Struct Scenario 

pub struct Scenario {
    pub id: String,
    pub turns: Vec<Turn>,
    pub expected: String,
    pub metadata: Value,
}

A single benchmark scenario loaded from a dataset file.

Each scenario represents one question/task that will be presented to the agent. The id field is used to correlate agent responses with ground-truth answers and to skip already-completed scenarios during a --resume run.

Construct via Scenario::single for single-turn scenarios (all built-in loaders), or push Turns directly into Scenario::turns for multi-turn scenarios.

Examples

use zeph_bench::Scenario;

let scenario = Scenario::single(
    "gaia_t42",
    "What is the boiling point of water in Celsius?",
    "100",
    serde_json::json!({"level": 1}),
);
assert_eq!(scenario.id, "gaia_t42");
assert_eq!(scenario.primary_prompt().unwrap(), "What is the boiling point of water in Celsius?");

Fields

id: String

Unique identifier within the dataset (e.g. "frames_0", "s1_2").

turns: Vec<Turn>

Ordered turns in this scenario. Non-empty by contract of Scenario::single.

Direct construction is allowed for multi-turn scenarios; callers must ensure at least one Role::User turn is present before calling Scenario::primary_prompt.

expected: String

The gold-standard answer used for scoring.

metadata: Value

Dataset-specific extras such as difficulty level or reasoning_types.

Set to serde_json::Value::Null when the dataset has no extra metadata.

Implementations

impl Scenario

pub fn single( id: impl Into<String>, prompt: impl Into<String>, expected: impl Into<String>, metadata: Value, ) -> Self

Convenience constructor for single-turn scenarios.

Wraps prompt in a one-element Vec<Turn> with Role::User. All built-in dataset loaders use this constructor.

Examples

use zeph_bench::Scenario;

let s = Scenario::single("id1", "What year?", "2026", serde_json::Value::Null);
assert_eq!(s.primary_prompt().unwrap(), "What year?");

pub fn primary_prompt(&self) -> Result<&str, BenchError>

Returns the content of the first Role::User turn.

Errors

Returns BenchError::InvalidFormat when turns is empty or contains no Role::User entry. Loaders must construct via Scenario::single or push at least one user turn.

Examples

use zeph_bench::Scenario;

let s = Scenario::single("id1", "hello", "world", serde_json::Value::Null);
assert_eq!(s.primary_prompt().unwrap(), "hello");

Trait Implementations

impl Clone for Scenario

fn clone(&self) -> Scenario

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Scenario

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.