Case context: A machine learning team has an idea for a new image model. They are excited and want to spend the next four weeks building it right away. Another engineer suggests first reviewing 100 recent misclassifications by hand, which would take about two hours, to see whether the idea addresses the real failure modes.

Question: What should the team do, and what is the main danger of ignoring the engineer's suggestion?

Sample answer: The team should pause and inspect the 100 misclassified images before starting the month-long build. The review takes very little time compared with four weeks of development, and it can prevent the team from investing heavily in a change that does not fix the actual problems. If they skip the review, they may spend a month building something that ends up improving the model very little or not at all.

Key points:

• Do the short manual error review first.
• Do not jump straight into a long implementation effort.
• The risk is spending a month on a change with little or no payoff.

Rubric: The response must recommend the manual review and explain the risk of losing a month of effort on a change that does not meaningfully help. It should also contrast the short review time with the much larger implementation cost.