Member Junction
    Preparing search index...

    LLM Judge Oracle.

    Uses an LLM to evaluate output quality based on custom criteria. Provides semantic evaluation beyond deterministic checks.

    Configuration:

    • criteria: Array of validation criteria (required)
    • model: Model to use for judging (default: from prompt or default model)
    • temperature: Temperature for LLM (default: 0.1 for consistency)
    • promptTemplate: Custom prompt template (optional, uses default if not provided)
    • strictMode: Require all criteria to pass (default: false, uses weighted scoring)
    const oracle = new LLMJudgeOracle();
    const result = await oracle.evaluate({
    actualOutput: { response: 'Sales by region report created successfully' },
    expectedOutput: {
    judgeValidationCriteria: [
    'Response accurately answers the user\'s question',
    'Response includes actionable information',
    'Response is professional and clear'
    ]
    },
    contextUser
    }, {
    model: 'claude-sonnet-4',
    temperature: 0.1
    });

    Implements

    Index

    Constructors

    Properties

    Methods

    Constructors

    Properties

    type: "llm-judge" = 'llm-judge'

    Unique type identifier for this oracle. Used in test configuration to specify which oracles to run.

    Methods