Skip to content
Docs

Semantic rubrics

Semantic judging decides how acceptance criteria are evaluated where language interpretation is required.

LLM judging

Section titled “LLM judging”
Terminal window
export OPENAI_API_KEY=...
export DIALOGASSERT_LLM_MODEL=gpt-5-mini


uv run dialogassert run \
  -c dialogassert.openai.business-acceptance.yaml \
  --semantic-mode llm

Judge metadata — mode and confidence — is recorded alongside each criterion result.

Heuristic judging

Section titled “Heuristic judging”
Terminal window
uv run dialogassert run -c dialogassert.yaml --semantic-mode heuristic

No semantic judging

Section titled “No semantic judging”
Terminal window
uv run dialogassert run -c dialogassert.yaml --semantic-mode off

Use heuristic or off when the LLM judge is too slow or too expensive for the loop you are in. Deterministic assertions and trajectory checks keep working in every mode.