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

# Dataset Coverage Agent

> Compares recent production traces with an existing dataset and adds real spans that close gaps in your curation policy, on a schedule.

> [!NOTE]
>
> The Dataset Coverage Agent is part of managed agents on **Enterprise**. Signal and managed agents are currently included at no additional charge for Enterprise. See [Pricing and usage](/docs/ax/security-and-settings/pricing-and-usage.md).

The **Dataset Coverage Agent** keeps a dataset in step with production. On a schedule, it compares a tracing project's recent spans with the dataset, finds the gaps your **curation policy** describes, and copies real spans into the dataset to close them. It never writes synthetic examples.

Use it to keep an eval or experiment dataset representative as traffic changes, without hand-picking spans.

## Run it

**Signal & Agents → + New Agent → Monitor & optimize → Dataset Coverage**. You can also open a tracing project and choose **Agents → Dataset Coverage → Enable Agent**, or select **New dataset agent** on the project's **Dataset Coverage** tab.

| Setting | Default |
| - | - |
| **Project** | Required. The tracing project whose spans it reads. |
| **Dataset** | Required. Examples go into its latest version, so it needs at least one example. |
| **Curation policy** | Required, up to 10,000 characters. Which coverage gaps the agent must fill. |
| **Target dataset size** | 5,000 (maximum 5,000). The agent never grows the dataset past this. |
| **Maximum additions per run** | 10 (maximum 25). |
| **Trigger** | Daily. Editable before you save. |
| **Agent** | Claude Code on the Arize-managed provider. You can switch harness or use your own AI provider. |
| **Name** | Required. |

You can run several Dataset Coverage agents on one project, for example one per dataset or policy.

### Write a curation policy

The policy is the agent's only definition of a gap, so be specific about what to keep and in what proportion. For example:

> Keep at least 20 examples per user intent, and prioritize traces where an evaluation failed.

The setup form offers starters for common policies: failing examples, balanced intent coverage, failure-stratified coverage, and rare or long-tail inputs. If a policy is empty, vague, or contradictory, or no span fills a gap, the run succeeds and adds nothing.

## What each run does

**1. Reads the dataset and recent spans**

The agent reads every example in the dataset's latest version, plus up to **200 spans** from the project that arrived since the previous run. The first run looks back 24 hours.

**2. Picks spans that close a gap**

It chooses spans that fill gaps in the policy, never more than **Maximum additions per run**, and stops once the dataset reaches **Target dataset size**.

**3. Adds them and files a report**

It copies the chosen spans into the latest dataset version, then files one run report.

The first run starts as soon as you save. Select **Run now** on the tab to trigger another run off schedule.

## Read run reports

Reports appear on the project's **Dataset Coverage** tab, which shows once the project has an agent, and on the space-wide **Issues** page with a **Dataset Coverage** badge. If the project has more than one agent, pick one from the selector on the tab.

The tab charts dataset size and confirmed additions per run. Each report lists the spans it added, links to each span's details, records the final dataset example count, and explains why each span was chosen. Reports are a record of what changed, so they have no actions to accept.

To stop the agent, choose **Pause Dataset Coverage** from the settings menu on the tab.

## Related docs

* [Managed agents](/docs/ax/agents/build-your-own-agent.md): the gallery, agent configs, and customization
* [Eval Health agent](/docs/ax/agents/eval-health-agent.md): find missing evals and fix weak ones
* [Cost agent](/docs/ax/agents/cost-agent.md): ranked cost issues with quantified savings
* [Datasets](/docs/ax/develop/datasets): create and version the datasets this agent fills