The playground runs a single message through a real decision. Choose a published preset, an API key, and a level, then send the message.

<Callout type="warning" title="Live and billable">
Each run is charged to the selected key at its normal rate and may call the inference provider.
The playground is for checking policy behavior, not for free trials.
</Callout>

## Before you run [#before-you-run]

- The preset must have a published version. System presets also need the deployment's system preset seed.
- The key must be active and allow the selected level.
- The workspace must still map to a Flagline tenant.

If no published preset exists, publish one from the [preset editor](/docs/preset-editor). If no key exists, create one in [Getting started](/docs/getting-started).

## Read the result [#read-the-result]

A completed run shows the final decision, the category scores, the level, and the model information. The history keeps recent runs from this browser tab so you can compare messages.

The playground does not save a preset or create a reusable test case. To turn a draft into a policy, edit the preset and test that draft in the editor. See [Preset editor](/docs/preset-editor).

## Level choice [#level-choice]

The selector offers only levels the key allows. The preset's default level is a starting point, not a restriction. A request may choose another allowed level.

- `light` is the most permissive band and uses the first inference tier only.
- `standard` uses the first tier and can escalate eligible grey-zone results.
- `heavy` uses the second tier for every decision and the tightest threshold band.