()

Configuration Status

Checking configuration ...

Current Settings

Setting Value
User Persona Simulator Model anthropic/claude-sonnet-4-5
Default Assistant Model — the model being evaluated anthropic/claude-sonnet-4-5
Default Judge Model(s) anthropic/claude-sonnet-4-5
Default Number of Conversation Rounds 4
Default Parallelism (number of concurrent conversations to run in parallel) 1

Evaluation Rubric Settings

Configure the scoring rubric used for evaluating assistant model responses.

Setting Value Description
Rubric to apply in evaluations Select a rubric from tblRubric to use for evaluating assistant responses.
Type of rubric algorithm to apply in evaluations

Loading rubric algorithm types ...

 If the rubric algorithm type is "word", the legacy rubric will be applied. It has dimensions (criteria) identified by a word or a short phrase.
With this algorithm, deriving a score of 1-10 for each dimension of the rubric is delegated to the Judge.

If the rubric algorithm type is "sentence", a more verbose rubric will be applied where each dimension (criterion) is identified by a sentence.
With this algorithm, if scoring guidance is enabled, it will help the Judge apply a more nuanced score of 1-5 per dimension.  (Scores of 1-5 are normalized to 1-10 for consistency with the legacy rubric type.)
Use scoring guidance (loading ...) Read-only. Controlled by tblRubricAlgorithm.blnRubricAlgorithm_UseScoringGuidance via the rubric's associated algorithm.
When enabled, evaluation uses a behavioral descriptor table from tblCriterionScoring that describes the quality of the assistant's behavior for each scoring level in the judge prompt:  1=Absent, 2=Minimal, 3=Adequate, 4=Skilled, 5=Exceptional.

Environment Variables

Configure these in your .env file or Azure App Settings:

OPENROUTER_API_KEY Checking ...
PERSONA_API_KEY (Optional -- falls back to OPENROUTER_API_KEY)
ASSISTANT_MODEL_API_KEY (Optional -- falls back to OPENROUTER_API_KEY)
JUDGE_API_KEY (Optional -- falls back to OPENROUTER_API_KEY)
SECRET_KEY (For Flask sessions -- auto-generated in dev)

Documentation