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

# Run comparisons

> Score one response with several evaluators and judge models.

Enter one response and score it with several evaluators and judge models at once, laid out as a grid. The models here do the scoring rather than being scored, so a comparison tells you how your evaluators and judges behave on text you already have.

## Before you start

You need at least one [model](/guard/app/models) to judge with, and at least one LLM-as-a-Judge [evaluator](/guard/app/evaluators).

## Create a comparison

<Steps>
  <Step title="New Comparison">
    On the **Comparisons** page, click **New Comparison**.
  </Step>

  <Step title="Create Comparison">
    * **LLM Response (for Evaluation)** – enter the response you want scored. A comparison does not pull responses from datasets or examinations.
    * **Evaluators** – choose one or more. Only LLM-as-a-Judge evaluators appear.
    * **Judge Models** – choose the models that will score the response.

    Click **Create Comparison**.
  </Step>
</Steps>

## Read the comparison grid

A comparison's page is titled with the response itself, and shows it in full under **Response Being Evaluated**. The grid below has one row per evaluator and one column per judge model, with each judge's provider beneath its name. **Add Models** brings more judges into an existing comparison.

Each cell shows that judge's score as a gauge, labelled with the direction the evaluator scores in, or an error icon if scoring failed. Expand an evaluator's row for its detailed results: one card per judge model, holding the score and the judge's written reasoning, or **Evaluation failed**.

The comparison's status appears beside its title.

| Status   | Meaning                                 |
| -------- | --------------------------------------- |
| Pending  | Scoring has not started.                |
| Running  | Judges are still scoring.               |
| Complete | Every cell returned a score.            |
| Partial  | Some cells returned and others did not. |
| Error    | Scoring failed.                         |

## Manage your comparisons

The **Comparisons** list shows one row per comparison, with **Name**, **Models**, **Status**, **Visibility**, **Created**, and **Actions**. The name is the response that was scored, and **Models** is the number of judge models.

Above the table, filter by **Visibility**.

Each row's **Actions** column opens the comparison, and offers a delete action for comparisons you created.