Confidence Scoring

​

How Confidence Scoring Works

• Analyzes multiple factors in real-time
• Provides scores from 0.0 to 1.0
• Considers context relevance and data quality
• Adapts based on your feedback

​

Confidence Factors

1. Data Relevance: How well the retrieved data matches the query
2. Source Quality: Reliability of the data sources
3. Context Alignment: Consistency with the conversation context
4. Model Certainty: The AI model’s confidence in its response

​

Using Confidence Scores

response = client.rag.chat(
    messages=[{"role": "user", "content": "Your question"}],
    confidence_threshold=0.7  # Only return high-confidence responses
)

print(f"Confidence: {response.confidence_score}")

​

Best Practices

• Set appropriate thresholds for your use case
• Monitor confidence trends over time
• Use scores to trigger human review when needed
• Combine with decision tracing for full transparency

​

Next Steps

• Explore Decision Tracing
• Learn about Execution Control
• View API Examples