Which two practices help minimize breaking changes when modifying dataset schemas?

Creating new columns instead of modifying or deleting existing ones effectively minimizes breaking changes when modifying dataset schemas because it preserves the existing data structure. By introducing new columns, data consumers can continue using the old columns without disruption, ensuring backward compatibility. This practice also allows for a smoother transition for users and applications that depend on the dataset, as they can adapt to the new columns at their own pace while retaining access to the original data.

In contrast, approaches like deleting old columns immediately after adding new ones could lead to issues for users who have not yet adapted to the new schema, resulting in data access problems. Announcing the deprecation of old columns and providing instructions can help users prepare for changes, but it does not directly prevent breaking changes in the dataset structure. Modifying existing columns to repurpose them can create confusion and mismatch for consumers who expect those columns to retain their original format and data type. Thus, introducing new columns is a strategy that safeguards against interruption in data access and usage.