Question: In a two-step pipeline like cat detector -> cat breed classifier, explain the logic behind why replacing the upstream component's output with a hand-labeled perfect output lets you definitively attribute an error to exactly one component. Discuss what each of the two possible outcomes reveals.

Sample answer: By replacing the detector's output with a perfect, hand-labeled bounding box, the only variable that changes is the quality of the upstream output; everything else about the classifier stays the same. If the classifier still misclassifies the image, its failure cannot be explained by a bad bounding box, so the fault must lie within the classifier itself. If instead the classifier now correctly classifies the image, this proves the classifier works correctly when given good input, meaning the original error was caused by the detector producing a flawed bounding box. This controlled substitution isolates the effect of the upstream component from the downstream one.

Key points:

• Only the upstream output is changed; the downstream component is unaffected otherwise
• If the classifier still fails, its own error is the cause
• If the classifier succeeds, the original detector output was the cause
• The test provides a controlled comparison to definitively attribute error to one component

Rubric: Full credit: explains that the test isolates the upstream component's influence by holding the downstream component's behavior constant against a perfect input, and correctly interprets both outcomes (classifier still fails vs. classifier now succeeds) with proper attribution. Partial credit: describes the procedure without explaining why it isolates the fault. No credit: does not address the causal logic of the test.