Expand description
Multi-turn simulation support for eval case generation and replay.
This module implements Phase 6 (US4) of spec 043-evals-adv-features:
ActorSimulatordrives a simulated user across multiple turns.ToolSimulatorproduces schema-valid tool responses, backed by aStateRegistryof bounded-historyStateBuckets.run_multiturn_simulationorchestrates an agent ↔ actor dialogue up tomax_turnsor goal-completion.
All surfaces gated by the crate-level simulation feature. Tests live in
eval/tests/simulation_test.rs, eval/tests/simulation_state_test.rs,
and eval/tests/us4_end_to_end_test.rs.
Re-exports§
pub use actor::ActorProfile;pub use actor::ActorSimulator;pub use actor::ActorTurn;pub use orchestrator::SimulationError;pub use orchestrator::SimulationOutcome;pub use orchestrator::run_multiturn_simulation;pub use tool::StateBucket;pub use tool::StateRegistry;pub use tool::ToolCallRecord;pub use tool::ToolSchema;pub use tool::ToolSimulator;
Modules§
- actor
- Simulated-user driver (US4, FR-024).
- orchestrator
- Multi-turn simulation orchestrator (US4, FR-026).
- tool
- Tool-call simulator and shared state registry (US4, FR-025).