Module evals

Module evals 

Source
Expand description

The fixture-driven quality harness.

Memory quality is not something a type system can hold up. These are the measurements that decide whether a change to ranking, extraction or policy made B better or merely different: a hand-written corpus, a set of cases stating what should happen, and the thresholds from the design’s acceptance criteria enforced as tests.

Re-exports§

pub use fixtures::IngestionCase;
pub use fixtures::RetrievalCase;
pub use fixtures::corpus;
pub use fixtures::eval_user;
pub use fixtures::ingestion_cases;
pub use fixtures::retrieval_cases;
pub use harness::IngestionReport;
pub use harness::RetrievalReport;
pub use harness::run_ingestion_eval;
pub use harness::run_retrieval_eval;

Modules§

fixtures
The evaluation corpus and cases.
harness
Running the evaluation corpus against the real engine.
metrics
Retrieval quality metrics.