# EvalTurn record

> The EvalTurn record of Novior.MentorAgent.Evaluation: One user turn of an EvalConversation, with what it must do. Every check is optional; a turn with none…

- URL: https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalturn
- Verified with: MentorAgent 1.4.0
- Last updated: 2026-11-18
- Packages: Novior.MentorAgent

One user turn of an [`EvalConversation`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalconversation), with what it must do. Every check is optional; a turn with none passes when it answers with some text and without an error.

## Definition

Namespace: `Novior.MentorAgent.Evaluation`. Package: `Novior.MentorAgent`.

`public sealed record EvalTurn : IEquatable<EvalTurn>`

```shell
dotnet add package Novior.MentorAgent
```

## Constructors

`public EvalTurn(string Input)`

One user turn of an [`EvalConversation`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalconversation), with what it must do. Every check is optional; a turn with none passes when it answers with some text and without an error.

**Parameters**

- `Input`: The user message of this turn.

## Properties

### `ExpectRefusal`

`public EvalRefusal? ExpectRefusal { get; init; }`

The refusal this turn must get: the mentor turns the message down instead of answering, and the user sees an error. With it, that refusal passes the turn and the conversation goes on; a turn that was answered, refused the other way, or failed, fails, and its notes say which. `null`, the default, expects none: a refusal fails the turn as any error does.

A refusal is one of the two judgements on the message: [`EvalRefusal.OutOfScope`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalrefusal) and [`EvalRefusal.Guardrail`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalrefusal). A safety check that could not run, a message too long, the rate limit and a turn without a tenant are not refusals: they stay errors, expected or not, so a case that must be refused does not pass because the guardrail is down.

### `ExpectTools`

`public IReadOnlyList<string>? ExpectTools { get; init; }`

Tools this turn must call, by technical name (for example `get_order`): the turn fails if any of them is not in [`MentorEvalResult.ToolsCalled`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/mentorevalresult#toolscalled). This is how a case says "at turn 2 it must call X".

### `Expectation`

`public string? Expectation { get; init; }`

The expected behaviour, for the built-in judge, as [`EvalCase.Expectation`](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalcase#expectation).

### `ForbidTools`

`public IReadOnlyList<string>? ForbidTools { get; init; }`

Tools this turn must not call, by technical name: the turn fails if any of them was called.

### `Input`

`public string Input { get; init; }`

The user message of this turn.

### `MaxTimeToFirstToken`

`public TimeSpan? MaxTimeToFirstToken { get; init; }`

The longest acceptable wait for the first streamed text of the answer; the turn fails if the first text came later, or never came.

### `MaxTokens`

`public long? MaxTokens { get; init; }`

Token budget of this turn — every chat-model call it made; the turn fails if exceeded.

---

- Previous: [EvalRefusal enum](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/evalrefusal.md)
- Next: [MentorEvalOptions class](https://docs.mentoragent.net/api/novior-mentoragent-evaluation/mentorevaloptions.md)
