source · Rust

Lib

crates/vak-eval/src/lib.rs
Raw
1.0 KB24 linesSnapshot ed4ab258
  1. 1//! vak-eval: deterministic, in-process eval harness.
  2. 2//!
  3. 3//! Each case pairs a scripted model trajectory with a real workspace and a
  4. 4//! verification command. The agent loop, tools, permissions, and session
  5. 5//! ledger are the production code paths — only the provider is scripted.
  6. 6//! Pass/fail plus token/cost metrics land in a JSON report so regressions
  7. 7//! block releases.
  8. 8
  9. 9pub mod cases;
  10. 10pub mod context_engine_gate;
  11. 11pub mod no_first_class_integrations;
  12. 12pub mod runner;
  13. 13pub mod scorecard;
  14. 14
  15. 15pub use cases::{builtin_suite, general_suite, held_out_suite, live_suite};
  16. 16pub use context_engine_gate::run_context_engine_scorecard;
  17. 17pub use no_first_class_integrations::{Offense, scan_banned_tokens};
  18. 18pub use runner::{
  19. 19 EvalCase, EvalComparisonReport, EvalComparisonSuiteReport, EvalReport, EvalSuiteReport,
  20. 20 ScriptedTurn, compare_case, compare_suite, run_case, run_case_brokered, run_case_with_provider,
  21. 21 run_case_with_provider_brokered, run_suite,
  22. 22};
  23. 23pub use scorecard::{ContextScorecard, MetricDirection, QualityMetric};
  24. 24