> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getoneprofile.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# An agent gave a wrong answer

> The column filled in, but the answer is wrong. Read what the agent did, then fix the prompt, the inputs, or the effort.

An agent can answer confidently and still be wrong. Read what it actually did before you touch the prompt.

## Work through the checks

<Steps>
  <Step title="Open what the agent did">
    Right-click the cell and choose **View run** to see what it searched, read and used to answer.
  </Step>

  <Step title="Check the inputs it was given">
    An agent reads only the columns bound to it. Open the column's editor and confirm the right fields are mapped under **Inputs**.
  </Step>

  <Step title="Tighten the prompt">
    Add the missing detail or limit to the agent's **Prompt**, then run the cell again.
  </Step>

  <Step title="Raise the effort">
    A higher option in the agent's **Effort** card reads more carefully before it answers.
  </Step>

  <Step title="Rate the cell">
    Click the thumbs-down icon on the cell and add a note on what was wrong. That note is what a training round acts on.
  </Step>
</Steps>

<Note>
  A built-in agent has no **Prompt** or **Effort** card to change. Keep a copy of it first, then edit the copy.
</Note>

## Good to know

<AccordionGroup>
  <Accordion title="Why doesn't View run show any steps?">
    A cell that needed no research or tools answered in one pass, so there is nothing to trace. An old run's trail may also have been cleared to save space. Run the cell again for a fresh one.
  </Accordion>

  <Accordion title="Does a higher effort always fix it?">
    Not if the cause was a missing input or a vague prompt. See [Choose an effort level](/agents/choose-an-effort-level) for what each option changes and costs.
  </Accordion>
</AccordionGroup>

## Next steps

<CardGroup cols={2}>
  <Card title="Read an agent run" icon="activity" href="/agents/read-an-agent-run">
    Every step an agent takes, explained.
  </Card>

  <Card title="Train an agent" icon="graduation-cap" href="/agents/train-an-agent">
    Turn rated cells into a better version.
  </Card>
</CardGroup>