[Dashboard](/user-guides/trust-lab/dashboard/)**: the real-time health view of your test coverage.
- **[Test Suites](/user-guides/trust-lab/test-suites/)**: groups of scenarios you run, schedule, and track together.
- **[Scenarios](/user-guides/trust-lab/scenarios/)**: individual test conversations and the evals that grade them.
- **[Runs](/user-guides/trust-lab/runs/)**: the execution log, where you drill into failures.
- **[Schedules](/user-guides/trust-lab/schedules/)**: automated runs so regressions surface on their own.

## Typical Trust Lab workflow

1. **Import a real conversation**: go to [Scenarios](/user-guides/trust-lab/scenarios/) → **+ New Scenario** → **Import Conversation**, and pick a production conversation that represents an important user journey.
2. **Review and refine the evals**: check the auto-generated evals and tighten any that are too vague; add tool use evals if the scenario involves actions or navigation.
3. **Group into a suite**: add the scenario to a Sanity or Regression [suite](/user-guides/trust-lab/test-suites/).
4. **Run manually once**: click **Run Suite** to confirm everything passes before activating automation.
5. **Set a daily schedule**: [Schedules](/user-guides/trust-lab/schedules/) → **+ Add Schedule** → daily at 08:00, scoped to your suite.
6. **Check the Dashboard each morning**: if the pass rate drops, open [Runs](/user-guides/trust-lab/runs/) to find the failed assertion and fix the underlying [Knowledge](/user-guides/knowledge-base/), [action](/user-guides/authoring-actions/), or navigation config.
7. **Repeat for every new bug**: any time a real user surfaces unexpected behavior, import that conversation as a new scenario.