Struct E 

pub struct E;

The E namespace — static factory methods for evaluation criteria.

Implementations

impl E

pub fn suite() -> EvalSuite

Create an evaluation suite.

pub fn response_match() -> ECriterion

Exact response match criterion.

pub fn contains_match() -> ECriterion

Substring containment criterion — scores 1.0 if output contains expected.

pub fn safety(llm: Arc<dyn BaseLlm>) -> ECriterion

Safety criterion — LLM judge; scores 1.0 when the output is safe.

Mirrors ADK’s safety evaluation, run locally against the judge llm.

pub fn semantic_match(llm: Arc<dyn BaseLlm>) -> ECriterion

Semantic match criterion — LLM judge comparing the output to the expected reference answer; scores 1.0 when they convey the same answer.

Mirrors ADK’s final_response_match_v2 (LLM-as-judge with a reference).

pub fn hallucination(llm: Arc<dyn BaseLlm>) -> ECriterion

Hallucination criterion — LLM judge; scores 1.0 when the output is free of fabricated claims relative to the expected reference.

pub fn trajectory() -> ECriterion

Tool-trajectory criterion (EXACT match) — mirrors ADK’s TrajectoryEvaluator. Both output and expected are parsed as tool-call sequences (a JSON array of names/objects, or a comma-separated list), so an eval harness can score the agent’s captured tool calls against an expected sequence. Scores 1.0 on an exact match, else 0.0.

pub fn trajectory_in_order() -> ECriterion

Tool-trajectory criterion requiring the expected calls in order (extras allowed in between) — ADK’s IN_ORDER mode.

pub fn trajectory_any_order() -> ECriterion

Tool-trajectory criterion requiring the expected calls in any order (extras allowed) — ADK’s ANY_ORDER mode.

pub fn custom( name: &'static str, f: impl Fn(&str, &str) -> f64 + Send + Sync + 'static, ) -> ECriterion

Custom evaluation criterion from a scoring function.

pub fn from_file(path: &str) -> EvalSuite

Load eval cases from a file path.

The file should contain one case per pair of consecutive lines: odd lines are prompts, even lines are expected responses. Lines starting with # are comments and blank lines are skipped.

pub fn persona(name: &'static str, description: &'static str) -> ECriterion

Create a persona-based evaluator for user simulation.

The persona describes a simulated user with a given name and description, which can be used to generate realistic test interactions.