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

# Jabrod Evals

> Find where your AI application fails, and why.

> [!NOTE]
>
> Evals is available at [jabrod.com/app/evals](https://jabrod.com/app/evals). Sign in with your Jabrod account. To work with the Jabrod team on an evaluation, [book a call](https://jabrod.com/evals).

Jabrod Evals tests retrieval systems, agents and support assistants. You give it:

* a **test set**: questions your users really ask, with what a good answer looks like;
* a **target**: the AI application to test, reached over HTTP, or a [Context](https://docs.jabrod.com/context/index.md) workspace.

Evals sends every question to the target, scores every answer with the **suites** you pick, and gives you a **report**: pass rates by suite, each failure with its reason, and the change since the last run.

## Suites

| Suite | Checks that |
| - | - |
| Grounded in sources | Every claim in the answer is supported by the passages the app retrieved |
| Correctness | The answer agrees with the expected answer and contains the required phrases |
| Correct refusals | The app declines out-of-scope questions, and only those |
| Tool call arguments | The app calls the expected tool with the expected arguments |
| Policy and tone | The answer follows a policy you write |

## Next

> [card] [Quickstart](https://docs.jabrod.com/evals/quickstart.md)
>
> Run a first evaluation.

> [card] [Suites and scoring](https://docs.jabrod.com/evals/suites.md)
>
> How each answer is judged.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.