self-healing-rag — eval history
1 run(s). Every push runs the golden set and fails the build if the hallucination rate exceeds 5% or the system answers more than 10% of the questions it should refuse.
| metric | latest | trend |
|---|---|---|
| Hallucination rate | 4.4% | needs 2+ runs |
| False-answer rate (unanswerable) | 0.0% | needs 2+ runs |
| False-abstention rate | 0.0% | needs 2+ runs |
| Answer relevancy | 0.958 | needs 2+ runs |
| Key-fact coverage | 0.86 | needs 2+ runs |
| Citation rate | 100.0% | needs 2+ runs |
| p95 latency | 8384ms | needs 2+ runs |
| Cost per query | $0.0005 | needs 2+ runs |