trace-001
Query
Response
Retrieved Context
Metadata
Evaluation Checklist
⚡ Quick Checks
Has Response
response length > 50
Professional Greeting
matches pattern /hello|hi|thank you|glad to help/i
Exposed Internal
must not contain "internal-only"
Debug Output
must not contain "DEBUG:"
Credentials Leaked
must not contain "api_key"
Uses Knowledge Base
uses context
Cites Sources
cites sources
Fast Response
code: metadata.latency < 3000
Not Too Verbose
code: response.split(' ').length < 500
Escalation Protocol
llm: If the issue is about legal matters, billing disputes over $1000, or safety concerns, does the response offer to connect with a human specialist?
Upsell Appropriately
llm: If mentioning premium features, is it relevant to the user's need and not pushy?
🔍 Deep Checks
Actually Helpful
llm: Does this response genuinely help solve the user's problem, not just acknowledge it?
No Hallucination
no hallucination
Empathetic Tone
llm: Is the response empathetic and understanding of the customer's situation?
Notes (optional)
Pass P
Fail F
Skip S
Prev/Next ← →
Toggle eval 1-9 1-9