What is the primary function of a grader in a prompt evaluation workflow?	To take model output and provide an objective signal, such as a number or a true/false value.
What are the three types of graders discussed in the transcript?	Code-based, Model-based, and Human-based.
What kind of checks can a code-based grader perform on model output?	Programmatic checks such as verifying output length, checking for specific words, or validating JSON/code syntax.
How does a model-based grader function?	It takes the original model output and feeds it into an additional model API request for evaluation.
What is the main drawback of using human-based grading?	It generally takes a lot of time and is a tedious process.
What are the three specific evaluation criteria mentioned for the use case?	1. Output format (Python, JSON, or regex) without extra explanation. 2. Valid syntax for the generated code/data. 3. General task following and accuracy.
Why is it recommended to ask a model for strengths and weaknesses when grading?	To prevent the model from giving middling scores (like 6) and encourage a more concrete, focused score.
Which type of grader was deemed most appropriate for checking the general response and task following?	The model grader, due to its flexibility.
What is the final step taken to derive a single objective metric from the evaluation results?	Calculating the average score of all the individual model grades.