adk_eval/
lib.rs

1//! # adk-eval
2//!
3//! Agent evaluation framework for ADK-Rust.
4//!
5//! This crate provides comprehensive tools for testing and validating agent behavior,
6//! enabling developers to ensure their agents perform correctly and consistently.
7//!
8//! ## Features
9//!
10//! - **Test Definitions**: Structured format for defining test cases (`.test.json`)
11//! - **Trajectory Evaluation**: Validate tool call sequences
12//! - **Response Quality**: Assess final output quality with multiple metrics
13//! - **Multiple Criteria**: Ground truth, rubric-based, and LLM-judged evaluation
14//! - **Automation**: Run evaluations programmatically or via CLI
15//!
16//! ## Quick Start
17//!
18//! ```rust,ignore
19//! use adk_eval::{Evaluator, EvaluationConfig, EvaluationCriteria};
20//! use std::sync::Arc;
21//!
22//! #[tokio::main]
23//! async fn main() -> Result<(), Box<dyn std::error::Error>> {
24//!     // Create your agent
25//!     let agent = create_my_agent()?;
26//!
27//!     // Configure evaluator
28//!     let config = EvaluationConfig {
29//!         criteria: EvaluationCriteria {
30//!             tool_trajectory_score: Some(1.0),  // Exact tool match
31//!             response_similarity: Some(0.8),    // 80% text similarity
32//!             ..Default::default()
33//!         },
34//!         ..Default::default()
35//!     };
36//!
37//!     let evaluator = Evaluator::new(config);
38//!
39//!     // Run evaluation
40//!     let result = evaluator
41//!         .evaluate_file(agent, "tests/my_agent.test.json")
42//!         .await?;
43//!
44//!     assert!(result.passed, "Evaluation failed: {:?}", result.failures);
45//!     Ok(())
46//! }
47//! ```
48
49pub mod criteria;
50pub mod error;
51pub mod evaluator;
52pub mod llm_judge;
53pub mod report;
54pub mod schema;
55pub mod scoring;
56
57#[cfg(feature = "personas")]
58pub mod personas;
59
60pub mod optimizer;
61
62// New unconditional modules
63pub mod annotation;
64pub mod baseline;
65pub mod conversation_scorer;
66pub mod cost_tracker;
67pub mod pricing;
68pub mod structured_judge;
69pub mod test_generator;
70pub mod trace_analyzer;
71
72// New feature-gated modules
73#[cfg(feature = "embedding")]
74pub mod embedding_scorer;
75
76#[cfg(feature = "ci-helpers")]
77pub mod junit_reporter;
78
79#[cfg(feature = "statistics")]
80pub mod ab_comparator;
81
82// Re-exports
83pub use criteria::{
84    EvaluationCriteria, ResponseMatchConfig, Rubric, RubricConfig, ToolTrajectoryConfig,
85};
86pub use error::{EvalError, Result};
87pub use evaluator::{EvaluationConfig, Evaluator};
88pub use llm_judge::{
89    LlmJudge, LlmJudgeConfig, RubricEvaluationResult, RubricScore, SemanticMatchResult,
90};
91pub use report::{EvaluationReport, EvaluationResult, Failure, TestCaseResult};
92pub use schema::{EvalCase, EvalSet, IntermediateData, SessionInput, TestFile, ToolUse, Turn};
93pub use scoring::{ResponseScorer, ToolTrajectoryScorer};
94
95// Optimizer re-exports
96pub use optimizer::{OptimizationResult, OptimizerConfig, PromptOptimizer};
97
98// New module re-exports
99pub use annotation::{AnnotationRecord, AnnotationStore, HumanVerdict};
100pub use baseline::{Baseline, BaselineStore, Regression};
101pub use conversation_scorer::{ConversationMetrics, ConversationScorer, ConversationScorerConfig};
102pub use cost_tracker::{CostMetrics, CostTracker};
103pub use pricing::ModelPricing;
104pub use structured_judge::{
105    JudgeRubric, ScalePoint, StructuredJudge, StructuredJudgeConfig, StructuredVerdict, Verdict,
106};
107pub use test_generator::{EvalCaseMetadata, GeneratorConfig, TestGenerator};
108pub use trace_analyzer::{
109    ToolCallRecord, TraceAnalysis, TraceAnalyzer, TraceDiagnostic, TracePattern,
110};
111
112#[cfg(feature = "embedding")]
113pub use embedding_scorer::EmbeddingScorer;
114
115#[cfg(feature = "ci-helpers")]
116pub use junit_reporter::JunitReporter;
117
118#[cfg(feature = "statistics")]
119pub use ab_comparator::AbComparator;
120
121/// Prelude for convenient imports
122pub mod prelude {
123    pub use crate::criteria::{
124        EvaluationCriteria, ResponseMatchConfig, Rubric, RubricConfig, ToolTrajectoryConfig,
125    };
126    pub use crate::error::{EvalError, Result};
127    pub use crate::evaluator::{EvaluationConfig, Evaluator};
128    pub use crate::llm_judge::{
129        LlmJudge, LlmJudgeConfig, RubricEvaluationResult, SemanticMatchResult,
130    };
131    pub use crate::report::{EvaluationReport, EvaluationResult, Failure, TestCaseResult};
132    pub use crate::schema::{EvalCase, EvalSet, TestFile, ToolUse, Turn};
133}
adk_eval/lib.rs

adk_eval/
lib.rs
