# evals

> Reference for the evals tool: create evals with cases and a scoring method, read them and their past runs.

Source: https://prompt-verse.io/docs/reference/evals/

An eval is a list of cases (an id, an input and what you expect) and a method that scores answers from 0 to 1. A threshold, 0.8 by default, says whether a version passes.

Call it as `geenius_promptverse(tool: "evals", action: "<action>", args: {...})` through Geenius OS, or as the `evals` tool of the Prompt Verse MCP server with `action` and `args`.

## Actions

| Action | What it does |
|---|---|
| `create` | A new eval with its cases and method. |
| `get` | One eval with its cases. Your AI can use these cases to test a prompt itself. |
| `list` | Your evals, newest first. |
| `run` | Score a prompt on a model with the saved cases. Switched off in the pre-beta. |
| `runs` | Past runs of an eval. |

## Arguments we have checked

### evals · create

A new eval with its cases and method.

- `entity_type` (required): prompt, workflow or agent.
- `entity_id` (required): Its id.
- `title` (required): What it checks.
- `cases` (required): A list of cases: objects with id, input and expected.
- `method` (required): exact, contains, structural, judge or hybrid.
- `threshold`: From 0 to 1. Default 0.8.

### evals · get

One eval with its cases. Your AI can use these cases to test a prompt itself.

- `id` (required): Eval id.

### evals · run

Score a prompt on a model with the saved cases. Switched off in the pre-beta.

- `eval_id` (required): Eval id.
- `prompt_id` (required): Prompt id.
- `model_id` (required): Model id.

### evals · runs

Past runs of an eval.

- `eval_id` (required): Eval id.
- `limit`: Default 20.

> [!TIP]
> Ask for the exact arguments of any action with `action: "help", of: "create"`. The server answers from its own definitions, so that answer is always current.

> [!HONEST]
> run is listed but refuses while the AI features are off. Your AI can still read the saved cases with get, run the prompt itself and score the answers in your conversation.