Question: In a model review table used to compare performance across two data sources, how is the table organized, and why is it useful to fill in the extra cells?

Sample answer: The table uses one axis for the two data sources being compared and the other axis for three error categories: human-level error, error on data the model was trained to handle, and error on data it has not seen in training. Completing the extra cells helps reveal whether the model is struggling differently on the two sources, which makes it easier to diagnose the underlying problem.

Key points:

• One axis lists the two data sources or distributions.
• The other axis lists human-level error, error on trained data, and error on untrained data.
• Extra cells help expose differences in behavior across the two sources.

Rubric: A strong answer should identify both axes correctly and explain that the additional entries provide diagnostic information about how the model behaves across the two data sources.