Agent evaluation harness: golden set regression + paired bootstrap CI + 16 scorers + 30 cases across 6 failure modes. v0.2.0 baseline: deterministic 30/30, flaky 5/30, 5 scorers regressed at 95% CI.
python regression-testing trajectory rag pydantic ai-agent llm ollama trulens ragas llm-as-judge tool-calling deepeval openai-compatible agent-evaluation golden-set bootstrap-ci ci-gate paired-bootstrap golden-set-regression
-
Updated
Oct 14, 2026 - Python