Case context: A team has a very small training set. Its end-to-end system lacks hand-engineered knowledge, while an alternative pipeline can incorporate components based on the team's insight.

Question: What should the team recognize about the two approaches, and which source of knowledge should it emphasize?

Sample answer: The team should recognize that the end-to-end system might do worse than the hand-engineered pipeline because the training set is small and the system lacks hand-engineered knowledge. It should emphasize human insight encoded through hand-engineered components, since most of the algorithm's knowledge may need to come from that source in this setting.

Key points:

• The training set is very small.
• The end-to-end system lacks hand-engineered knowledge.
• It might underperform the hand-engineered pipeline.
• Human insight should be encoded through hand-engineered components.

Rubric: The response should diagnose the risk caused by scarce training data, compare the end-to-end system with the hand-engineered pipeline without claiming guaranteed outcomes, and recommend emphasizing human insight through hand-engineered components.