> ## Documentation Index
> Fetch the complete documentation index at: https://critiqor-71f5274a.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Observation Workflow: Start, Finalize, and Review Runs

> Walk through the complete Critiqor observation workflow — start a session, use your agent, finalize the run, and review reliability results.

Critiqor's observation workflow is designed to be minimal friction: start a monitored session, use your agent as you normally would, then finalize the run to generate a reliability report. This guide walks you through the full cycle — from the first command to reviewing your results and revisiting past evaluations.

<Steps>
  <Step title="Start an observation">
    Run the following command to begin monitoring your agent:

    ```bash theme={null}
    critiqor monitor openclaw
    ```

    Critiqor starts capturing evidence and then launches OpenClaw in the same terminal window. From this point forward, your agent is being observed. No additional setup or instrumentation is required.
  </Step>

  <Step title="Use your agent normally">
    With the session running, interact with OpenClaw exactly as you would in production. Run prompts, ask questions, invoke tools, and work through realistic tasks.

    Critiqor operates passively in the background — it does not intercept, slow down, or alter your agent's behavior. You don't need to change anything about how you work.

    The more representative your usage, the more useful your evaluation will be. See [Best Practices](/guides/best-practices) for guidance on what to test during a session.
  </Step>

  <Step title="Finalize the run">
    When you're done with the session, run:

    ```bash theme={null}
    critiqor finalize
    ```

    Critiqor completes the evidence capture, generates your reliability report, and automatically opens the dashboard so you can review results right away.

    **Skip the dashboard on finalize**

    If you want to finalize without immediately opening the dashboard — for example, in a scripted or CI workflow — use the `--no-dashboard` flag:

    ```bash theme={null}
    critiqor finalize --no-dashboard
    ```

    The evaluation is still saved and accessible at any time.
  </Step>

  <Step title="Review your results">
    After finalizing, the dashboard opens automatically and presents your evaluation. The key sections to review are:

    * **Executive Summary** — a plain-language overview of what Critiqor observed during the run and the overall reliability picture.
    * **Trust Score** — a single 0–100 score representing your agent's evaluated reliability. Higher is better. The score is accompanied by a trust level: Low, Moderate, High, or Very High.
    * **Primary Diagnosis** — the most significant reliability concern Critiqor identified. This is where to start when addressing issues.
    * **Recommendations** — a prioritized list of actionable improvements. Work through these in order of severity to raise your trust score over time.
  </Step>

  <Step title="Switch between runs">
    To see all recorded runs, use:

    ```bash theme={null}
    critiqor runs
    ```

    This lists every saved evaluation with its run ID, timestamp, and trust score. To open the dashboard for a specific run, pass its ID:

    ```bash theme={null}
    critiqor dashboard run_001
    ```

    You can jump to any previous run at any time — your full evaluation history is always available locally.
  </Step>

  <Step title="Open previous evaluations">
    Any past run can be reopened as a full interactive dashboard. This is especially useful when comparing reliability before and after a change — such as a prompt update, tool addition, or model swap.

    Open a previous run's dashboard the same way:

    ```bash theme={null}
    critiqor dashboard run_003
    ```

    Reviewing earlier runs alongside recent ones helps you spot trends, confirm improvements, and catch regressions before they reach production.
  </Step>
</Steps>

<Note>
  If you close the dashboard at any point, you can always reopen it with:

  ```bash theme={null}
  critiqor dashboard
  ```

  This reopens the dashboard for your most recent run. To open a specific run, pass the run ID: `critiqor dashboard run_001`.
</Note>