> For the complete documentation index, see [llms.txt](https://docs.radar.citadel-ai.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.radar.citadel-ai.com/en/overview/faq.md).

# Frequently Asked Questions (FAQ)

This page summarizes frequently asked questions when using Citadel Radar.

Start by reviewing the section closest to your role.

<table><thead><tr><th width="148.96484375">Role</th><th>Main topics to check</th></tr></thead><tbody><tr><td>Common</td><td>Basic concepts of Citadel Radar, Node, Application, and more</td></tr><tr><td>Application Owners</td><td>Evaluation libraries, metrics, evaluation packages, monitoring, alerts, and more</td></tr><tr><td>Administrators</td><td>User management, role permissions, resource permissions, billing, and more</td></tr><tr><td>Developers</td><td>API keys, API endpoints, Swagger Playground, and more</td></tr></tbody></table>

### Common

#### What is a Node?

> This unit is used to isolate processing infrastructure.&#x20;
>
> By setting up independent processing unit by environment or department, you can limit the impact of heavy workloads and failures.
>
> For example, Nodes can be used to separate Production / Staging environments or separate departments.
>
> Reference: [Citadel Radar Overview](/en/overview/overview.md)

#### What is an Application?

> An Application is the target evaluated and monitored by Citadel Radar.
>
> Register chatbots, internal business support tools, external user-facing services, and other applications that use LLMs by application.
>
> Reference: [Citadel Radar Overview](/en/overview/overview.md), [Register a New Application](/en/application-owner-guide/new-application.md)

### For Application Owners

#### What is an Evaluation Library?

> The Evaluation Library is a feature for centrally managing resources used to evaluate inputs and outputs, such as Evaluation Categories, Evaluation Metrics, and Evaluation Packages.
>
> You can review Built-In Metrics, create Custom Metrics, create Evaluation Packages, and more.
>
> Reference: [Manage the Evaluation Library](/en/application-owner-guide/evaluation-library.md)

#### What is the difference between an Evaluation Category and an Evaluation Metric?

> An Evaluation Category is a unit that defines a risk or threat perspective.
>
> An Evaluation Metric is tied to that category and is the criterion for judging whether input logs or output logs match the detection target.
>
> Reference: [Manage the Evaluation Library](/en/application-owner-guide/evaluation-library.md)

#### What is an Evaluation Package?

> An Evaluation Package is a set that groups multiple evaluation metrics and each metric's runtime parameters.
>
> You can preconfigure not only the metrics to use, but also each metric's action, borderline handling, and Remote LLM. Created Evaluation Packages can be selected when applying metrics to an application.
>
> Reference: [Create Evaluation Packages](/en/application-owner-guide/evaluation-library/evaluation-packages.md)

#### Are there any precautions when configuring metrics that Block?

> Evaluation Metrics for which `Block` is selected are judged synchronously by Radar, and the evaluation result is returned to the application after judgment completes. Therefore, requests to each configured evaluation model are executed, and the overall response time is affected by the slowest responding model.
>
> To operate with stable response times, we recommend using the same model for evaluation metrics that Block.
>
> Evaluation Metrics for which `Flag` is selected are processed asynchronously, so this does not apply.
>
> Reference: [Configure Evaluation Metrics for an Application](/en/application-owner-guide/new-application/set-metrics.md)

#### What is the difference between Local type and Remote type metrics?

> Local type metrics use evaluation models managed by Citadel AI for judgment. They perform evaluation without using external Remote LLM services.
>
> Remote type metrics use external Remote LLM services for judgment. With Remote type metrics, you can choose a fast model or high-accuracy model as the model used for evaluation.
>
> Choose according to the metric type and operational policy while considering the balance among evaluation accuracy, response time, and cost.
>
> Reference: [Test metric accuracy](/en/application-owner-guide/evaluation-library/playground.md)

#### What is the difference between fast models and high-accuracy models?

> Fast models are used when response time and evaluation cost are prioritized.
>
> High-accuracy models are used when evaluation accuracy is prioritized. They are suitable for metrics where you want to prioritize judgment correctness or when evaluating high-importance risks.
>
> If the selected metric is Local type, fast model / high-accuracy model selection is not available.
>
> Reference: [Test metric accuracy](/en/application-owner-guide/evaluation-library/playground.md)

#### What is the difference between Block and Flag?

> Block is used when you want to control the behavior of the target AI system for inputs or outputs where a problem has been detected.
>
> Flag records detected problem logs without blocking. Select Flag when you want to avoid user impact and focus on follow-up review or analysis.
>
> Reference: [Configure Evaluation Metrics for an Application](/en/application-owner-guide/new-application/set-metrics.md)

#### What is borderline handling?

> Borderline handling is a setting that specifies whether to treat an evaluation result as Pass or Fail when it is near the judgment criterion boundary.
>
> You can use Pass when you want to reduce false positives, or Fail when you want to prevent missed detections.
>
> Reference: [Configure Evaluation Metrics for an Application](/en/application-owner-guide/new-application/set-metrics.md)

#### Can I check metric evaluation accuracy in advance?

> You can check the evaluation accuracy of built-in metrics and custom metrics in the metrics playground.
>
> In free input, you can enter any test data and check judgment details one by one. In AI-generated datasets, you can generate multiple test data items based on a specified use case and check metric accuracy and misjudgments.
>
> Reference: [Test metric accuracy](/en/application-owner-guide/evaluation-library/playground.md)

#### Can I review the details of problematic logs?

> Yes. In traffic page details, you can review the input content, output content, judgment results for each evaluation metric, judgment reasons, metadata, and more.
>
> Additionally, by using the AI explanation feature, you can generate an explanation detailing exactly which parts of a flagged log were determined to be problematic.
>
> Reference: [Traffic](/en/application-owner-guide/monitor-applications/traffic.md)

### For administrators

#### Can I create any number of user accounts?

> Yes. You can create as many user accounts as needed.
>
> No additional charges are incurred based on the number of users.
>
> Reference: [Manage Permissions](/en/administrator-guide/manage-permissions.md)

#### Can I configure different permissions for the same user by application?

> Yes. Resource permissions can be configured per application.
>
> The same user can have edit permission for one application and view-only permission for another application.
>
> Reference: [Manage Permissions](/en/administrator-guide/manage-permissions.md)

### For developers

#### How do I send logs to Citadel Radar?

> Send input logs and output logs from the monitored application to the Citadel Radar API.
>
> Endpoints for input evaluation and output evaluation are available. You can check API specifications and request formats on the API endpoints page.
>
> Reference: [API endpoints](/en/developer-guide/api-endpoints.md)

#### At what unit are API keys issued?

> API keys are issued per application.
>
> Separating API keys by application makes it easier to manage which application is sending logs. It also lets you limit the impact scope when deleting keys that are no longer needed or updating keys.
>
> Reference: [Issue and Manage API Keys](/en/developer-guide/api-keys.md)

#### Can I configure scopes or permissions for each API key?

> Currently, there is no feature for configuring permissions per API key.
>
> API keys are used as authentication information for log-sending POST APIs. API keys are managed in association with the application that issued them.
>
> Reference: [Issue and Manage API Keys](/en/developer-guide/api-keys.md)

#### Is there a way to try API request formats?

> Yes. You can review the API documentation from `Open API documentation` on the Node details page.
>
> In the API documentation, you can use Swagger Playground to check Citadel Radar API request and response formats. You can also use an issued API key to try log-sending POST APIs.
>
> Reference: [Issue and Manage API Keys](/en/developer-guide/api-keys.md#try-requests-in-swagger-playground)