Question: When a machine learning team is deciding what to fix next, why is the share of errors in each category not enough to choose the top priority? What other considerations should guide the decision?

Sample answer: Error counts show how often each problem appears, but frequency by itself does not tell you what to fix first. Some categories are straightforward to improve and can produce a large gain, while others are costly to address or have only a small upside. A good prioritization also considers the likely improvement available in each category and the amount of time, engineering effort, or data work required to achieve that improvement. In practice, the best next task is often the one with a strong balance of impact and feasibility, not simply the most common error.

Key points:

• Error percentages alone are not enough for prioritization
• You must estimate the likely improvement available in each category
• You must weigh the work, time, and resources needed to fix each category

Rubric: A strong response should explain that frequency does not equal priority. It must mention both the expected benefit from fixing a category and the effort required to do so.