Question: In a concise analytical response, explain how the construction of the training and test sets led to a misleading view of the cat detector's deployed performance.

Sample answer: The team trained and tested the detector using cat and non-cat images downloaded from websites, so strong test performance showed that it handled that website-image distribution well. The mobile app instead received user-uploaded mobile-phone pictures, which tended to have lower resolution, more blur, and poorer lighting. Because the original test set did not represent this actual distribution of interest, its strong result did not demonstrate that the classifier would generalize to deployed inputs.

Key points:

• Training and test data came from websites.
• Deployment inputs were user-uploaded mobile-phone pictures.
• Mobile-phone pictures tended to be lower resolution, blurrier, and poorly lit.
• Success on the website-image test set did not establish generalization to the deployed distribution.

Rubric: A strong response identifies both data sources, explains the appearance mismatch, and connects the unrepresentative test distribution to poor deployed generalization. Partial credit should be given for identifying the mismatch without fully explaining why the test result was misleading.