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

# Datasets

> Curate collections of inputs, outputs, and expected outputs from real traces to test and improve your agent.

<Info>
  **Where this fits:** Datasets are part of **Refine**, after [Signals](../signals/overview). They turn real traces into reusable test cases for [regression testing](../test-and-fix/regression-testing).
</Info>

A **dataset** is a collection of rows you curate for testing and improving your agent. Each row holds an **input**, the agent's **output**, an optional **expected output**, and arbitrary **metadata**. Teams use them as golden datasets: stable, known-good test sets that a fix has to keep passing.

<Frame caption="The Datasets page lists each dataset with its description and when it was last updated.">
  <img src="https://mintcdn.com/latitude-monitoring/9O40jWPK25W0XUhc/images/datasets/datasets-list.png?fit=max&auto=format&n=9O40jWPK25W0XUhc&q=85&s=c2ab3e71b728d3a608afce7ca3f0826e" alt="The Datasets page listing golden datasets with name, description, and last updated" width="2292" height="528" data-path="images/datasets/datasets-list.png" />
</Frame>

## What a dataset row contains

| Column              | Description                                                                                                     |
| ------------------- | --------------------------------------------------------------------------------------------------------------- |
| **Input**           | The input your agent received, for example the user message.                                                    |
| **Output**          | What your agent actually returned.                                                                              |
| **Expected output** | The correct or desired answer, used to check the agent. Optional, see [Add expected output](./expected-output). |
| **Metadata**        | Arbitrary fields carried alongside the row.                                                                     |

<Frame caption="A dataset's rows, each with an input, the agent's output, and an optional expected output. Import, export, or add rows from the top bar.">
  <img src="https://mintcdn.com/latitude-monitoring/9O40jWPK25W0XUhc/images/datasets/dataset-detail.png?fit=max&auto=format&n=9O40jWPK25W0XUhc&q=85&s=2c9cc54603fd139fb718bf5ca04e3a34" alt="A dataset detail view showing rows with input, output, and expected output columns" width="2288" height="1296" data-path="images/datasets/dataset-detail.png" />
</Frame>

## Create a dataset

You can build a dataset three ways:

<CardGroup cols={2}>
  <Card title="From real traces" icon="route" href="./add-traces">
    Select traces from the trace list, search results, or a signal, and add them to a dataset. The most realistic test cases come straight from production.
  </Card>

  <Card title="Manually" icon="table">
    Open **Datasets** in your project, create a new dataset, then **Import** a CSV or **Add row** to enter cases by hand.
  </Card>

  <Card title="From your coding agent" icon="robot" href="../getting-started/mcp">
    Through the [MCP server](../getting-started/mcp), an agent like Claude or Cursor can create datasets and pull in the traces behind a signal for you.
  </Card>
</CardGroup>

## How datasets are used

* **Regression testing**: replay a dataset's inputs against your agent and compare results to the expected outputs and your evaluations. See [Regression testing](../test-and-fix/regression-testing).
* **Curating test sets**: collect representative traces from [Search](../search/overview) and [Signals](../signals/overview) into a stable, reusable set.
* **Sharing with your harness**: export a dataset as CSV to drive tests in your own pipeline.

## Next step

* [Add traces to a dataset](./add-traces): build a test set from real production traces.