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

# Scorecards

> Turn each call's verdicts into one score and a band, weigh the judges that matter most, and cap the score when a critical judge fails.

A scorecard turns a call's verdicts into a score, such as 82 out of 100, and a band, such as *Good*. Every run is scored with one scorecard. A run's page shows the average and median score, and how many calls landed in each band.

Every account has a built-in scorecard, **Default**. It can't be changed or deleted.

## Manage scorecards

You pick a scorecard wherever one is used: in [Run evaluation](https://docs.zeroruntime.ai/agent-builder/evaluations/run-an-evaluation.md), in the agent's [auto run](https://docs.zeroruntime.ai/agent-builder/evaluations/run-an-evaluation.md#grade-calls-automatically) settings, and in [Re-score](https://docs.zeroruntime.ai/agent-builder/evaluations/review-results.md#re-score-a-run).

Open the scorecard dropdown and select **New scorecard** to create one. It's picked as soon as you save it. Select **Manage scorecards** next to the dropdown to see every scorecard, and to edit, duplicate, or delete one.

## Create a scorecard

![New scorecard form named Strict booking, out of 100, with Accuracy 0.6 and Experience 0.4, a Safety & Compliance judge weight of 2 set as a knockout capping the score at 40, judges not listed at 1, can't tell left out and reweighted, and the Good, Needs review, and Poor bands](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/scorecard-form.png)

### Basics

| Field | Required | What to enter |
| - | - | - |
| **Name** | Yes | A short name, for example *Strict booking*. Up to 256 characters. |
| **Out of** | Yes | The top score, from 1 to 1000. Defaults to 100. |

### Weights

| Field | What it does | Default |
| - | - | - |
| **Accuracy**, **Experience** | How much each category counts, from 0 to 1. `0.6` and `0.4` make a 60 / 40 split. They can't both be 0. | `0.6` and `0.4` |
| **Judge weights** | Weigh a judge more or less inside its category, from 0 to 10. `0` leaves it out. Add up to 50 with **Add judge weight**. | none |
| **Knockout** | On a judge weight: when that judge fails, the whole score is capped at **Cap the score at**. | off |
| **Judges not listed** | The weight of every judge you didn't list, from 0 to 10. | `1` |
| **When a judge can't tell** | How a can't tell verdict counts: **Leave it out and reweight the rest**, **Count it as a pass**, or **Count it as a fail**. | Leave it out |

Diagnostic judges never count toward the score.

### Bands

Each call gets the highest band its score reaches. Give each band a label and a minimum score. One band must start at 0, so every call gets a band. Add up to 10 bands.

New scorecards start with three bands:

| Band | Minimum |
| - | - |
| Good | 85 |
| Needs review | 60 |
| Poor | 0 |

Select **Create scorecard** to save it.

## How a call's score is worked out

1. Each judge gives a score: 1 for a pass, 0 for a fail. A can't tell counts as set in **When a judge can't tell**.
2. Each category's score is the weighted average of its judges' scores.
3. The call's score is the average of the Accuracy and Experience scores, weighted by their category weights, times **Out of**. A category with no graded judges is left out.
4. If a knockout judge failed, the score is capped at its cap. With more than one, the lowest cap applies.

For example, with the default 0.6 / 0.4 split, a call that passes every accuracy judge and half of its experience judges scores `(0.6 × 1 + 0.4 × 0.5) × 100 = 80`.

> [!TIP]
>
> Make the judges you can't ship without, such as safety or compliance, knockouts. A call that breaks one can't score well, however well the rest went.

## Edit and delete

Editing a scorecard applies to new runs. Every run keeps a copy of the scorecard it was scored with, so past scores don't change. To score a past run with your changes, [re-score](https://docs.zeroruntime.ai/agent-builder/evaluations/review-results.md#re-score-a-run) it.

Deleting a scorecard that an agent's auto run uses moves that agent to **Default**.

> [!NOTE]
>
> In the CLI, scorecards can also use a custom formula, and judges can grade on a scale or a curve. See [Evaluations in the CLI](https://docs.zeroruntime.ai/runtime-sdk/deployments/evaluations.md#scorecard). A scorecard with a formula keeps it when you edit it in the dashboard.


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