- LLM graders use a language model to judge behavior from conversation evidence. The built-in Expected behaviors grader checks the statements in a test. Custom graders let you write your own grading instructions.
- Code graders apply a defined calculation. The built-in Response latency grader compares measured p90 response time with your chosen maximum.
Graders
Types of graders
Egma supports LLM graders and code graders. Choose a check that can use the
evidence available in your conversations.