Expand description
GraphRAG evaluation harness for retrieval and code-impact observations.
The harness is intentionally pure: callers run CLI/API/storage workflows, then submit compact observations here so exact fact, multi-hop, temporal, negative, stale-index, ambiguous-entity, and code-impact checks share one scoring contract.
Structs§
- Evaluation
Case - Expected behavior for one evaluation query or workflow.
- Evaluation
Observation - Compact observation submitted by integration tests or diagnostics commands.
- Evaluation
Report - Aggregated report for a Phase 4 evaluation run.
- Evaluation
Result - Per-case evaluation result with concrete failure reasons.
Enums§
- Evaluation
Case Kind - Phase 4 GraphRAG evaluation scenario family.
- Evaluation
Error - Evaluation input validation error.
Functions§
- evaluate_
case - Scores one case against one observation.
- evaluate_
suite - Scores a suite where observations are supplied in case order.
- phase4_
fixture_ cases - Small deterministic Phase 4 fixture suite used by integration and CI gates.