Question: In a document-processing pipeline, suppose stage 1 extracts names, stage 2 classifies each name, and stage 3 builds a final report. Explain why replacing stage 1’s output with the correct names and then running stages 2 and 3 can reveal whether stage 1 is the main source of the overall error.

Sample answer: The idea is to change only one part of the system while keeping the rest unchanged. If you feed stages 2 and 3 the correct output from stage 1, then any improvement in the final report can be attributed to the quality of stage 1’s original output, because the later stages are still behaving the same way. If the final result becomes much better, stage 1 was likely the main bottleneck. If the result barely changes, then stage 1 is probably not the main cause, and the problem is more likely in stage 2 or stage 3. This lets you localize the failure without rebuilding the whole pipeline.

Key points:

• Replace only one stage’s output at a time
• Keep the remaining stages fixed for a fair comparison
• A noticeable gain means the replaced stage was contributing strongly to the error
• Little or no gain suggests a later stage is responsible
• This is a practical way to debug a multi-stage system

Rubric: Full credit explains that substituting a correct intermediate output while holding later stages constant isolates that stage’s causal role in the final error; also gives credit for noting that little or no improvement points to downstream stages.