Case context: A company is building a classifier to flag defective parts on an assembly line. Experienced inspectors can label examples reliably, and the engineering team can review many of the model's mistakes. The model's error rate is still far higher than the specialist baseline.

Question: What does this comparison imply, and how should the team use the expert benchmark while improving the system?

Sample answer: A large gap between the model and expert performance points to substantial avoidable bias. The expert benchmark should be used to estimate the lowest error the task is likely to permit and to choose a realistic target error rate that the project can actually reach. The team can also rely on experts to generate training data and to help diagnose failures. Since the gap is large, there are likely several promising ways to improve the system.

Key points:

• A wide gap signals substantial avoidable bias.
• Expert performance helps set the practical floor and the improvement target.
• The target must be realistic, not merely aspirational.
• Experts can contribute labeled examples.
• Experts can help interpret errors.
• A large gap suggests multiple paths for progress.

Rubric: The response should identify substantial avoidable bias, explain that expert performance is useful for estimating the practical floor and setting a reachable target, and describe how experts support labeling and error review. It should also connect the diagnosis to the fact that several improvement strategies may be available.