Skip to content

DeepEval

DeepEval is a separate agent-quality check, not permission to bypass backend validation. Project history records a corrected 0.95 score on a bounded two-task, three-call read evaluation; an earlier 0.0 adapter run had zero evaluable tasks and cannot be interpreted as failure.

# In an isolated evaluation environment
python -m pip install deepeval
deepeval --help
deepeval test run --help
# Run ONLY an actual reviewed DeepEval test file from the repository:
deepeval test run PATH_TO_REVIEWED_TEST_FILE.py

A runnable test must supply its own LLMTestCase (inputs, observed actual output and reference/context), configured metrics and any authorized evaluator-model credentials. This documentation does not invent a project test-file path or claim the above CLI has been rerun on the current deployment. See DeepEval CLI and quickstart.


Developer Guide: Overview · ← IBM watsonx Orchestrate · Phoenix tracing →