Question: In the context of model debugging, explain why error analysis is usually repeated in several rounds. Describe the typical sequence a practitioner follows when grouping misclassified examples.

Sample answer: Error analysis is repeated because the list of categories usually changes as you inspect more examples. A practitioner may start with no formal labels, look at a small set of errors to invent an initial set of categories, and then tag more examples using those categories. While doing that, new patterns often appear, so earlier examples need to be reviewed again with the expanded set of categories.

Key points:

• Categories may begin as an empty list.
• Early examples help generate the first categories.
• New patterns can appear during manual review.
• Earlier examples should be reconsidered after categories change.

Rubric: Full credit for explaining that error analysis starts without fixed categories, uses initial examples to create categories, discovers additional categories during review, and requires revisiting earlier examples after updates.