Case context: A robotics team is developing software that converts streams from thermal sensors, vibration readings, and a few camera measurements into control commands for a sorting machine. A manager wants one model to learn the entire mapping from the raw signals to the final actions. The team, however, has only a small number of verified examples showing the correct command for each sensor reading.

Question: What should you tell the manager about using a fully direct model here, and what is the main reason?

Sample answer: I would not recommend relying on a fully direct model at this point. Approaches that learn the full input-to-output mapping usually depend on a large collection of labeled examples that pair each input with the correct output. Since this project has only a small set of verified sensor-command pairs, the model is unlikely to learn the task well.

Key points:

• Do not rely on a fully direct model yet
• These methods usually need many labeled input-output pairs
• Too little paired data makes the model hard to train well

Rubric: 1 point for advising against a fully direct model at this stage. 1 point for explaining that such models need many labeled input-output pairs. 1 point for stating that the limited paired data makes training difficult.