Describe data skew, the different types, and how each affects sharing.

Two types matter for sharing, and they have different remedies.

1. Ownership skew: One user owns a very large number of records — commonly an integration user or a placeholder. Because the role hierarchy grants access upward, any change to that user's role, or to the hierarchy above them, forces recalculation across all of those records. The remedies are to distribute ownership across many users, or to move the owning user outside the role hierarchy so there is no upward path to recalculate.
2. Account or lookup data skew: Very many child records under one parent, conventionally described as more than roughly ten thousand. The platform locks the parent during child DML and maintains implicit shares for the children, producing row-lock errors and slow recalculation. The remedy is to distribute the children across a set of bucket parent records.

The diagnostic habit I would want to demonstrate is separating them by symptom. Lock errors during a bulk child load point at parent skew. Slow recalculation after a role change points at ownership skew.